Merge pull request #109 from fosrl/dev

1.5.0

.github/ISSUE_TEMPLATE/1.bug_report.yml

@@ -14,12 +14,13 @@ body:
           label: Environment
           description: Please fill out the relevant details below for your environment.
           value: |
-              - OS Type & Version: (e.g., Ubuntu 22.04)
+              - OS Type & Version:
               - Pangolin Version:
+              - Edition (Community or Enterprise):
               - Gerbil Version:
               - Traefik Version:
               - Newt Version:
-              - Olm Version: (if applicable)
+              - Client Version:
       validations:
           required: true
 

.github/dependabot.yml

@@ -1,40 +1,32 @@
 version: 2
+
 updates:
   - package-ecosystem: "gomod"
     directory: "/"
     schedule:
       interval: "daily"
+    open-pull-requests-limit: 1
     groups:
-      dev-patch-updates:
+      go-dependencies:
-        dependency-type: "development"
+        patterns:
-        update-types:
+          - "*"
-          - "patch"
-      dev-minor-updates:
-        dependency-type: "development"
-        update-types:
-          - "minor"
-      prod-patch-updates:
-        dependency-type: "production"
-        update-types:
-          - "patch"
-      prod-minor-updates:
-        dependency-type: "production"
-        update-types:
-          - "minor"
 
   - package-ecosystem: "docker"
     directory: "/"
     schedule:
       interval: "daily"
+    open-pull-requests-limit: 1
     groups:
-      patch-updates:
+      docker-dependencies:
-        update-types:
+        patterns:
-          - "patch"
+          - "*"
-      minor-updates:
-        update-types:
-          - "minor"
 
   - package-ecosystem: "github-actions"
     directory: "/"
     schedule:
       interval: "weekly"
+    open-pull-requests-limit: 1
+    groups:
+      github-actions-dependencies:
+        patterns:
+          - "*"

.github/workflows/cicd.yml

@@ -36,16 +36,16 @@ jobs:
 
     steps:
       - name: Checkout code
-        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+        uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
 
       - name: Set up QEMU
-        uses: docker/setup-qemu-action@c7c53464625b32c7a7e944ae62b3e17d2b600130 # v3.7.0
+        uses: docker/setup-qemu-action@96fe6ef7f33517b61c61be40b68a1882f3264fb8 # v4.2.0
 
       - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@8d2750c68a42422c14e847fe6c8ac0403b4cbd6f # v3.12.0
+        uses: docker/setup-buildx-action@bb05f3f5519dd87d3ba754cc423b652a5edd6d2c # v4.2.0
 
       - name: Log in to Docker Hub
-        uses: docker/login-action@b45d80f862d83dbcd57f89517bcf500b2ab88fb2 # v4.0.0
+        uses: docker/login-action@af1e73f918a031802d376d3c8bbc3fe56130a9b0 # v4.4.0
         with:
           registry: docker.io
           username: ${{ secrets.DOCKER_HUB_USERNAME }}
@@ -57,9 +57,9 @@ jobs:
         shell: bash
 
       - name: Install Go
-        uses: actions/setup-go@4b73464bb391d4059bd26b0524d20df3927bd417 # v6.3.0
+        uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16 # v6.5.0
         with:
-          go-version: 1.25
+          go-version: 1.26
 
       - name: Update version in main.go
         run: |
@@ -80,7 +80,7 @@ jobs:
         shell: bash
 
       - name: Login in to GHCR
-        uses: docker/login-action@b45d80f862d83dbcd57f89517bcf500b2ab88fb2 # v4.0.0
+        uses: docker/login-action@af1e73f918a031802d376d3c8bbc3fe56130a9b0 # v4.4.0
         with:
           registry: ghcr.io
           username: ${{ github.actor }}
@@ -107,8 +107,9 @@ jobs:
         shell: bash
 
       - name: Install cosign
-        # cosign is used to sign and verify container images (key and keyless)
+        uses: sigstore/cosign-installer@6f9f17788090df1f26f669e9d70d6ae9567deba6 # v4.1.2
-        uses: sigstore/cosign-installer@ba7bc0a3fef59531c69a25acd34668d6d3fe6f22 # v4.1.0
+        with:
+          cosign-release: v3.0.6
 
       - name: Dual-sign and verify (GHCR & Docker Hub)
         # Sign each image by digest using keyless (OIDC) and key-based signing,
@@ -155,7 +156,7 @@ jobs:
         shell: bash
 
       - name: Upload artifacts from /bin
-        uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7.0.0
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
           name: binaries
           path: bin/

.github/workflows/mirror.yaml

@@ -23,7 +23,7 @@ jobs:
           skopeo --version
 
       - name: Install cosign
-        uses: sigstore/cosign-installer@ba7bc0a3fef59531c69a25acd34668d6d3fe6f22 # v4.1.0
+        uses: sigstore/cosign-installer@6f9f17788090df1f26f669e9d70d6ae9567deba6 # v4.1.2
 
       - name: Input check
         run: |

.github/workflows/test.yml

@@ -14,12 +14,16 @@ jobs:
     runs-on: amd64-runner
 
     steps:
-      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
+      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+
+      - name: Read go version
+        id: goversion
+        run: echo "version=$(cat .go-version)" >> $GITHUB_OUTPUT
 
       - name: Set up Go
-        uses: actions/setup-go@4b73464bb391d4059bd26b0524d20df3927bd417 # v6.3.0
+        uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16 # v6.5.0
         with:
-          go-version: 1.26
+          go-version: ${{ steps.goversion.outputs.version }}
 
       - name: Build go
         run: go build

.go-version

@@ -1 +1 @@
-1.25
+1.26

Dockerfile

@@ -1,4 +1,5 @@
-FROM golang:1.26-alpine AS builder
+ARG GO_VERSION=1.26
+FROM golang:${GO_VERSION}-alpine AS builder
 
 # Set the working directory inside the container
 WORKDIR /app
@@ -16,7 +17,7 @@ COPY . .
 RUN CGO_ENABLED=0 GOOS=linux go build -o /gerbil
 
 # Start a new stage from scratch
-FROM alpine:3.23 AS runner
+FROM alpine:3.24 AS runner
 
 RUN apk add --no-cache iptables iproute2
 
@@ -25,4 +26,4 @@ COPY entrypoint.sh /
 
 RUN chmod +x /entrypoint.sh
 ENTRYPOINT ["/entrypoint.sh"]
-CMD ["gerbil"]
+CMD ["gerbil"]

Makefile

@@ -1,4 +1,6 @@
 
+GO_VERSION := $(shell cat .go-version 2>/dev/null)
+
 all: build push
 
 docker-build-release:
@@ -7,10 +9,10 @@ docker-build-release:
 		exit 1; \
 	fi
 	docker buildx build --platform linux/arm64,linux/amd64 -t fosrl/gerbil:latest -f Dockerfile --push .
-	docker buildx build --platform linux/arm64,linux/amd64 -t fosrl/gerbil:$(tag) -f Dockerfile --push .
+	docker buildx build --platform linux/arm64,linux/amd64 -t fosrl/gerbil:$(tag) -f Dockerfile --build-arg GO_VERSION=$(GO_VERSION) --push .
 
 build:
-	docker build -t fosrl/gerbil:latest .
+	docker build -t fosrl/gerbil:latest --build-arg GO_VERSION=$(GO_VERSION) .
 
 push:
 	docker push fosrl/gerbil:latest

README.md

@@ -138,7 +138,7 @@ make
 
 ### Binary
 
-Make sure to have Go 1.23.1 installed.
+Make sure to have Go 1.26 installed.
 
 ```bash
 make local

docs/observability.md

@@ -83,6 +83,7 @@ type OTelConfig struct {
     Endpoint       string        // default: "localhost:4317"
     Insecure       bool          // default: true
     ExportInterval time.Duration // default: 60s
+    Timeout        time.Duration // default: 10s
 }
 ```
 
@@ -97,6 +98,7 @@ type OTelConfig struct {
 | `OTEL_METRICS_ENDPOINT` | `localhost:4317` | OTLP collector address |
 | `OTEL_METRICS_INSECURE` | `true` | Disable TLS for OTLP |
 | `OTEL_METRICS_EXPORT_INTERVAL` | `60s` | Push interval (e.g. `10s`, `1m`) |
+| `OTEL_METRICS_TIMEOUT` | `10s` | Timeout for OTLP exporter connection setup |
 | `DEPLOYMENT_ENVIRONMENT` | _(unset)_ | OTel deployment.environment attribute |
 
 ### CLI flags
@@ -108,7 +110,8 @@ type OTelConfig struct {
 --otel-metrics-protocol      string  (default: grpc)
 --otel-metrics-endpoint      string  (default: localhost:4317)
 --otel-metrics-insecure      bool    (default: true)
---otel-metrics-export-interval  duration  (default: 1m0s)
+--otel-metrics-export-interval  duration  (default: 60s)
+--otel-metrics-timeout          duration  (default: 10s)
 ```
 
 ---
@@ -164,6 +167,7 @@ export OTEL_METRICS_PROTOCOL=grpc
 export OTEL_METRICS_ENDPOINT=otel-collector:4317
 export OTEL_METRICS_INSECURE=true
 export OTEL_METRICS_EXPORT_INTERVAL=10s
+export OTEL_METRICS_TIMEOUT=10s
 export DEPLOYMENT_ENVIRONMENT=production
 ```
 
@@ -176,6 +180,7 @@ export DEPLOYMENT_ENVIRONMENT=production
          --otel-metrics-endpoint=otel-collector:4317 \
          --otel-metrics-insecure \
          --otel-metrics-export-interval=10s \
+         --otel-metrics-timeout=10s \
          --config=/etc/gerbil/config.json
 ```
 
@@ -225,7 +230,6 @@ All metrics use the prefix `gerbil_<component>_<name>`.
 | Metric | Type | Labels |
 |--------|------|--------|
 | `gerbil_proxy_mapping_active` | UpDownCounter | `ifname` |
-| `gerbil_session_active` | UpDownCounter | `ifname` |
 | `gerbil_active_sessions` | UpDownCounter | `ifname` |
 | `gerbil_udp_packets_total` | Counter | `ifname`, `type`, `direction` |
 | `gerbil_hole_punch_events_total` | Counter | `ifname`, `result` |
@@ -256,7 +260,7 @@ The `docker-compose.metrics.yml` provides a complete observability stack.
 **Prometheus mode:**
 
 ```bash
-METRICS_BACKEND=prometheus docker-compose -f docker-compose.metrics.yml up -d
+METRICS_BACKEND=prometheus docker-compose -f docker compose.metrics.yml up -d
 # Scrape at http://localhost:3003/metrics
 # Grafana at http://localhost:3000 (admin/admin)
 ```
@@ -265,5 +269,5 @@ METRICS_BACKEND=prometheus docker-compose -f docker-compose.metrics.yml up -d
 
 ```bash
 METRICS_BACKEND=otel OTEL_METRICS_ENDPOINT=otel-collector:4317 \
-  docker-compose -f docker-compose.metrics.yml up -d
+  docker compose -f docker-compose.metrics.yml up -d
 ```

examples/otel-collector-config.yaml

@@ -1,3 +1,4 @@
+file_format: '1.0'
 receivers:
   otlp:
     protocols:
@@ -43,4 +44,4 @@ service:
     metrics:
       receivers: [otlp]
       processors: [batch, resource]
-      exporters: [prometheus, prometheusremotewrite, debug]
+      exporters: [prometheus, prometheusremotewrite, debug]

go.mod

@@ -4,16 +4,16 @@ go 1.26.0
 
 require (
 	github.com/patrickmn/go-cache v2.1.0+incompatible
-	github.com/prometheus/client_golang v1.20.5
+	github.com/prometheus/client_golang v1.23.2
 	github.com/vishvananda/netlink v1.3.1
-	go.opentelemetry.io/otel v1.42.0
+	go.opentelemetry.io/otel v1.44.0
-	go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetricgrpc v1.42.0
+	go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetricgrpc v1.44.0
-	go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetrichttp v1.42.0
+	go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetrichttp v1.44.0
-	go.opentelemetry.io/otel/metric v1.42.0
+	go.opentelemetry.io/otel/metric v1.44.0
-	go.opentelemetry.io/otel/sdk v1.42.0
+	go.opentelemetry.io/otel/sdk v1.44.0
-	go.opentelemetry.io/otel/sdk/metric v1.42.0
+	go.opentelemetry.io/otel/sdk/metric v1.44.0
-	golang.org/x/crypto v0.49.0
+	golang.org/x/crypto v0.53.0
-	golang.org/x/sync v0.20.0
+	golang.org/x/sync v0.21.0
 	golang.zx2c4.com/wireguard/wgctrl v0.0.0-20230429144221-925a1e7659e6
 )
 
@@ -25,26 +25,26 @@ require (
 	github.com/go-logr/stdr v1.2.2 // indirect
 	github.com/google/go-cmp v0.7.0 // indirect
 	github.com/google/uuid v1.6.0 // indirect
-	github.com/grpc-ecosystem/grpc-gateway/v2 v2.28.0 // indirect
+	github.com/grpc-ecosystem/grpc-gateway/v2 v2.29.0 // indirect
 	github.com/josharian/native v1.1.0 // indirect
-	github.com/klauspost/compress v1.17.9 // indirect
 	github.com/mdlayher/genetlink v1.3.2 // indirect
 	github.com/mdlayher/netlink v1.7.2 // indirect
 	github.com/mdlayher/socket v0.4.1 // indirect
 	github.com/munnerz/goautoneg v0.0.0-20191010083416-a7dc8b61c822 // indirect
-	github.com/prometheus/client_model v0.6.1 // indirect
+	github.com/prometheus/client_model v0.6.2 // indirect
-	github.com/prometheus/common v0.61.0 // indirect
+	github.com/prometheus/common v0.66.1 // indirect
-	github.com/prometheus/procfs v0.15.1 // indirect
+	github.com/prometheus/procfs v0.16.1 // indirect
 	github.com/vishvananda/netns v0.0.5 // indirect
 	go.opentelemetry.io/auto/sdk v1.2.1 // indirect
-	go.opentelemetry.io/otel/trace v1.42.0 // indirect
+	go.opentelemetry.io/otel/trace v1.44.0 // indirect
-	go.opentelemetry.io/proto/otlp v1.9.0 // indirect
+	go.opentelemetry.io/proto/otlp v1.10.0 // indirect
-	golang.org/x/net v0.51.0 // indirect
+	go.yaml.in/yaml/v2 v2.4.2 // indirect
-	golang.org/x/sys v0.42.0 // indirect
+	golang.org/x/net v0.55.0 // indirect
-	golang.org/x/text v0.35.0 // indirect
+	golang.org/x/sys v0.46.0 // indirect
+	golang.org/x/text v0.38.0 // indirect
 	golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b // indirect
-	google.golang.org/genproto/googleapis/api v0.0.0-20260209200024-4cfbd4190f57 // indirect
+	google.golang.org/genproto/googleapis/api v0.0.0-20260526163538-3dc84a4a5aaa // indirect
-	google.golang.org/genproto/googleapis/rpc v0.0.0-20260209200024-4cfbd4190f57 // indirect
+	google.golang.org/genproto/googleapis/rpc v0.0.0-20260526163538-3dc84a4a5aaa // indirect
-	google.golang.org/grpc v1.79.3 // indirect
+	google.golang.org/grpc v1.81.1 // indirect
 	google.golang.org/protobuf v1.36.11 // indirect
 )

go.sum

@@ -17,12 +17,16 @@ github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
 github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
 github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
-github.com/grpc-ecosystem/grpc-gateway/v2 v2.28.0 h1:HWRh5R2+9EifMyIHV7ZV+MIZqgz+PMpZ14Jynv3O2Zs=
+github.com/grpc-ecosystem/grpc-gateway/v2 v2.29.0 h1:5VipnvEpbqr2gA2VbM+nYVbkIF28c5ZQfqCBQ5g2xfk=
-github.com/grpc-ecosystem/grpc-gateway/v2 v2.28.0/go.mod h1:JfhWUomR1baixubs02l85lZYYOm7LV6om4ceouMv45c=
+github.com/grpc-ecosystem/grpc-gateway/v2 v2.29.0/go.mod h1:Hyl3n6Twe1hvtd9XUXDec4pTvgMSEixRuQKPTMH2bNs=
 github.com/josharian/native v1.1.0 h1:uuaP0hAbW7Y4l0ZRQ6C9zfb7Mg1mbFKry/xzDAfmtLA=
 github.com/josharian/native v1.1.0/go.mod h1:7X/raswPFr05uY3HiLlYeyQntB6OO7E/d2Cu7qoaN2w=
-github.com/klauspost/compress v1.17.9 h1:6KIumPrER1LHsvBVuDa0r5xaG0Es51mhhB9BQB2qeMA=
+github.com/klauspost/compress v1.18.0 h1:c/Cqfb0r+Yi+JtIEq73FWXVkRonBlf0CRNYc8Zttxdo=
-github.com/klauspost/compress v1.17.9/go.mod h1:Di0epgTjJY877eYKx5yC51cX2A2Vl2ibi7bDH9ttBbw=
+github.com/klauspost/compress v1.18.0/go.mod h1:2Pp+KzxcywXVXMr50+X0Q/Lsb43OQHYWRCY2AiWywWQ=
+github.com/kr/pretty v0.3.1 h1:flRD4NNwYAUpkphVc1HcthR4KEIFJ65n8Mw5qdRn3LE=
+github.com/kr/pretty v0.3.1/go.mod h1:hoEshYVHaxMs3cyo3Yncou5ZscifuDolrwPKZanG3xk=
+github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
+github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
 github.com/kylelemons/godebug v1.1.0 h1:RPNrshWIDI6G2gRW9EHilWtl7Z6Sb1BR0xunSBf0SNc=
 github.com/kylelemons/godebug v1.1.0/go.mod h1:9/0rRGxNHcop5bhtWyNeEfOS8JIWk580+fNqagV/RAw=
 github.com/mdlayher/genetlink v1.3.2 h1:KdrNKe+CTu+IbZnm/GVUMXSqBBLqcGpRDa0xkQy56gw=
@@ -39,14 +43,16 @@ github.com/patrickmn/go-cache v2.1.0+incompatible h1:HRMgzkcYKYpi3C8ajMPV8OFXaaR
 github.com/patrickmn/go-cache v2.1.0+incompatible/go.mod h1:3Qf8kWWT7OJRJbdiICTKqZju1ZixQ/KpMGzzAfe6+WQ=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
-github.com/prometheus/client_golang v1.20.5 h1:cxppBPuYhUnsO6yo/aoRol4L7q7UFfdm+bR9r+8l63Y=
+github.com/prometheus/client_golang v1.23.2 h1:Je96obch5RDVy3FDMndoUsjAhG5Edi49h0RJWRi/o0o=
-github.com/prometheus/client_golang v1.20.5/go.mod h1:PIEt8X02hGcP8JWbeHyeZ53Y/jReSnHgO035n//V5WE=
+github.com/prometheus/client_golang v1.23.2/go.mod h1:Tb1a6LWHB3/SPIzCoaDXI4I8UHKeFTEQ1YCr+0Gyqmg=
-github.com/prometheus/client_model v0.6.1 h1:ZKSh/rekM+n3CeS952MLRAdFwIKqeY8b62p8ais2e9E=
+github.com/prometheus/client_model v0.6.2 h1:oBsgwpGs7iVziMvrGhE53c/GrLUsZdHnqNwqPLxwZyk=
-github.com/prometheus/client_model v0.6.1/go.mod h1:OrxVMOVHjw3lKMa8+x6HeMGkHMQyHDk9E3jmP2AmGiY=
+github.com/prometheus/client_model v0.6.2/go.mod h1:y3m2F6Gdpfy6Ut/GBsUqTWZqCUvMVzSfMLjcu6wAwpE=
-github.com/prometheus/common v0.61.0 h1:3gv/GThfX0cV2lpO7gkTUwZru38mxevy90Bj8YFSRQQ=
+github.com/prometheus/common v0.66.1 h1:h5E0h5/Y8niHc5DlaLlWLArTQI7tMrsfQjHV+d9ZoGs=
-github.com/prometheus/common v0.61.0/go.mod h1:zr29OCN/2BsJRaFwG8QOBr41D6kkchKbpeNH7pAjb/s=
+github.com/prometheus/common v0.66.1/go.mod h1:gcaUsgf3KfRSwHY4dIMXLPV0K/Wg1oZ8+SbZk/HH/dA=
-github.com/prometheus/procfs v0.15.1 h1:YagwOFzUgYfKKHX6Dr+sHT7km/hxC76UB0learggepc=
+github.com/prometheus/procfs v0.16.1 h1:hZ15bTNuirocR6u0JZ6BAHHmwS1p8B4P6MRqxtzMyRg=
-github.com/prometheus/procfs v0.15.1/go.mod h1:fB45yRUv8NstnjriLhBQLuOUt+WW4BsoGhij/e3PBqk=
+github.com/prometheus/procfs v0.16.1/go.mod h1:teAbpZRB1iIAJYREa1LsoWUXykVXA1KlTmWl8x/U+Is=
+github.com/rogpeppe/go-internal v1.14.1 h1:UQB4HGPB6osV0SQTLymcB4TgvyWu6ZyliaW0tI/otEQ=
+github.com/rogpeppe/go-internal v1.14.1/go.mod h1:MaRKkUm5W0goXpeCfT7UZI6fk/L7L7so1lCWt35ZSgc=
 github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu7U=
 github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=
 github.com/vishvananda/netlink v1.3.1 h1:3AEMt62VKqz90r0tmNhog0r/PpWKmrEShJU0wJW6bV0=
@@ -55,47 +61,56 @@ github.com/vishvananda/netns v0.0.5 h1:DfiHV+j8bA32MFM7bfEunvT8IAqQ/NzSJHtcmW5zd
 github.com/vishvananda/netns v0.0.5/go.mod h1:SpkAiCQRtJ6TvvxPnOSyH3BMl6unz3xZlaprSwhNNJM=
 go.opentelemetry.io/auto/sdk v1.2.1 h1:jXsnJ4Lmnqd11kwkBV2LgLoFMZKizbCi5fNZ/ipaZ64=
 go.opentelemetry.io/auto/sdk v1.2.1/go.mod h1:KRTj+aOaElaLi+wW1kO/DZRXwkF4C5xPbEe3ZiIhN7Y=
-go.opentelemetry.io/otel v1.42.0 h1:lSQGzTgVR3+sgJDAU/7/ZMjN9Z+vUip7leaqBKy4sho=
+go.opentelemetry.io/otel v1.44.0 h1:JjwHmHpA4iZ3wBxluu2fbbE7j4kqlE8jXyAyPXH7HqU=
-go.opentelemetry.io/otel v1.42.0/go.mod h1:lJNsdRMxCUIWuMlVJWzecSMuNjE7dOYyWlqOXWkdqCc=
+go.opentelemetry.io/otel v1.44.0/go.mod h1:BMgjTHL9WPRlRjL2oZCBTL4whCGtXch2H4BhOPIAyYc=
-go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetricgrpc v1.42.0 h1:MdKucPl/HbzckWWEisiNqMPhRrAOQX8r4jTuGr636gk=
+go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetricgrpc v1.44.0 h1:SUplec5dp06reu1zaXmOXdvqH398taqrDXqUl99jxSc=
-go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetricgrpc v1.42.0/go.mod h1:RolT8tWtfHcjajEH5wFIZ4Dgh5jpPdFXYV9pTAk/qjc=
+go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetricgrpc v1.44.0/go.mod h1:ho2g4N+ane+swq5I/VBkKWnRDY4kUINH3FuqyZqX/Ug=
-go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetrichttp v1.42.0 h1:H7O6RlGOMTizyl3R08Kn5pdM06bnH8oscSj7o11tmLA=
+go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetrichttp v1.44.0 h1:RuynHbfU8JUEw7DyONgkVYg2SVtsoF28y0LGIr69jgA=
-go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetrichttp v1.42.0/go.mod h1:mBFWu/WOVDkWWsR7Tx7h6EpQB8wsv7P0Yrh0Pb7othc=
+go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetrichttp v1.44.0/go.mod h1:qZF+/lBs71APw8mlnEZcqZHMzqrYrsFiJOv83lX1OGo=
-go.opentelemetry.io/otel/metric v1.42.0 h1:2jXG+3oZLNXEPfNmnpxKDeZsFI5o4J+nz6xUlaFdF/4=
+go.opentelemetry.io/otel/metric v1.44.0 h1:1w0gILTcHdr3YI+ixLyjemwrVnsMURbTZFrSYCdDdmc=
-go.opentelemetry.io/otel/metric v1.42.0/go.mod h1:RlUN/7vTU7Ao/diDkEpQpnz3/92J9ko05BIwxYa2SSI=
+go.opentelemetry.io/otel/metric v1.44.0/go.mod h1:8O7hanEPBNgEMmybD3s2VBKcgWOCsA6tzHBPODAiquo=
-go.opentelemetry.io/otel/sdk v1.42.0 h1:LyC8+jqk6UJwdrI/8VydAq/hvkFKNHZVIWuslJXYsDo=
+go.opentelemetry.io/otel/metric/x v0.66.0 h1:YkCrx1zLOChi9ZcZ6euupOcsgzbVlec7D/xoEU1+cTA=
-go.opentelemetry.io/otel/sdk v1.42.0/go.mod h1:rGHCAxd9DAph0joO4W6OPwxjNTYWghRWmkHuGbayMts=
+go.opentelemetry.io/otel/metric/x v0.66.0/go.mod h1:d1+BDj9t96do0/1LoU1ayfCv79ZgNE41qbhBvnMOBZk=
-go.opentelemetry.io/otel/sdk/metric v1.42.0 h1:D/1QR46Clz6ajyZ3G8SgNlTJKBdGp84q9RKCAZ3YGuA=
+go.opentelemetry.io/otel/sdk v1.44.0 h1:nHYwb9lK+fJPU/dnT6s7W7Z8itMWyqrnVfbheVYrZ58=
-go.opentelemetry.io/otel/sdk/metric v1.42.0/go.mod h1:Ua6AAlDKdZ7tdvaQKfSmnFTdHx37+J4ba8MwVCYM5hc=
+go.opentelemetry.io/otel/sdk v1.44.0/go.mod h1:Osuydd3Se74nqjAKxid74N5eC+jfEqfTegHRnq58oK0=
-go.opentelemetry.io/otel/trace v1.42.0 h1:OUCgIPt+mzOnaUTpOQcBiM/PLQ/Op7oq6g4LenLmOYY=
+go.opentelemetry.io/otel/sdk/metric v1.44.0 h1:3LlKgI+VjbVsjNRFZJZAJ30WjXC5VkNRks6si09iEfI=
-go.opentelemetry.io/otel/trace v1.42.0/go.mod h1:f3K9S+IFqnumBkKhRJMeaZeNk9epyhnCmQh/EysQCdc=
+go.opentelemetry.io/otel/sdk/metric v1.44.0/go.mod h1:5B5pMARnXxKhltooO4xUuCBorl65a4EpnTalObqOigA=
-go.opentelemetry.io/proto/otlp v1.9.0 h1:l706jCMITVouPOqEnii2fIAuO3IVGBRPV5ICjceRb/A=
+go.opentelemetry.io/otel/trace v1.44.0 h1:jxF5CsGYCe74MCRx2X4g7WsY/VBKRqqpNvXlX/6gtIk=
-go.opentelemetry.io/proto/otlp v1.9.0/go.mod h1:xE+Cx5E/eEHw+ISFkwPLwCZefwVjY+pqKg1qcK03+/4=
+go.opentelemetry.io/otel/trace v1.44.0/go.mod h1:oLl1jrMQAVo6v3GAggN+1VH9VIz9iUSvW53sW1Q8PIE=
-golang.org/x/crypto v0.49.0 h1:+Ng2ULVvLHnJ/ZFEq4KdcDd/cfjrrjjNSXNzxg0Y4U4=
+go.opentelemetry.io/proto/otlp v1.10.0 h1:IQRWgT5srOCYfiWnpqUYz9CVmbO8bFmKcwYxpuCSL2g=
-golang.org/x/crypto v0.49.0/go.mod h1:ErX4dUh2UM+CFYiXZRTcMpEcN8b/1gxEuv3nODoYtCA=
+go.opentelemetry.io/proto/otlp v1.10.0/go.mod h1:/CV4QoCR/S9yaPj8utp3lvQPoqMtxXdzn7ozvvozVqk=
-golang.org/x/net v0.51.0 h1:94R/GTO7mt3/4wIKpcR5gkGmRLOuE/2hNGeWq/GBIFo=
+go.uber.org/goleak v1.3.0 h1:2K3zAYmnTNqV73imy9J1T3WC+gmCePx2hEGkimedGto=
-golang.org/x/net v0.51.0/go.mod h1:aamm+2QF5ogm02fjy5Bb7CQ0WMt1/WVM7FtyaTLlA9Y=
+go.uber.org/goleak v1.3.0/go.mod h1:CoHD4mav9JJNrW/WLlf7HGZPjdw8EucARQHekz1X6bE=
-golang.org/x/sync v0.20.0 h1:e0PTpb7pjO8GAtTs2dQ6jYa5BWYlMuX047Dco/pItO4=
+go.yaml.in/yaml/v2 v2.4.2 h1:DzmwEr2rDGHl7lsFgAHxmNz/1NlQ7xLIrlN2h5d1eGI=
-golang.org/x/sync v0.20.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
+go.yaml.in/yaml/v2 v2.4.2/go.mod h1:081UH+NErpNdqlCXm3TtEran0rJZGxAYx9hb/ELlsPU=
+golang.org/x/crypto v0.53.0 h1:QZ4Muo8THX6CizN2vPPd5fBGHyogrdK9fG4wLPFUsto=
+golang.org/x/crypto v0.53.0/go.mod h1:DNLU434OwVakk9PzuwV8w62mAJpRJL3vsgcfp4Qnsio=
+golang.org/x/net v0.55.0 h1:bcvxaJn3e1U6InsFWt1JUq1aSjnRxLzT2rtD2KfkDF8=
+golang.org/x/net v0.55.0/go.mod h1:L5U2KuzuOe1lY7Z+aWVIKK6qEeJXnXV9yzGA+WCHJww=
+golang.org/x/sync v0.21.0 h1:HLII4xRRTtCRkxYp4HNFF0Js/Og6q2i++KXbg0gHCwM=
+golang.org/x/sync v0.21.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.2.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
 golang.org/x/sys v0.10.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.42.0 h1:omrd2nAlyT5ESRdCLYdm3+fMfNFE/+Rf4bDIQImRJeo=
+golang.org/x/sys v0.46.0 h1:noSf2Fq6F8DBgS+LysIkx7rIExoNHJsxOAtPp4rthXw=
-golang.org/x/sys v0.42.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
+golang.org/x/sys v0.46.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
-golang.org/x/text v0.35.0 h1:JOVx6vVDFokkpaq1AEptVzLTpDe9KGpj5tR4/X+ybL8=
+golang.org/x/text v0.38.0 h1:sXmwo9DwP3OK9EZ7PqAdaooSGozfl/3a6/xJcbzPRhE=
-golang.org/x/text v0.35.0/go.mod h1:khi/HExzZJ2pGnjenulevKNX1W67CUy0AsXcNubPGCA=
+golang.org/x/text v0.38.0/go.mod h1:YXZt3QhHUKYT53r2lLKFIVi6Ao1jdzrTR/KQ09qyxF4=
 golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b h1:J1CaxgLerRR5lgx3wnr6L04cJFbWoceSK9JWBdglINo=
 golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b/go.mod h1:tqur9LnfstdR9ep2LaJT4lFUl0EjlHtge+gAjmsHUG4=
 golang.zx2c4.com/wireguard/wgctrl v0.0.0-20230429144221-925a1e7659e6 h1:CawjfCvYQH2OU3/TnxLx97WDSUDRABfT18pCOYwc2GE=
 golang.zx2c4.com/wireguard/wgctrl v0.0.0-20230429144221-925a1e7659e6/go.mod h1:3rxYc4HtVcSG9gVaTs2GEBdehh+sYPOwKtyUWEOTb80=
-gonum.org/v1/gonum v0.16.0 h1:5+ul4Swaf3ESvrOnidPp4GZbzf0mxVQpDCYUQE7OJfk=
+gonum.org/v1/gonum v0.17.0 h1:VbpOemQlsSMrYmn7T2OUvQ4dqxQXU+ouZFQsZOx50z4=
-gonum.org/v1/gonum v0.16.0/go.mod h1:fef3am4MQ93R2HHpKnLk4/Tbh/s0+wqD5nfa6Pnwy4E=
+gonum.org/v1/gonum v0.17.0/go.mod h1:El3tOrEuMpv2UdMrbNlKEh9vd86bmQ6vqIcDwxEOc1E=
-google.golang.org/genproto/googleapis/api v0.0.0-20260209200024-4cfbd4190f57 h1:JLQynH/LBHfCTSbDWl+py8C+Rg/k1OVH3xfcaiANuF0=
+google.golang.org/genproto/googleapis/api v0.0.0-20260526163538-3dc84a4a5aaa h1:Kjn0N0tCrDgiAFW+lGO4JZ3ck44CehvJQMAwj9QF0G8=
-google.golang.org/genproto/googleapis/api v0.0.0-20260209200024-4cfbd4190f57/go.mod h1:kSJwQxqmFXeo79zOmbrALdflXQeAYcUbgS7PbpMknCY=
+google.golang.org/genproto/googleapis/api v0.0.0-20260526163538-3dc84a4a5aaa/go.mod h1:q4lMZS6kskjT5HvCPrnnypcDPVJqT/f4nfxmkE7gryY=
-google.golang.org/genproto/googleapis/rpc v0.0.0-20260209200024-4cfbd4190f57 h1:mWPCjDEyshlQYzBpMNHaEof6UX1PmHcaUODUywQ0uac=
+google.golang.org/genproto/googleapis/rpc v0.0.0-20260526163538-3dc84a4a5aaa h1:mZHHdPZl0dbGHCflZgAq/Q468DWVFcU2whhB2KAo8fk=
-google.golang.org/genproto/googleapis/rpc v0.0.0-20260209200024-4cfbd4190f57/go.mod h1:j9x/tPzZkyxcgEFkiKEEGxfvyumM01BEtsW8xzOahRQ=
+google.golang.org/genproto/googleapis/rpc v0.0.0-20260526163538-3dc84a4a5aaa/go.mod h1:4Hqkh8ycfw05ld/3BWL7rJOSfebL2Q+DVDeRgYgxUU8=
-google.golang.org/grpc v1.79.3 h1:sybAEdRIEtvcD68Gx7dmnwjZKlyfuc61Dyo9pGXXkKE=
+google.golang.org/grpc v1.81.1 h1:VnnIIZ88UzOOKLukQi+ImGz8O1Wdp8nAGGnvOfEIWQQ=
-google.golang.org/grpc v1.79.3/go.mod h1:KmT0Kjez+0dde/v2j9vzwoAScgEPx/Bw1CYChhHLrHQ=
+google.golang.org/grpc v1.81.1/go.mod h1:xGH9GfzOyMTGIOXBJmXt+BX/V0kcdQbdcuwQ/zNw42I=
 google.golang.org/protobuf v1.36.11 h1:fV6ZwhNocDyBLK0dj+fg8ektcVegBBuEolpbTQyBNVE=
 google.golang.org/protobuf v1.36.11/go.mod h1:HTf+CrKn2C3g5S8VImy6tdcUvCska2kB7j23XfzDpco=
+gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c h1:Hei/4ADfdWqJk1ZMxUNpqntNwaWcugrBjAiHlqqRiVk=
+gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c/go.mod h1:JHkPIbrfpd72SG/EVd6muEfDQjcINNoR0C8j2r3qZ4Q=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=

internal/metrics/metrics.go

@@ -7,7 +7,9 @@ package metrics
 
 import (
 	"context"
+	"fmt"
 	"net/http"
+	"sync"
 
 	"github.com/fosrl/gerbil/internal/observability"
 )
@@ -24,6 +26,7 @@ type OTelConfig = observability.OTelConfig
 
 var (
 	backend observability.Backend
+	initMu  sync.Mutex
 
 	// Interface and peer metrics
 	wgInterfaceUp      observability.Int64Gauge
@@ -55,7 +58,7 @@ var (
 	udpPacketSizeBytes         observability.Histogram
 	holePunchEventsTotal       observability.Counter
 	proxyMappingActive         observability.UpDownCounter
-	sessionActive              observability.UpDownCounter
+	relayUDPConnectionsActive  observability.UpDownCounter
 	sessionRebuiltTotal        observability.Counter
 	commPatternActive          observability.UpDownCounter
 	proxyCleanupRemovedTotal   observability.Counter
@@ -107,6 +110,13 @@ func DefaultConfig() Config {
 // Initialize sets up the metrics system using the selected backend.
 // It returns the /metrics HTTP handler (non-nil only for Prometheus backend).
 func Initialize(cfg Config) (http.Handler, error) {
+	initMu.Lock()
+	defer initMu.Unlock()
+
+	if backend != nil {
+		return backend.HTTPHandler(), nil
+	}
+
 	b, err := observability.New(cfg)
 	if err != nil {
 		return nil, err
@@ -114,6 +124,7 @@ func Initialize(cfg Config) (http.Handler, error) {
 	backend = b
 
 	if err := createInstruments(); err != nil {
+		backend = nil
 		return nil, err
 	}
 
@@ -122,8 +133,13 @@ func Initialize(cfg Config) (http.Handler, error) {
 
 // Shutdown gracefully shuts down the metrics backend.
 func Shutdown(ctx context.Context) error {
-	if backend != nil {
+	initMu.Lock()
-		return backend.Shutdown(ctx)
+	b := backend
+	backend = nil
+	initMu.Unlock()
+
+	if b != nil {
+		return b.Shutdown(ctx)
 	}
 	return nil
 }
@@ -135,129 +151,351 @@ func createInstruments() error {
 
 	b := backend
 
-	wgInterfaceUp = b.NewInt64Gauge("gerbil_wg_interface_up",
+	newCounter := func(name, desc string, labelNames ...string) (observability.Counter, error) {
+		c, err := b.NewCounter(name, desc, labelNames...)
+		if err != nil {
+			return nil, fmt.Errorf("create counter %q: %w", name, err)
+		}
+		return c, nil
+	}
+
+	newUpDownCounter := func(name, desc string, labelNames ...string) (observability.UpDownCounter, error) {
+		c, err := b.NewUpDownCounter(name, desc, labelNames...)
+		if err != nil {
+			return nil, fmt.Errorf("create updown counter %q: %w", name, err)
+		}
+		return c, nil
+	}
+
+	newInt64Gauge := func(name, desc string, labelNames ...string) (observability.Int64Gauge, error) {
+		g, err := b.NewInt64Gauge(name, desc, labelNames...)
+		if err != nil {
+			return nil, fmt.Errorf("create int64 gauge %q: %w", name, err)
+		}
+		return g, nil
+	}
+
+	newFloat64Gauge := func(name, desc string, labelNames ...string) (observability.Float64Gauge, error) {
+		g, err := b.NewFloat64Gauge(name, desc, labelNames...)
+		if err != nil {
+			return nil, fmt.Errorf("create float64 gauge %q: %w", name, err)
+		}
+		return g, nil
+	}
+
+	newHistogram := func(name, desc string, buckets []float64, labelNames ...string) (observability.Histogram, error) {
+		h, err := b.NewHistogram(name, desc, buckets, labelNames...)
+		if err != nil {
+			return nil, fmt.Errorf("create histogram %q: %w", name, err)
+		}
+		return h, nil
+	}
+
+	var err error
+
+	wgInterfaceUp, err = newInt64Gauge("gerbil_wg_interface_up",
 		"Operational state of a WireGuard interface (1=up, 0=down)", "ifname", "instance")
-	wgPeersTotal = b.NewUpDownCounter("gerbil_wg_peers_total",
+	if err != nil {
+		return err
+	}
+	wgPeersTotal, err = newUpDownCounter("gerbil_wg_peers_total",
 		"Total number of configured peers per interface", "ifname")
-	wgPeerConnected = b.NewInt64Gauge("gerbil_wg_peer_connected",
+	if err != nil {
+		return err
+	}
+	wgPeerConnected, err = newInt64Gauge("gerbil_wg_peer_connected",
 		"Whether a specific peer is connected (1=connected, 0=disconnected)", "ifname", "peer")
-	allowedIPsCount = b.NewUpDownCounter("gerbil_allowed_ips_count",
+	if err != nil {
+		return err
+	}
+	allowedIPsCount, err = newUpDownCounter("gerbil_allowed_ips_count",
 		"Number of allowed IPs configured per peer", "ifname", "peer")
-	keyRotationTotal = b.NewCounter("gerbil_key_rotation_total",
+	if err != nil {
+		return err
+	}
+	keyRotationTotal, err = newCounter("gerbil_key_rotation_total",
 		"Key rotation events", "ifname", "reason")
-	wgHandshakesTotal = b.NewCounter("gerbil_wg_handshakes_total",
+	if err != nil {
+		return err
+	}
+	wgHandshakesTotal, err = newCounter("gerbil_wg_handshakes_total",
 		"Count of handshake attempts with their result status", "ifname", "peer", "result")
-	wgHandshakeLatency = b.NewHistogram("gerbil_wg_handshake_latency_seconds",
+	if err != nil {
+		return err
+	}
+	wgHandshakeLatency, err = newHistogram("gerbil_wg_handshake_latency_seconds",
 		"Distribution of handshake latencies in seconds", durationBuckets, "ifname", "peer")
-	wgPeerRTT = b.NewHistogram("gerbil_wg_peer_rtt_seconds",
+	if err != nil {
+		return err
+	}
+	wgPeerRTT, err = newHistogram("gerbil_wg_peer_rtt_seconds",
 		"Observed round-trip time to a peer in seconds", durationBuckets, "ifname", "peer")
-	wgBytesReceived = b.NewCounter("gerbil_wg_bytes_received_total",
+	if err != nil {
+		return err
+	}
+	wgBytesReceived, err = newCounter("gerbil_wg_bytes_received_total",
 		"Number of bytes received from a peer", "ifname", "peer")
-	wgBytesTransmitted = b.NewCounter("gerbil_wg_bytes_transmitted_total",
+	if err != nil {
+		return err
+	}
+	wgBytesTransmitted, err = newCounter("gerbil_wg_bytes_transmitted_total",
 		"Number of bytes transmitted to a peer", "ifname", "peer")
-	netlinkEventsTotal = b.NewCounter("gerbil_netlink_events_total",
+	if err != nil {
+		return err
+	}
+	netlinkEventsTotal, err = newCounter("gerbil_netlink_events_total",
 		"Number of netlink events processed", "event_type")
-	netlinkErrorsTotal = b.NewCounter("gerbil_netlink_errors_total",
+	if err != nil {
+		return err
+	}
+	netlinkErrorsTotal, err = newCounter("gerbil_netlink_errors_total",
 		"Count of netlink or kernel errors", "component", "error_type")
-	syncDuration = b.NewHistogram("gerbil_sync_duration_seconds",
+	if err != nil {
+		return err
+	}
+	syncDuration, err = newHistogram("gerbil_sync_duration_seconds",
 		"Duration of reconciliation/sync loops in seconds", durationBuckets, "component")
-	workqueueDepth = b.NewUpDownCounter("gerbil_workqueue_depth",
+	if err != nil {
+		return err
+	}
+	workqueueDepth, err = newUpDownCounter("gerbil_workqueue_depth",
 		"Current length of internal work queues", "queue")
-	kernelModuleLoads = b.NewCounter("gerbil_kernel_module_loads_total",
+	if err != nil {
+		return err
+	}
+	kernelModuleLoads, err = newCounter("gerbil_kernel_module_loads_total",
 		"Count of kernel module load attempts", "result")
-	firewallRulesApplied = b.NewCounter("gerbil_firewall_rules_applied_total",
+	if err != nil {
+		return err
+	}
+	firewallRulesApplied, err = newCounter("gerbil_firewall_rules_applied_total",
 		"IPTables/NFT rules applied", "result", "chain")
-	activeSessions = b.NewUpDownCounter("gerbil_active_sessions",
+	if err != nil {
+		return err
+	}
+	activeSessions, err = newUpDownCounter("gerbil_active_sessions",
 		"Number of active UDP relay sessions", "ifname")
-	activeProxyConnections = b.NewUpDownCounter("gerbil_active_proxy_connections",
+	if err != nil {
+		return err
+	}
+	activeProxyConnections, err = newUpDownCounter("gerbil_active_proxy_connections",
 		"Active SNI proxy connections")
-	proxyRouteLookups = b.NewCounter("gerbil_proxy_route_lookups_total",
+	if err != nil {
+		return err
+	}
+	proxyRouteLookups, err = newCounter("gerbil_proxy_route_lookups_total",
 		"Number of route lookups", "result")
-	proxyTLSHandshake = b.NewHistogram("gerbil_proxy_tls_handshake_seconds",
+	if err != nil {
+		return err
+	}
+	proxyTLSHandshake, err = newHistogram("gerbil_proxy_tls_handshake_seconds",
 		"TLS handshake duration for SNI proxy in seconds", durationBuckets)
-	proxyBytesTransmitted = b.NewCounter("gerbil_proxy_bytes_transmitted_total",
+	if err != nil {
+		return err
+	}
+	proxyBytesTransmitted, err = newCounter("gerbil_proxy_bytes_transmitted_total",
 		"Bytes sent/received by the SNI proxy", "direction")
-	configReloadsTotal = b.NewCounter("gerbil_config_reloads_total",
+	if err != nil {
+		return err
+	}
+	configReloadsTotal, err = newCounter("gerbil_config_reloads_total",
 		"Number of configuration reloads", "result")
-	restartTotal = b.NewCounter("gerbil_restart_total",
+	if err != nil {
+		return err
+	}
+	restartTotal, err = newCounter("gerbil_restart_total",
 		"Process restart count")
-	authFailuresTotal = b.NewCounter("gerbil_auth_failures_total",
+	if err != nil {
+		return err
+	}
+	authFailuresTotal, err = newCounter("gerbil_auth_failures_total",
 		"Count of authentication or peer validation failures", "peer", "reason")
-	aclDeniedTotal = b.NewCounter("gerbil_acl_denied_total",
+	if err != nil {
+		return err
+	}
+	aclDeniedTotal, err = newCounter("gerbil_acl_denied_total",
 		"Access control denied events", "ifname", "peer", "policy")
-	certificateExpiryDays = b.NewFloat64Gauge("gerbil_certificate_expiry_days",
+	if err != nil {
+		return err
+	}
+	certificateExpiryDays, err = newFloat64Gauge("gerbil_certificate_expiry_days",
 		"Days until certificate expiry", "cert_name", "ifname")
-	udpPacketsTotal = b.NewCounter("gerbil_udp_packets_total",
+	if err != nil {
+		return err
+	}
+	udpPacketsTotal, err = newCounter("gerbil_udp_packets_total",
 		"Count of UDP packets processed by relay workers", "ifname", "type", "direction")
-	udpPacketSizeBytes = b.NewHistogram("gerbil_udp_packet_size_bytes",
+	if err != nil {
+		return err
+	}
+	udpPacketSizeBytes, err = newHistogram("gerbil_udp_packet_size_bytes",
 		"Size distribution of packets forwarded through relay", sizeBuckets, "ifname", "type")
-	holePunchEventsTotal = b.NewCounter("gerbil_hole_punch_events_total",
+	if err != nil {
+		return err
+	}
+	holePunchEventsTotal, err = newCounter("gerbil_hole_punch_events_total",
 		"Count of hole punch messages processed", "ifname", "result")
-	proxyMappingActive = b.NewUpDownCounter("gerbil_proxy_mapping_active",
+	if err != nil {
+		return err
+	}
+	proxyMappingActive, err = newUpDownCounter("gerbil_proxy_mapping_active",
 		"Number of active proxy mappings", "ifname")
-	sessionActive = b.NewUpDownCounter("gerbil_session_active",
+	if err != nil {
-		"Number of active WireGuard sessions", "ifname")
+		return err
-	sessionRebuiltTotal = b.NewCounter("gerbil_session_rebuilt_total",
+	}
+	relayUDPConnectionsActive, err = newUpDownCounter("gerbil_relay_udp_connections_active",
+		"Number of open per-peer outbound UDP sockets held by the relay connection pool", "ifname")
+	if err != nil {
+		return err
+	}
+	sessionRebuiltTotal, err = newCounter("gerbil_session_rebuilt_total",
 		"Count of sessions rebuilt from communication patterns", "ifname")
-	commPatternActive = b.NewUpDownCounter("gerbil_comm_pattern_active",
+	if err != nil {
+		return err
+	}
+	commPatternActive, err = newUpDownCounter("gerbil_comm_pattern_active",
 		"Number of active communication patterns", "ifname")
-	proxyCleanupRemovedTotal = b.NewCounter("gerbil_proxy_cleanup_removed_total",
+	if err != nil {
+		return err
+	}
+	proxyCleanupRemovedTotal, err = newCounter("gerbil_proxy_cleanup_removed_total",
 		"Count of items removed during cleanup routines", "ifname", "component")
-	proxyConnectionErrorsTotal = b.NewCounter("gerbil_proxy_connection_errors_total",
+	if err != nil {
+		return err
+	}
+	proxyConnectionErrorsTotal, err = newCounter("gerbil_proxy_connection_errors_total",
 		"Count of connection errors in proxy operations", "ifname", "error_type")
-	proxyInitialMappingsTotal = b.NewInt64Gauge("gerbil_proxy_initial_mappings",
+	if err != nil {
+		return err
+	}
+	proxyInitialMappingsTotal, err = newInt64Gauge("gerbil_proxy_initial_mappings",
 		"Number of initial proxy mappings loaded", "ifname")
-	proxyMappingUpdatesTotal = b.NewCounter("gerbil_proxy_mapping_updates_total",
+	if err != nil {
+		return err
+	}
+	proxyMappingUpdatesTotal, err = newCounter("gerbil_proxy_mapping_updates_total",
 		"Count of proxy mapping updates", "ifname")
-	proxyIdleCleanupDuration = b.NewHistogram("gerbil_proxy_idle_cleanup_duration_seconds",
+	if err != nil {
+		return err
+	}
+	proxyIdleCleanupDuration, err = newHistogram("gerbil_proxy_idle_cleanup_duration_seconds",
 		"Duration of cleanup cycles", durationBuckets, "ifname", "component")
-	sniConnectionsTotal = b.NewCounter("gerbil_sni_connections_total",
+	if err != nil {
+		return err
+	}
+	sniConnectionsTotal, err = newCounter("gerbil_sni_connections_total",
 		"Count of connections processed by SNI proxy", "result")
-	sniConnectionDuration = b.NewHistogram("gerbil_sni_connection_duration_seconds",
+	if err != nil {
+		return err
+	}
+	sniConnectionDuration, err = newHistogram("gerbil_sni_connection_duration_seconds",
 		"Lifetime distribution of proxied TLS connections", sniDurationBuckets)
-	sniActiveConnections = b.NewUpDownCounter("gerbil_sni_active_connections",
+	if err != nil {
+		return err
+	}
+	sniActiveConnections, err = newUpDownCounter("gerbil_sni_active_connections",
 		"Number of active SNI tunnels")
-	sniRouteCacheHitsTotal = b.NewCounter("gerbil_sni_route_cache_hits_total",
+	if err != nil {
+		return err
+	}
+	sniRouteCacheHitsTotal, err = newCounter("gerbil_sni_route_cache_hits_total",
 		"Count of route cache hits and misses", "result")
-	sniRouteAPIRequestsTotal = b.NewCounter("gerbil_sni_route_api_requests_total",
+	if err != nil {
+		return err
+	}
+	sniRouteAPIRequestsTotal, err = newCounter("gerbil_sni_route_api_requests_total",
 		"Count of route API requests", "result")
-	sniRouteAPILatency = b.NewHistogram("gerbil_sni_route_api_latency_seconds",
+	if err != nil {
+		return err
+	}
+	sniRouteAPILatency, err = newHistogram("gerbil_sni_route_api_latency_seconds",
 		"Distribution of route API call latencies", durationBuckets)
-	sniLocalOverrideTotal = b.NewCounter("gerbil_sni_local_override_total",
+	if err != nil {
+		return err
+	}
+	sniLocalOverrideTotal, err = newCounter("gerbil_sni_local_override_total",
 		"Count of routes using local overrides", "hit")
-	sniTrustedProxyEventsTotal = b.NewCounter("gerbil_sni_trusted_proxy_events_total",
+	if err != nil {
+		return err
+	}
+	sniTrustedProxyEventsTotal, err = newCounter("gerbil_sni_trusted_proxy_events_total",
 		"Count of PROXY protocol events", "event")
-	sniProxyProtocolParseErrorsTotal = b.NewCounter("gerbil_sni_proxy_protocol_parse_errors_total",
+	if err != nil {
+		return err
+	}
+	sniProxyProtocolParseErrorsTotal, err = newCounter("gerbil_sni_proxy_protocol_parse_errors_total",
 		"Count of PROXY protocol parse failures")
-	sniDataBytesTotal = b.NewCounter("gerbil_sni_data_bytes_total",
+	if err != nil {
+		return err
+	}
+	sniDataBytesTotal, err = newCounter("gerbil_sni_data_bytes_total",
 		"Count of bytes proxied through SNI tunnels", "direction")
-	sniTunnelTerminationsTotal = b.NewCounter("gerbil_sni_tunnel_terminations_total",
+	if err != nil {
+		return err
+	}
+	sniTunnelTerminationsTotal, err = newCounter("gerbil_sni_tunnel_terminations_total",
 		"Count of tunnel terminations by reason", "reason")
-	httpRequestsTotal = b.NewCounter("gerbil_http_requests_total",
+	if err != nil {
+		return err
+	}
+	httpRequestsTotal, err = newCounter("gerbil_http_requests_total",
 		"Count of HTTP requests to management API", "endpoint", "method", "status_code")
-	httpRequestDuration = b.NewHistogram("gerbil_http_request_duration_seconds",
+	if err != nil {
+		return err
+	}
+	httpRequestDuration, err = newHistogram("gerbil_http_request_duration_seconds",
 		"Distribution of HTTP request handling time", durationBuckets, "endpoint", "method")
-	peerOperationsTotal = b.NewCounter("gerbil_peer_operations_total",
+	if err != nil {
+		return err
+	}
+	peerOperationsTotal, err = newCounter("gerbil_peer_operations_total",
 		"Count of peer lifecycle operations", "operation", "result")
-	proxyMappingUpdateRequestsTotal = b.NewCounter("gerbil_proxy_mapping_update_requests_total",
+	if err != nil {
+		return err
+	}
+	proxyMappingUpdateRequestsTotal, err = newCounter("gerbil_proxy_mapping_update_requests_total",
 		"Count of proxy mapping update API calls", "result")
-	destinationsUpdateRequestsTotal = b.NewCounter("gerbil_destinations_update_requests_total",
+	if err != nil {
+		return err
+	}
+	destinationsUpdateRequestsTotal, err = newCounter("gerbil_destinations_update_requests_total",
 		"Count of destinations update API calls", "result")
-	remoteConfigFetchesTotal = b.NewCounter("gerbil_remote_config_fetches_total",
+	if err != nil {
+		return err
+	}
+	remoteConfigFetchesTotal, err = newCounter("gerbil_remote_config_fetches_total",
 		"Count of remote configuration fetch attempts", "result")
-	bandwidthReportsTotal = b.NewCounter("gerbil_bandwidth_reports_total",
+	if err != nil {
+		return err
+	}
+	bandwidthReportsTotal, err = newCounter("gerbil_bandwidth_reports_total",
 		"Count of bandwidth report transmissions", "result")
-	peerBandwidthBytesTotal = b.NewCounter("gerbil_peer_bandwidth_bytes_total",
+	if err != nil {
+		return err
+	}
+	peerBandwidthBytesTotal, err = newCounter("gerbil_peer_bandwidth_bytes_total",
 		"Bytes per peer tracked by bandwidth calculation", "peer", "direction")
-	memorySpikeTotal = b.NewCounter("gerbil_memory_spike_total",
+	if err != nil {
+		return err
+	}
+	memorySpikeTotal, err = newCounter("gerbil_memory_spike_total",
 		"Count of memory spikes detected", "severity")
-	heapProfilesWrittenTotal = b.NewCounter("gerbil_heap_profiles_written_total",
+	if err != nil {
+		return err
+	}
+	heapProfilesWrittenTotal, err = newCounter("gerbil_heap_profiles_written_total",
 		"Count of heap profile files generated")
+	if err != nil {
+		return err
+	}
 
 	return nil
 }
 
 func RecordInterfaceUp(ifname, instance string, up bool) {
+	if wgInterfaceUp == nil {
+		return
+	}
 	value := int64(0)
 	if up {
 		value = 1
@@ -266,10 +504,16 @@ func RecordInterfaceUp(ifname, instance string, up bool) {
 }
 
 func RecordPeersTotal(ifname string, delta int64) {
+	if wgPeersTotal == nil {
+		return
+	}
 	wgPeersTotal.Add(context.Background(), delta, observability.Labels{"ifname": ifname})
 }
 
 func RecordPeerConnected(ifname, peer string, connected bool) {
+	if wgPeerConnected == nil {
+		return
+	}
 	value := int64(0)
 	if connected {
 		value = 1
@@ -278,229 +522,400 @@ func RecordPeerConnected(ifname, peer string, connected bool) {
 }
 
 func RecordHandshake(ifname, peer, result string) {
+	if wgHandshakesTotal == nil {
+		return
+	}
 	wgHandshakesTotal.Add(context.Background(), 1, observability.Labels{"ifname": ifname, "peer": peer, "result": result})
 }
 
 func RecordHandshakeLatency(ifname, peer string, seconds float64) {
+	if wgHandshakeLatency == nil {
+		return
+	}
 	wgHandshakeLatency.Record(context.Background(), seconds, observability.Labels{"ifname": ifname, "peer": peer})
 }
 
 func RecordPeerRTT(ifname, peer string, seconds float64) {
+	if wgPeerRTT == nil {
+		return
+	}
 	wgPeerRTT.Record(context.Background(), seconds, observability.Labels{"ifname": ifname, "peer": peer})
 }
 
 func RecordBytesReceived(ifname, peer string, bytes int64) {
+	if wgBytesReceived == nil {
+		return
+	}
 	wgBytesReceived.Add(context.Background(), bytes, observability.Labels{"ifname": ifname, "peer": peer})
 }
 
 func RecordBytesTransmitted(ifname, peer string, bytes int64) {
+	if wgBytesTransmitted == nil {
+		return
+	}
 	wgBytesTransmitted.Add(context.Background(), bytes, observability.Labels{"ifname": ifname, "peer": peer})
 }
 
 func RecordAllowedIPsCount(ifname, peer string, delta int64) {
+	if allowedIPsCount == nil {
+		return
+	}
 	allowedIPsCount.Add(context.Background(), delta, observability.Labels{"ifname": ifname, "peer": peer})
 }
 
 func RecordKeyRotation(ifname, reason string) {
+	if keyRotationTotal == nil {
+		return
+	}
 	keyRotationTotal.Add(context.Background(), 1, observability.Labels{"ifname": ifname, "reason": reason})
 }
 
 func RecordNetlinkEvent(eventType string) {
+	if netlinkEventsTotal == nil {
+		return
+	}
 	netlinkEventsTotal.Add(context.Background(), 1, observability.Labels{"event_type": eventType})
 }
 
 func RecordNetlinkError(component, errorType string) {
+	if netlinkErrorsTotal == nil {
+		return
+	}
 	netlinkErrorsTotal.Add(context.Background(), 1, observability.Labels{"component": component, "error_type": errorType})
 }
 
 func RecordSyncDuration(component string, seconds float64) {
+	if syncDuration == nil {
+		return
+	}
 	syncDuration.Record(context.Background(), seconds, observability.Labels{"component": component})
 }
 
 func RecordWorkqueueDepth(queue string, delta int64) {
+	if workqueueDepth == nil {
+		return
+	}
 	workqueueDepth.Add(context.Background(), delta, observability.Labels{"queue": queue})
 }
 
 func RecordKernelModuleLoad(result string) {
+	if kernelModuleLoads == nil {
+		return
+	}
 	kernelModuleLoads.Add(context.Background(), 1, observability.Labels{"result": result})
 }
 
 func RecordFirewallRuleApplied(result, chain string) {
+	if firewallRulesApplied == nil {
+		return
+	}
 	firewallRulesApplied.Add(context.Background(), 1, observability.Labels{"result": result, "chain": chain})
 }
 
 func RecordActiveSession(ifname string, delta int64) {
+	if activeSessions == nil {
+		return
+	}
 	activeSessions.Add(context.Background(), delta, observability.Labels{"ifname": ifname})
 }
 
-func RecordActiveProxyConnection(hostname string, delta int64) {
+func RecordActiveProxyConnection(delta int64) {
-	_ = hostname
+	if activeProxyConnections == nil {
+		return
+	}
 	activeProxyConnections.Add(context.Background(), delta, nil)
 }
 
-func RecordProxyRouteLookup(result, hostname string) {
+func RecordProxyRouteLookup(result string) {
-	_ = hostname
+	if proxyRouteLookups == nil {
+		return
+	}
 	proxyRouteLookups.Add(context.Background(), 1, observability.Labels{"result": result})
 }
 
-func RecordProxyTLSHandshake(hostname string, seconds float64) {
+func RecordProxyTLSHandshake(seconds float64) {
-	_ = hostname
+	if proxyTLSHandshake == nil {
+		return
+	}
 	proxyTLSHandshake.Record(context.Background(), seconds, nil)
 }
 
-func RecordProxyBytesTransmitted(hostname, direction string, bytes int64) {
+func RecordProxyBytesTransmitted(direction string, bytes int64) {
-	_ = hostname
+	if proxyBytesTransmitted == nil {
+		return
+	}
 	proxyBytesTransmitted.Add(context.Background(), bytes, observability.Labels{"direction": direction})
 }
 
 func RecordConfigReload(result string) {
+	if configReloadsTotal == nil {
+		return
+	}
 	configReloadsTotal.Add(context.Background(), 1, observability.Labels{"result": result})
 }
 
 func RecordRestart() {
+	if restartTotal == nil {
+		return
+	}
 	restartTotal.Add(context.Background(), 1, nil)
 }
 
 func RecordAuthFailure(peer, reason string) {
+	if authFailuresTotal == nil {
+		return
+	}
 	authFailuresTotal.Add(context.Background(), 1, observability.Labels{"peer": peer, "reason": reason})
 }
 
 func RecordACLDenied(ifname, peer, policy string) {
+	if aclDeniedTotal == nil {
+		return
+	}
 	aclDeniedTotal.Add(context.Background(), 1, observability.Labels{"ifname": ifname, "peer": peer, "policy": policy})
 }
 
 func RecordCertificateExpiry(certName, ifname string, days float64) {
+	if certificateExpiryDays == nil {
+		return
+	}
 	certificateExpiryDays.Record(context.Background(), days, observability.Labels{"cert_name": certName, "ifname": ifname})
 }
 
 func RecordUDPPacket(ifname, packetType, direction string) {
+	if udpPacketsTotal == nil {
+		return
+	}
 	udpPacketsTotal.Add(context.Background(), 1, observability.Labels{"ifname": ifname, "type": packetType, "direction": direction})
 }
 
 func RecordUDPPacketSize(ifname, packetType string, bytes float64) {
+	if udpPacketSizeBytes == nil {
+		return
+	}
 	udpPacketSizeBytes.Record(context.Background(), bytes, observability.Labels{"ifname": ifname, "type": packetType})
 }
 
 func RecordHolePunchEvent(ifname, result string) {
+	if holePunchEventsTotal == nil {
+		return
+	}
 	holePunchEventsTotal.Add(context.Background(), 1, observability.Labels{"ifname": ifname, "result": result})
 }
 
 func RecordProxyMapping(ifname string, delta int64) {
+	if proxyMappingActive == nil {
+		return
+	}
 	proxyMappingActive.Add(context.Background(), delta, observability.Labels{"ifname": ifname})
 }
 
 func RecordSession(ifname string, delta int64) {
-	sessionActive.Add(context.Background(), delta, observability.Labels{"ifname": ifname})
+	if activeSessions == nil {
+		return
+	}
+	activeSessions.Add(context.Background(), delta, observability.Labels{"ifname": ifname})
+}
+
+func RecordUDPConnection(ifname string, delta int64) {
+	if relayUDPConnectionsActive == nil {
+		return
+	}
+	relayUDPConnectionsActive.Add(context.Background(), delta, observability.Labels{"ifname": ifname})
 }
 
 func RecordSessionRebuilt(ifname string) {
+	if sessionRebuiltTotal == nil {
+		return
+	}
 	sessionRebuiltTotal.Add(context.Background(), 1, observability.Labels{"ifname": ifname})
 }
 
 func RecordCommPattern(ifname string, delta int64) {
+	if commPatternActive == nil {
+		return
+	}
 	commPatternActive.Add(context.Background(), delta, observability.Labels{"ifname": ifname})
 }
 
 func RecordProxyCleanupRemoved(ifname, component string, count int64) {
+	if proxyCleanupRemovedTotal == nil {
+		return
+	}
 	proxyCleanupRemovedTotal.Add(context.Background(), count, observability.Labels{"ifname": ifname, "component": component})
 }
 
 func RecordProxyConnectionError(ifname, errorType string) {
+	if proxyConnectionErrorsTotal == nil {
+		return
+	}
 	proxyConnectionErrorsTotal.Add(context.Background(), 1, observability.Labels{"ifname": ifname, "error_type": errorType})
 }
 
 func RecordProxyInitialMappings(ifname string, count int64) {
+	if proxyInitialMappingsTotal == nil {
+		return
+	}
 	proxyInitialMappingsTotal.Record(context.Background(), count, observability.Labels{"ifname": ifname})
 }
 
 func RecordProxyMappingUpdate(ifname string) {
+	if proxyMappingUpdatesTotal == nil {
+		return
+	}
 	proxyMappingUpdatesTotal.Add(context.Background(), 1, observability.Labels{"ifname": ifname})
 }
 
 func RecordProxyIdleCleanupDuration(ifname, component string, seconds float64) {
+	if proxyIdleCleanupDuration == nil {
+		return
+	}
 	proxyIdleCleanupDuration.Record(context.Background(), seconds, observability.Labels{"ifname": ifname, "component": component})
 }
 
 func RecordSNIConnection(result string) {
+	if sniConnectionsTotal == nil {
+		return
+	}
 	sniConnectionsTotal.Add(context.Background(), 1, observability.Labels{"result": result})
 }
 
 func RecordSNIConnectionDuration(seconds float64) {
+	if sniConnectionDuration == nil {
+		return
+	}
 	sniConnectionDuration.Record(context.Background(), seconds, nil)
 }
 
 func RecordSNIActiveConnection(delta int64) {
+	if sniActiveConnections == nil {
+		return
+	}
 	sniActiveConnections.Add(context.Background(), delta, nil)
 }
 
 func RecordSNIRouteCacheHit(result string) {
+	if sniRouteCacheHitsTotal == nil {
+		return
+	}
 	sniRouteCacheHitsTotal.Add(context.Background(), 1, observability.Labels{"result": result})
 }
 
 func RecordSNIRouteAPIRequest(result string) {
+	if sniRouteAPIRequestsTotal == nil {
+		return
+	}
 	sniRouteAPIRequestsTotal.Add(context.Background(), 1, observability.Labels{"result": result})
 }
 
 func RecordSNIRouteAPILatency(seconds float64) {
+	if sniRouteAPILatency == nil {
+		return
+	}
 	sniRouteAPILatency.Record(context.Background(), seconds, nil)
 }
 
 func RecordSNILocalOverride(hit string) {
+	if sniLocalOverrideTotal == nil {
+		return
+	}
 	sniLocalOverrideTotal.Add(context.Background(), 1, observability.Labels{"hit": hit})
 }
 
 func RecordSNITrustedProxyEvent(event string) {
+	if sniTrustedProxyEventsTotal == nil {
+		return
+	}
 	sniTrustedProxyEventsTotal.Add(context.Background(), 1, observability.Labels{"event": event})
 }
 
 func RecordSNIProxyProtocolParseError() {
+	if sniProxyProtocolParseErrorsTotal == nil {
+		return
+	}
 	sniProxyProtocolParseErrorsTotal.Add(context.Background(), 1, nil)
 }
 
 func RecordSNIDataBytes(direction string, bytes int64) {
+	if sniDataBytesTotal == nil {
+		return
+	}
 	sniDataBytesTotal.Add(context.Background(), bytes, observability.Labels{"direction": direction})
 }
 
 func RecordSNITunnelTermination(reason string) {
+	if sniTunnelTerminationsTotal == nil {
+		return
+	}
 	sniTunnelTerminationsTotal.Add(context.Background(), 1, observability.Labels{"reason": reason})
 }
 
 func RecordHTTPRequest(endpoint, method, statusCode string) {
+	if httpRequestsTotal == nil {
+		return
+	}
 	httpRequestsTotal.Add(context.Background(), 1, observability.Labels{"endpoint": endpoint, "method": method, "status_code": statusCode})
 }
 
 func RecordHTTPRequestDuration(endpoint, method string, seconds float64) {
+	if httpRequestDuration == nil {
+		return
+	}
 	httpRequestDuration.Record(context.Background(), seconds, observability.Labels{"endpoint": endpoint, "method": method})
 }
 
 func RecordPeerOperation(operation, result string) {
+	if peerOperationsTotal == nil {
+		return
+	}
 	peerOperationsTotal.Add(context.Background(), 1, observability.Labels{"operation": operation, "result": result})
 }
 
 func RecordProxyMappingUpdateRequest(result string) {
+	if proxyMappingUpdateRequestsTotal == nil {
+		return
+	}
 	proxyMappingUpdateRequestsTotal.Add(context.Background(), 1, observability.Labels{"result": result})
 }
 
 func RecordDestinationsUpdateRequest(result string) {
+	if destinationsUpdateRequestsTotal == nil {
+		return
+	}
 	destinationsUpdateRequestsTotal.Add(context.Background(), 1, observability.Labels{"result": result})
 }
 
 func RecordRemoteConfigFetch(result string) {
+	if remoteConfigFetchesTotal == nil {
+		return
+	}
 	remoteConfigFetchesTotal.Add(context.Background(), 1, observability.Labels{"result": result})
 }
 
 func RecordBandwidthReport(result string) {
+	if bandwidthReportsTotal == nil {
+		return
+	}
 	bandwidthReportsTotal.Add(context.Background(), 1, observability.Labels{"result": result})
 }
 
 func RecordPeerBandwidthBytes(peer, direction string, bytes int64) {
+	if peerBandwidthBytesTotal == nil {
+		return
+	}
 	peerBandwidthBytesTotal.Add(context.Background(), bytes, observability.Labels{"peer": peer, "direction": direction})
 }
 
 func RecordMemorySpike(severity string) {
+	if memorySpikeTotal == nil {
+		return
+	}
 	memorySpikeTotal.Add(context.Background(), 1, observability.Labels{"severity": severity})
 }
 
 func RecordHeapProfileWritten() {
+	if heapProfilesWrittenTotal == nil {
+		return
+	}
 	heapProfilesWrittenTotal.Add(context.Background(), 1, nil)
 }

internal/metrics/metrics_test.go

@@ -89,6 +89,9 @@ func TestDefaultConfig(t *testing.T) {
 }
 
 func TestShutdownNoInit(t *testing.T) {
+	// Ensure a known clean global state before testing no-init shutdown behavior.
+	_ = metrics.Shutdown(context.Background())
+
 	// Shutdown without Initialize should not panic or error.
 	if err := metrics.Shutdown(context.Background()); err != nil {
 		t.Errorf("unexpected error: %v", err)
@@ -168,6 +171,7 @@ func TestRecordRelay(t *testing.T) {
 	body := scrape(t, h)
 	assertContains(t, body, "gerbil_udp_packets_total")
 	assertContains(t, body, "gerbil_proxy_mapping_active")
+	assertContains(t, body, "gerbil_active_sessions")
 }
 
 func TestRecordWireGuard(t *testing.T) {
@@ -216,10 +220,10 @@ func TestRecordNetlink(t *testing.T) {
 	metrics.RecordKernelModuleLoad("success")
 	metrics.RecordFirewallRuleApplied("success", "INPUT")
 	metrics.RecordActiveSession("wg0", 1)
-	metrics.RecordActiveProxyConnection(exampleHostname, 1)
+	metrics.RecordActiveProxyConnection(1)
-	metrics.RecordProxyRouteLookup("hit", exampleHostname)
+	metrics.RecordProxyRouteLookup("hit")
-	metrics.RecordProxyTLSHandshake(exampleHostname, 0.05)
+	metrics.RecordProxyTLSHandshake(0.05)
-	metrics.RecordProxyBytesTransmitted(exampleHostname, "tx", 1024)
+	metrics.RecordProxyBytesTransmitted("tx", 1024)
 	body := scrape(t, h)
 	assertContains(t, body, "gerbil_netlink_events_total")
 	assertContains(t, body, "gerbil_active_sessions")

internal/observability/config.go

@@ -60,6 +60,10 @@ type OTelConfig struct {
 	// ExportInterval is how often metrics are pushed to the collector.
 	// Defaults to 60 s.
 	ExportInterval time.Duration
+
+	// Timeout bounds OTLP exporter construction calls.
+	// Defaults to 10 s.
+	Timeout time.Duration
 }
 
 // DefaultMetricsConfig returns a MetricsConfig with sensible defaults.
@@ -75,6 +79,7 @@ func DefaultMetricsConfig() MetricsConfig {
 			Endpoint:       "localhost:4317",
 			Insecure:       true,
 			ExportInterval: 60 * time.Second,
+			Timeout:        10 * time.Second,
 		},
 		ServiceName:    "gerbil",
 		ServiceVersion: "1.0.0",
@@ -88,8 +93,10 @@ func (c *MetricsConfig) Validate() error {
 	}
 
 	switch c.Backend {
-	case "prometheus", "none", "":
+	case "prometheus", "none":
 		// valid
+	case "":
+		return fmt.Errorf("metrics: enabled requires a non-empty backend")
 	case "otel":
 		if c.OTel.Endpoint == "" {
 			return fmt.Errorf("metrics: backend=otel requires a non-empty OTel endpoint")
@@ -100,6 +107,9 @@ func (c *MetricsConfig) Validate() error {
 		if c.OTel.ExportInterval <= 0 {
 			return fmt.Errorf("metrics: otel export interval must be positive")
 		}
+		if c.OTel.Timeout <= 0 {
+			return fmt.Errorf("metrics: otel timeout must be positive")
+		}
 	default:
 		return fmt.Errorf("metrics: unknown backend %q (must be \"prometheus\", \"otel\", or \"none\")", c.Backend)
 	}

internal/observability/metrics.go

@@ -43,20 +43,20 @@ type Histogram interface {
 type Backend interface {
 	// NewCounter creates a counter metric.
 	// labelNames declares the set of label keys that will be passed at observation time.
-	NewCounter(name, desc string, labelNames ...string) Counter
+	NewCounter(name, desc string, labelNames ...string) (Counter, error)
 
 	// NewUpDownCounter creates an up-down counter metric.
-	NewUpDownCounter(name, desc string, labelNames ...string) UpDownCounter
+	NewUpDownCounter(name, desc string, labelNames ...string) (UpDownCounter, error)
 
 	// NewInt64Gauge creates an integer gauge metric.
-	NewInt64Gauge(name, desc string, labelNames ...string) Int64Gauge
+	NewInt64Gauge(name, desc string, labelNames ...string) (Int64Gauge, error)
 
 	// NewFloat64Gauge creates a float gauge metric.
-	NewFloat64Gauge(name, desc string, labelNames ...string) Float64Gauge
+	NewFloat64Gauge(name, desc string, labelNames ...string) (Float64Gauge, error)
 
 	// NewHistogram creates a histogram metric.
 	// buckets are the explicit upper-bound bucket boundaries.
-	NewHistogram(name, desc string, buckets []float64, labelNames ...string) Histogram
+	NewHistogram(name, desc string, buckets []float64, labelNames ...string) (Histogram, error)
 
 	// HTTPHandler returns the /metrics HTTP handler.
 	// Implementations that do not expose an HTTP endpoint return nil.
@@ -88,6 +88,7 @@ func New(cfg MetricsConfig) (Backend, error) {
 			Endpoint:              cfg.OTel.Endpoint,
 			Insecure:              cfg.OTel.Insecure,
 			ExportInterval:        cfg.OTel.ExportInterval,
+			Timeout:               cfg.OTel.Timeout,
 			ServiceName:           cfg.ServiceName,
 			ServiceVersion:        cfg.ServiceVersion,
 			DeploymentEnvironment: cfg.DeploymentEnvironment,
@@ -110,19 +111,19 @@ type promAdapter struct {
 	b *obsprom.Backend
 }
 
-func (a *promAdapter) NewCounter(name, desc string, labelNames ...string) Counter {
+func (a *promAdapter) NewCounter(name, desc string, labelNames ...string) (Counter, error) {
 	return a.b.NewCounter(name, desc, labelNames...)
 }
-func (a *promAdapter) NewUpDownCounter(name, desc string, labelNames ...string) UpDownCounter {
+func (a *promAdapter) NewUpDownCounter(name, desc string, labelNames ...string) (UpDownCounter, error) {
 	return a.b.NewUpDownCounter(name, desc, labelNames...)
 }
-func (a *promAdapter) NewInt64Gauge(name, desc string, labelNames ...string) Int64Gauge {
+func (a *promAdapter) NewInt64Gauge(name, desc string, labelNames ...string) (Int64Gauge, error) {
 	return a.b.NewInt64Gauge(name, desc, labelNames...)
 }
-func (a *promAdapter) NewFloat64Gauge(name, desc string, labelNames ...string) Float64Gauge {
+func (a *promAdapter) NewFloat64Gauge(name, desc string, labelNames ...string) (Float64Gauge, error) {
 	return a.b.NewFloat64Gauge(name, desc, labelNames...)
 }
-func (a *promAdapter) NewHistogram(name, desc string, buckets []float64, labelNames ...string) Histogram {
+func (a *promAdapter) NewHistogram(name, desc string, buckets []float64, labelNames ...string) (Histogram, error) {
 	return a.b.NewHistogram(name, desc, buckets, labelNames...)
 }
 func (a *promAdapter) HTTPHandler() http.Handler          { return a.b.HTTPHandler() }
@@ -133,19 +134,19 @@ type otelAdapter struct {
 	b *obsotel.Backend
 }
 
-func (a *otelAdapter) NewCounter(name, desc string, labelNames ...string) Counter {
+func (a *otelAdapter) NewCounter(name, desc string, labelNames ...string) (Counter, error) {
 	return a.b.NewCounter(name, desc, labelNames...)
 }
-func (a *otelAdapter) NewUpDownCounter(name, desc string, labelNames ...string) UpDownCounter {
+func (a *otelAdapter) NewUpDownCounter(name, desc string, labelNames ...string) (UpDownCounter, error) {
 	return a.b.NewUpDownCounter(name, desc, labelNames...)
 }
-func (a *otelAdapter) NewInt64Gauge(name, desc string, labelNames ...string) Int64Gauge {
+func (a *otelAdapter) NewInt64Gauge(name, desc string, labelNames ...string) (Int64Gauge, error) {
 	return a.b.NewInt64Gauge(name, desc, labelNames...)
 }
-func (a *otelAdapter) NewFloat64Gauge(name, desc string, labelNames ...string) Float64Gauge {
+func (a *otelAdapter) NewFloat64Gauge(name, desc string, labelNames ...string) (Float64Gauge, error) {
 	return a.b.NewFloat64Gauge(name, desc, labelNames...)
 }
-func (a *otelAdapter) NewHistogram(name, desc string, buckets []float64, labelNames ...string) Histogram {
+func (a *otelAdapter) NewHistogram(name, desc string, buckets []float64, labelNames ...string) (Histogram, error) {
 	return a.b.NewHistogram(name, desc, buckets, labelNames...)
 }
 func (a *otelAdapter) HTTPHandler() http.Handler          { return a.b.HTTPHandler() }

internal/observability/metrics_test.go

@@ -2,6 +2,8 @@ package observability_test
 
 import (
 	"context"
+	"net"
+	"os"
 	"testing"
 	"time"
 
@@ -39,20 +41,19 @@ func TestValidateValidConfigs(t *testing.T) {
 	}{
 		{name: "disabled", cfg: observability.MetricsConfig{Enabled: false}},
 		{name: "backend none", cfg: observability.MetricsConfig{Enabled: true, Backend: "none"}},
-		{name: "backend empty", cfg: observability.MetricsConfig{Enabled: true, Backend: ""}},
 		{name: "prometheus", cfg: observability.MetricsConfig{Enabled: true, Backend: "prometheus"}},
 		{
 			name: "otel grpc",
 			cfg: observability.MetricsConfig{
 				Enabled: true, Backend: "otel",
-				OTel: observability.OTelConfig{Protocol: "grpc", Endpoint: otelGRPCEndpoint, ExportInterval: 10 * time.Second},
+				OTel: observability.OTelConfig{Protocol: "grpc", Endpoint: otelGRPCEndpoint, ExportInterval: 10 * time.Second, Timeout: 2 * time.Second},
 			},
 		},
 		{
 			name: "otel http",
 			cfg: observability.MetricsConfig{
 				Enabled: true, Backend: "otel",
-				OTel: observability.OTelConfig{Protocol: "http", Endpoint: "localhost:4318", ExportInterval: 30 * time.Second},
+				OTel: observability.OTelConfig{Protocol: "http", Endpoint: "localhost:4318", ExportInterval: 30 * time.Second, Timeout: 2 * time.Second},
 			},
 		},
 	}
@@ -71,25 +72,36 @@ func TestValidateInvalidConfigs(t *testing.T) {
 		cfg  observability.MetricsConfig
 	}{
 		{name: "unknown backend", cfg: observability.MetricsConfig{Enabled: true, Backend: "datadog"}},
+		{
+			name: "backend empty while enabled",
+			cfg:  observability.MetricsConfig{Enabled: true, Backend: ""},
+		},
 		{
 			name: "otel missing endpoint",
 			cfg: observability.MetricsConfig{
 				Enabled: true, Backend: "otel",
-				OTel: observability.OTelConfig{Protocol: "grpc", Endpoint: "", ExportInterval: 10 * time.Second},
+				OTel: observability.OTelConfig{Protocol: "grpc", Endpoint: "", ExportInterval: 10 * time.Second, Timeout: 2 * time.Second},
 			},
 		},
 		{
 			name: "otel invalid protocol",
 			cfg: observability.MetricsConfig{
 				Enabled: true, Backend: "otel",
-				OTel: observability.OTelConfig{Protocol: "tcp", Endpoint: otelGRPCEndpoint, ExportInterval: 10 * time.Second},
+				OTel: observability.OTelConfig{Protocol: "tcp", Endpoint: otelGRPCEndpoint, ExportInterval: 10 * time.Second, Timeout: 2 * time.Second},
 			},
 		},
 		{
 			name: "otel zero interval",
 			cfg: observability.MetricsConfig{
 				Enabled: true, Backend: "otel",
-				OTel: observability.OTelConfig{Protocol: "grpc", Endpoint: otelGRPCEndpoint, ExportInterval: 0},
+				OTel: observability.OTelConfig{Protocol: "grpc", Endpoint: otelGRPCEndpoint, ExportInterval: 0, Timeout: 2 * time.Second},
+			},
+		},
+		{
+			name: "otel zero timeout",
+			cfg: observability.MetricsConfig{
+				Enabled: true, Backend: "otel",
+				OTel: observability.OTelConfig{Protocol: "grpc", Endpoint: otelGRPCEndpoint, ExportInterval: 10 * time.Second, Timeout: 0},
 			},
 		},
 	}
@@ -157,11 +169,32 @@ func TestPrometheusAdapterAllInstruments(t *testing.T) {
 	ctx := context.Background()
 	labels := observability.Labels{"k": "v"}
 
-	b.NewCounter("prom_adapter_counter_total", "desc", "k").Add(ctx, 1, labels)
+	c, err := b.NewCounter("prom_adapter_counter_total", "desc", "k")
-	b.NewUpDownCounter("prom_adapter_updown", "desc", "k").Add(ctx, 2, labels)
+	if err != nil {
-	b.NewInt64Gauge("prom_adapter_int_gauge", "desc", "k").Record(ctx, 99, labels)
+		t.Fatalf("NewCounter error: %v", err)
-	b.NewFloat64Gauge("prom_adapter_float_gauge", "desc", "k").Record(ctx, 1.23, labels)
+	}
-	b.NewHistogram("prom_adapter_histogram", "desc", []float64{0.1, 1.0}, "k").Record(ctx, 0.5, labels)
+	u, err := b.NewUpDownCounter("prom_adapter_updown", "desc", "k")
+	if err != nil {
+		t.Fatalf("NewUpDownCounter error: %v", err)
+	}
+	ig, err := b.NewInt64Gauge("prom_adapter_int_gauge", "desc", "k")
+	if err != nil {
+		t.Fatalf("NewInt64Gauge error: %v", err)
+	}
+	fg, err := b.NewFloat64Gauge("prom_adapter_float_gauge", "desc", "k")
+	if err != nil {
+		t.Fatalf("NewFloat64Gauge error: %v", err)
+	}
+	h, err := b.NewHistogram("prom_adapter_histogram", "desc", []float64{0.1, 1.0}, "k")
+	if err != nil {
+		t.Fatalf("NewHistogram error: %v", err)
+	}
+
+	c.Add(ctx, 1, labels)
+	u.Add(ctx, 2, labels)
+	ig.Record(ctx, 99, labels)
+	fg.Record(ctx, 1.23, labels)
+	h.Record(ctx, 0.5, labels)
 
 	if b.HTTPHandler() == nil {
 		t.Error("prometheus adapter HTTPHandler should not be nil")
@@ -172,9 +205,20 @@ func TestPrometheusAdapterAllInstruments(t *testing.T) {
 }
 
 func TestOtelAdapterAllInstruments(t *testing.T) {
+	if os.Getenv("SKIP_OTEL_INTEGRATION") != "" {
+		t.Skip("skipping OTel integration test because SKIP_OTEL_INTEGRATION is set")
+	}
+
+	dialTimeout := 300 * time.Millisecond
+	conn, err := net.DialTimeout("tcp", otelGRPCEndpoint, dialTimeout)
+	if err != nil {
+		t.Skipf("skipping OTel integration test; collector %s not reachable: %v", otelGRPCEndpoint, err)
+	}
+	_ = conn.Close()
+
 	b, err := observability.New(observability.MetricsConfig{
 		Enabled: true, Backend: "otel",
-		OTel: observability.OTelConfig{Protocol: "grpc", Endpoint: otelGRPCEndpoint, Insecure: true, ExportInterval: 100 * time.Millisecond},
+		OTel: observability.OTelConfig{Protocol: "grpc", Endpoint: otelGRPCEndpoint, Insecure: true, ExportInterval: 100 * time.Millisecond, Timeout: 2 * time.Second},
 	})
 	if err != nil {
 		t.Fatalf("failed to create otel backend: %v", err)
@@ -182,11 +226,32 @@ func TestOtelAdapterAllInstruments(t *testing.T) {
 	ctx := context.Background()
 	labels := observability.Labels{"k": "v"}
 
-	b.NewCounter("otel_adapter_counter_total", "desc", "k").Add(ctx, 1, labels)
+	c, err := b.NewCounter("otel_adapter_counter_total", "desc", "k")
-	b.NewUpDownCounter("otel_adapter_updown", "desc", "k").Add(ctx, 2, labels)
+	if err != nil {
-	b.NewInt64Gauge("otel_adapter_int_gauge", "desc", "k").Record(ctx, 99, labels)
+		t.Fatalf("NewCounter error: %v", err)
-	b.NewFloat64Gauge("otel_adapter_float_gauge", "desc", "k").Record(ctx, 1.23, labels)
+	}
-	b.NewHistogram("otel_adapter_histogram", "desc", []float64{0.1, 1.0}, "k").Record(ctx, 0.5, labels)
+	u, err := b.NewUpDownCounter("otel_adapter_updown", "desc", "k")
+	if err != nil {
+		t.Fatalf("NewUpDownCounter error: %v", err)
+	}
+	ig, err := b.NewInt64Gauge("otel_adapter_int_gauge", "desc", "k")
+	if err != nil {
+		t.Fatalf("NewInt64Gauge error: %v", err)
+	}
+	fg, err := b.NewFloat64Gauge("otel_adapter_float_gauge", "desc", "k")
+	if err != nil {
+		t.Fatalf("NewFloat64Gauge error: %v", err)
+	}
+	h, err := b.NewHistogram("otel_adapter_histogram", "desc", []float64{0.1, 1.0}, "k")
+	if err != nil {
+		t.Fatalf("NewHistogram error: %v", err)
+	}
+
+	c.Add(ctx, 1, labels)
+	u.Add(ctx, 2, labels)
+	ig.Record(ctx, 99, labels)
+	fg.Record(ctx, 1.23, labels)
+	h.Record(ctx, 0.5, labels)
 
 	if b.HTTPHandler() != nil {
 		t.Error("OTel adapter HTTPHandler should be nil")

internal/observability/noop.go

@@ -13,38 +13,31 @@ type NoopBackend struct{}
 // Compile-time interface check.
 var _ Backend = (*NoopBackend)(nil)
 
-func (n *NoopBackend) NewCounter(_ string, _ string, _ ...string) Counter {
+func (n *NoopBackend) NewCounter(_ string, _ string, _ ...string) (Counter, error) {
-	_ = n
+	return noopCounter{}, nil
-	return noopCounter{}
 }
 
-func (n *NoopBackend) NewUpDownCounter(_ string, _ string, _ ...string) UpDownCounter {
+func (n *NoopBackend) NewUpDownCounter(_ string, _ string, _ ...string) (UpDownCounter, error) {
-	_ = n
+	return noopUpDownCounter{}, nil
-	return noopUpDownCounter{}
 }
 
-func (n *NoopBackend) NewInt64Gauge(_ string, _ string, _ ...string) Int64Gauge {
+func (n *NoopBackend) NewInt64Gauge(_ string, _ string, _ ...string) (Int64Gauge, error) {
-	_ = n
+	return noopInt64Gauge{}, nil
-	return noopInt64Gauge{}
 }
 
-func (n *NoopBackend) NewFloat64Gauge(_ string, _ string, _ ...string) Float64Gauge {
+func (n *NoopBackend) NewFloat64Gauge(_ string, _ string, _ ...string) (Float64Gauge, error) {
-	_ = n
+	return noopFloat64Gauge{}, nil
-	return noopFloat64Gauge{}
 }
 
-func (n *NoopBackend) NewHistogram(_ string, _ string, _ []float64, _ ...string) Histogram {
+func (n *NoopBackend) NewHistogram(_ string, _ string, _ []float64, _ ...string) (Histogram, error) {
-	_ = n
+	return noopHistogram{}, nil
-	return noopHistogram{}
 }
 
 func (n *NoopBackend) HTTPHandler() http.Handler {
-	_ = n
 	return nil
 }
 
 func (n *NoopBackend) Shutdown(_ context.Context) error {
-	_ = n
 	return nil
 }
 

internal/observability/noop_test.go

@@ -13,32 +13,32 @@ func TestNoopBackendAllInstruments(t *testing.T) {
 	ctx := context.Background()
 	labels := observability.Labels{"k": "v"}
 
-	t.Run("Counter", func(_ *testing.T) {
+	t.Run("Counter", func(t *testing.T) {
-		c := n.NewCounter("test_counter", "desc")
+		c, _ := n.NewCounter("test_counter", "desc")
 		c.Add(ctx, 1, labels)
 		c.Add(ctx, 0, nil)
 	})
 
-	t.Run("UpDownCounter", func(_ *testing.T) {
+	t.Run("UpDownCounter", func(t *testing.T) {
-		u := n.NewUpDownCounter("test_updown", "desc")
+		u, _ := n.NewUpDownCounter("test_updown", "desc")
 		u.Add(ctx, 1, labels)
 		u.Add(ctx, -1, nil)
 	})
 
-	t.Run("Int64Gauge", func(_ *testing.T) {
+	t.Run("Int64Gauge", func(t *testing.T) {
-		g := n.NewInt64Gauge("test_int64gauge", "desc")
+		g, _ := n.NewInt64Gauge("test_int64gauge", "desc")
 		g.Record(ctx, 42, labels)
 		g.Record(ctx, 0, nil)
 	})
 
-	t.Run("Float64Gauge", func(_ *testing.T) {
+	t.Run("Float64Gauge", func(t *testing.T) {
-		g := n.NewFloat64Gauge("test_float64gauge", "desc")
+		g, _ := n.NewFloat64Gauge("test_float64gauge", "desc")
 		g.Record(ctx, 3.14, labels)
 		g.Record(ctx, 0, nil)
 	})
 
-	t.Run("Histogram", func(_ *testing.T) {
+	t.Run("Histogram", func(t *testing.T) {
-		h := n.NewHistogram("test_histogram", "desc", []float64{1, 5, 10})
+		h, _ := n.NewHistogram("test_histogram", "desc", []float64{1, 5, 10})
 		h.Record(ctx, 2.5, labels)
 		h.Record(ctx, 0, nil)
 	})
@@ -56,12 +56,47 @@ func TestNoopBackendAllInstruments(t *testing.T) {
 	})
 }
 
-func TestNoopBackendLabelNames(_ *testing.T) {
+func TestNoopBackendLabelNames(t *testing.T) {
 	// Verify that label names passed at creation time are accepted without panic.
 	n := &observability.NoopBackend{}
-	n.NewCounter("c", "d", "label1", "label2")
+
-	n.NewUpDownCounter("u", "d", "l1")
+	assertNoPanic := func(t *testing.T, constructor string, fn func()) {
-	n.NewInt64Gauge("g1", "d", "l1", "l2", "l3")
+		t.Helper()
-	n.NewFloat64Gauge("g2", "d")
+		defer func() {
-	n.NewHistogram("h", "d", []float64{0.1, 1.0}, "l1")
+			if r := recover(); r != nil {
+				t.Fatalf("%s panicked: %v", constructor, r)
+			}
+		}()
+		fn()
+	}
+
+	t.Run("NewCounter", func(t *testing.T) {
+		assertNoPanic(t, "NewCounter", func() {
+			_, _ = n.NewCounter("c", "d", "label1", "label2")
+		})
+	})
+
+	t.Run("NewUpDownCounter", func(t *testing.T) {
+		assertNoPanic(t, "NewUpDownCounter", func() {
+			_, _ = n.NewUpDownCounter("u", "d", "l1")
+		})
+	})
+
+	t.Run("NewInt64Gauge", func(t *testing.T) {
+		assertNoPanic(t, "NewInt64Gauge", func() {
+			_, _ = n.NewInt64Gauge("g1", "d", "l1", "l2", "l3")
+		})
+	})
+
+	t.Run("NewFloat64Gauge", func(t *testing.T) {
+		assertNoPanic(t, "NewFloat64Gauge", func() {
+			_, _ = n.NewFloat64Gauge("g2", "d")
+		})
+	})
+
+	t.Run("NewHistogram", func(t *testing.T) {
+		assertNoPanic(t, "NewHistogram", func() {
+			_, _ = n.NewHistogram("h", "d", []float64{0.1, 1.0}, "l1")
+		})
+	})
 }

internal/observability/otel/backend.go

@@ -9,7 +9,10 @@ package otel
 import (
 	"context"
 	"fmt"
+	"log"
 	"net/http"
+	"regexp"
+	"strings"
 	"time"
 
 	"go.opentelemetry.io/otel/attribute"
@@ -17,6 +20,8 @@ import (
 	sdkmetric "go.opentelemetry.io/otel/sdk/metric"
 )
 
+var metricLabelNameRE = regexp.MustCompile(`^[a-zA-Z_][a-zA-Z0-9_]*$`)
+
 // Config holds OTel backend configuration.
 type Config struct {
 	// Protocol is "grpc" (default) or "http".
@@ -31,6 +36,9 @@ type Config struct {
 	// ExportInterval is the period between pushes to the collector.
 	ExportInterval time.Duration
 
+	// Timeout bounds exporter construction calls.
+	Timeout time.Duration
+
 	ServiceName           string
 	ServiceVersion        string
 	DeploymentEnvironment string
@@ -57,9 +65,15 @@ func New(cfg Config) (*Backend, error) {
 	if cfg.Protocol == "" {
 		cfg.Protocol = "grpc"
 	}
+	if strings.TrimSpace(cfg.Endpoint) == "" {
+		return nil, fmt.Errorf("otel backend: empty cfg.Endpoint")
+	}
 	if cfg.ExportInterval <= 0 {
 		cfg.ExportInterval = 60 * time.Second
 	}
+	if cfg.Timeout <= 0 {
+		cfg.Timeout = 10 * time.Second
+	}
 	if cfg.ServiceName == "" {
 		cfg.ServiceName = "gerbil"
 	}
@@ -100,111 +114,196 @@ func (b *Backend) Shutdown(ctx context.Context) error {
 }
 
 // NewCounter creates an OTel Int64Counter.
-func (b *Backend) NewCounter(name, desc string, _ ...string) *Counter {
+func (b *Backend) NewCounter(name, desc string, labelNames ...string) (*Counter, error) {
+	normalizedLabelNames, err := validateLabelNames(labelNames)
+	if err != nil {
+		return nil, fmt.Errorf("otel: create counter %q: %w", name, err)
+	}
 	c, err := b.meter.Int64Counter(name, metric.WithDescription(desc))
 	if err != nil {
-		panic(fmt.Sprintf("otel: create counter %q: %v", name, err))
+		return nil, fmt.Errorf("otel: create counter %q: %w", name, err)
 	}
-	return &Counter{c: c}
+	return &Counter{c: c, labelNames: normalizedLabelNames}, nil
 }
 
 // NewUpDownCounter creates an OTel Int64UpDownCounter.
-func (b *Backend) NewUpDownCounter(name, desc string, _ ...string) *UpDownCounter {
+func (b *Backend) NewUpDownCounter(name, desc string, labelNames ...string) (*UpDownCounter, error) {
+	normalizedLabelNames, err := validateLabelNames(labelNames)
+	if err != nil {
+		return nil, fmt.Errorf("otel: create up-down counter %q: %w", name, err)
+	}
 	c, err := b.meter.Int64UpDownCounter(name, metric.WithDescription(desc))
 	if err != nil {
-		panic(fmt.Sprintf("otel: create up-down counter %q: %v", name, err))
+		return nil, fmt.Errorf("otel: create up-down counter %q: %w", name, err)
 	}
-	return &UpDownCounter{c: c}
+	return &UpDownCounter{c: c, labelNames: normalizedLabelNames}, nil
 }
 
 // NewInt64Gauge creates an OTel Int64Gauge.
-func (b *Backend) NewInt64Gauge(name, desc string, _ ...string) *Int64Gauge {
+func (b *Backend) NewInt64Gauge(name, desc string, labelNames ...string) (*Int64Gauge, error) {
+	normalizedLabelNames, err := validateLabelNames(labelNames)
+	if err != nil {
+		return nil, fmt.Errorf("otel: create int64 gauge %q: %w", name, err)
+	}
 	g, err := b.meter.Int64Gauge(name, metric.WithDescription(desc))
 	if err != nil {
-		panic(fmt.Sprintf("otel: create int64 gauge %q: %v", name, err))
+		return nil, fmt.Errorf("otel: create int64 gauge %q: %w", name, err)
 	}
-	return &Int64Gauge{g: g}
+	return &Int64Gauge{g: g, labelNames: normalizedLabelNames}, nil
 }
 
 // NewFloat64Gauge creates an OTel Float64Gauge.
-func (b *Backend) NewFloat64Gauge(name, desc string, _ ...string) *Float64Gauge {
+func (b *Backend) NewFloat64Gauge(name, desc string, labelNames ...string) (*Float64Gauge, error) {
+	normalizedLabelNames, err := validateLabelNames(labelNames)
+	if err != nil {
+		return nil, fmt.Errorf("otel: create float64 gauge %q: %w", name, err)
+	}
 	g, err := b.meter.Float64Gauge(name, metric.WithDescription(desc))
 	if err != nil {
-		panic(fmt.Sprintf("otel: create float64 gauge %q: %v", name, err))
+		return nil, fmt.Errorf("otel: create float64 gauge %q: %w", name, err)
 	}
-	return &Float64Gauge{g: g}
+	return &Float64Gauge{g: g, labelNames: normalizedLabelNames}, nil
 }
 
 // NewHistogram creates an OTel Float64Histogram with explicit bucket boundaries.
-func (b *Backend) NewHistogram(name, desc string, buckets []float64, _ ...string) *Histogram {
+func (b *Backend) NewHistogram(name, desc string, buckets []float64, labelNames ...string) (*Histogram, error) {
+	normalizedLabelNames, err := validateLabelNames(labelNames)
+	if err != nil {
+		return nil, fmt.Errorf("otel: create histogram %q: %w", name, err)
+	}
 	h, err := b.meter.Float64Histogram(name,
 		metric.WithDescription(desc),
 		metric.WithExplicitBucketBoundaries(buckets...),
 	)
 	if err != nil {
-		panic(fmt.Sprintf("otel: create histogram %q: %v", name, err))
+		return nil, fmt.Errorf("otel: create histogram %q: %w", name, err)
 	}
-	return &Histogram{h: h}
+	return &Histogram{h: h, labelNames: normalizedLabelNames}, nil
 }
 
-// labelsToAttrs converts a Labels map to OTel attribute key-value pairs.
+func validateLabelNames(labelNames []string) ([]string, error) {
-func labelsToAttrs(labels map[string]string) []attribute.KeyValue {
+	if len(labelNames) == 0 {
-	if len(labels) == 0 {
+		return nil, nil
-		return nil
 	}
-	attrs := make([]attribute.KeyValue, 0, len(labels))
+
-	for k, v := range labels {
+	normalized := make([]string, len(labelNames))
-		attrs = append(attrs, attribute.String(k, v))
+	seen := make(map[string]struct{}, len(labelNames))
+	for i, name := range labelNames {
+		if !metricLabelNameRE.MatchString(name) {
+			return nil, fmt.Errorf("invalid label name %q", name)
+		}
+		if _, exists := seen[name]; exists {
+			return nil, fmt.Errorf("duplicate label name %q", name)
+		}
+		seen[name] = struct{}{}
+		normalized[i] = name
 	}
+
+	return normalized, nil
+}
+
+func labelsToAttrs(labelNames []string, labels map[string]string) []attribute.KeyValue {
+	if len(labelNames) == 0 {
+		if len(labels) > 0 {
+			log.Printf("WARN: dropping otel metric sample due to unexpected labels: got=%v expected=none", labels)
+			return nil
+		}
+		return []attribute.KeyValue{}
+	}
+
+	attrs := make([]attribute.KeyValue, 0, len(labelNames))
+	for _, labelName := range labelNames {
+		attrs = append(attrs, attribute.String(labelName, labels[labelName]))
+	}
+
+	for got := range labels {
+		found := false
+		for _, expected := range labelNames {
+			if got == expected {
+				found = true
+				break
+			}
+		}
+		if !found {
+			log.Printf("WARN: dropping otel metric sample due to unexpected label key %q (expected=%v)", got, labelNames)
+			return nil
+		}
+	}
+
 	return attrs
 }
 
 // Counter wraps an OTel Int64Counter.
 type Counter struct {
-	c metric.Int64Counter
+	c          metric.Int64Counter
+	labelNames []string
 }
 
 // Add increments the counter by value.
 func (c *Counter) Add(ctx context.Context, value int64, labels map[string]string) {
-	c.c.Add(ctx, value, metric.WithAttributes(labelsToAttrs(labels)...))
+	attrs := labelsToAttrs(c.labelNames, labels)
+	if attrs == nil {
+		return
+	}
+	c.c.Add(ctx, value, metric.WithAttributes(attrs...))
 }
 
 // UpDownCounter wraps an OTel Int64UpDownCounter.
 type UpDownCounter struct {
-	c metric.Int64UpDownCounter
+	c          metric.Int64UpDownCounter
+	labelNames []string
 }
 
 // Add adjusts the up-down counter by value.
 func (u *UpDownCounter) Add(ctx context.Context, value int64, labels map[string]string) {
-	u.c.Add(ctx, value, metric.WithAttributes(labelsToAttrs(labels)...))
+	attrs := labelsToAttrs(u.labelNames, labels)
+	if attrs == nil {
+		return
+	}
+	u.c.Add(ctx, value, metric.WithAttributes(attrs...))
 }
 
 // Int64Gauge wraps an OTel Int64Gauge.
 type Int64Gauge struct {
-	g metric.Int64Gauge
+	g          metric.Int64Gauge
+	labelNames []string
 }
 
 // Record sets the gauge to value.
 func (g *Int64Gauge) Record(ctx context.Context, value int64, labels map[string]string) {
-	g.g.Record(ctx, value, metric.WithAttributes(labelsToAttrs(labels)...))
+	attrs := labelsToAttrs(g.labelNames, labels)
+	if attrs == nil {
+		return
+	}
+	g.g.Record(ctx, value, metric.WithAttributes(attrs...))
 }
 
 // Float64Gauge wraps an OTel Float64Gauge.
 type Float64Gauge struct {
-	g metric.Float64Gauge
+	g          metric.Float64Gauge
+	labelNames []string
 }
 
 // Record sets the gauge to value.
 func (g *Float64Gauge) Record(ctx context.Context, value float64, labels map[string]string) {
-	g.g.Record(ctx, value, metric.WithAttributes(labelsToAttrs(labels)...))
+	attrs := labelsToAttrs(g.labelNames, labels)
+	if attrs == nil {
+		return
+	}
+	g.g.Record(ctx, value, metric.WithAttributes(attrs...))
 }
 
 // Histogram wraps an OTel Float64Histogram.
 type Histogram struct {
-	h metric.Float64Histogram
+	h          metric.Float64Histogram
+	labelNames []string
 }
 
 // Record observes value in the histogram.
 func (h *Histogram) Record(ctx context.Context, value float64, labels map[string]string) {
-	h.h.Record(ctx, value, metric.WithAttributes(labelsToAttrs(labels)...))
+	attrs := labelsToAttrs(h.labelNames, labels)
+	if attrs == nil {
+		return
+	}
+	h.h.Record(ctx, value, metric.WithAttributes(attrs...))
 }

internal/observability/otel/backend_test.go

@@ -55,7 +55,10 @@ func TestOtelBackendCounter(t *testing.T) {
 	b := newInMemoryBackend(t)
 	defer b.Shutdown(context.Background()) //nolint:errcheck
 
-	c := b.NewCounter("gerbil_test_counter_total", "test counter", "result")
+	c, err := b.NewCounter("gerbil_test_counter_total", "test counter", "result")
+	if err != nil {
+		t.Fatalf("NewCounter returned error: %v", err)
+	}
 	// Should not panic
 	c.Add(context.Background(), 1, map[string]string{"result": "ok"})
 	c.Add(context.Background(), 5, nil)
@@ -65,7 +68,10 @@ func TestOtelBackendUpDownCounter(t *testing.T) {
 	b := newInMemoryBackend(t)
 	defer b.Shutdown(context.Background()) //nolint:errcheck
 
-	u := b.NewUpDownCounter("gerbil_test_updown", "test updown", "state")
+	u, err := b.NewUpDownCounter("gerbil_test_updown", "test updown", "state")
+	if err != nil {
+		t.Fatalf("NewUpDownCounter returned error: %v", err)
+	}
 	u.Add(context.Background(), 3, map[string]string{"state": "active"})
 	u.Add(context.Background(), -1, map[string]string{"state": "active"})
 }
@@ -74,7 +80,10 @@ func TestOtelBackendInt64Gauge(t *testing.T) {
 	b := newInMemoryBackend(t)
 	defer b.Shutdown(context.Background()) //nolint:errcheck
 
-	g := b.NewInt64Gauge("gerbil_test_int_gauge", "test gauge")
+	g, err := b.NewInt64Gauge("gerbil_test_int_gauge", "test gauge")
+	if err != nil {
+		t.Fatalf("NewInt64Gauge returned error: %v", err)
+	}
 	g.Record(context.Background(), 42, nil)
 }
 
@@ -82,7 +91,10 @@ func TestOtelBackendFloat64Gauge(t *testing.T) {
 	b := newInMemoryBackend(t)
 	defer b.Shutdown(context.Background()) //nolint:errcheck
 
-	g := b.NewFloat64Gauge("gerbil_test_float_gauge", "test float gauge")
+	g, err := b.NewFloat64Gauge("gerbil_test_float_gauge", "test float gauge")
+	if err != nil {
+		t.Fatalf("NewFloat64Gauge returned error: %v", err)
+	}
 	g.Record(context.Background(), 3.14, nil)
 }
 
@@ -90,8 +102,11 @@ func TestOtelBackendHistogram(t *testing.T) {
 	b := newInMemoryBackend(t)
 	defer b.Shutdown(context.Background()) //nolint:errcheck
 
-	h := b.NewHistogram("gerbil_test_duration_seconds", "test histogram",
+	h, err := b.NewHistogram("gerbil_test_duration_seconds", "test histogram",
 		[]float64{0.1, 0.5, 1.0}, "method")
+	if err != nil {
+		t.Fatalf("NewHistogram returned error: %v", err)
+	}
 	h.Record(context.Background(), 0.3, map[string]string{"method": "GET"})
 }
 
@@ -139,3 +154,22 @@ func TestOtelBackendDeploymentEnvironment(t *testing.T) {
 	}
 	defer b.Shutdown(context.Background()) //nolint:errcheck
 }
+
+func TestOtelBackendRejectsInvalidLabelNames(t *testing.T) {
+	b := newInMemoryBackend(t)
+	defer b.Shutdown(context.Background()) //nolint:errcheck
+
+	t.Run("duplicate labels", func(t *testing.T) {
+		_, err := b.NewCounter("gerbil_test_invalid_labels_total", "test counter", "result", "result")
+		if err == nil {
+			t.Fatal("expected error for duplicate label names")
+		}
+	})
+
+	t.Run("invalid label name", func(t *testing.T) {
+		_, err := b.NewHistogram("gerbil_test_invalid_histogram", "test histogram", []float64{0.1, 1.0}, "status-code")
+		if err == nil {
+			t.Fatal("expected error for invalid label name")
+		}
+	})
+}

internal/observability/otel/exporter.go

@@ -3,6 +3,8 @@ package otel
 import (
 	"context"
 	"fmt"
+	"net/url"
+	"strings"
 
 	"go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetricgrpc"
 	"go.opentelemetry.io/otel/exporters/otlp/otlpmetric/otlpmetrichttp"
@@ -11,6 +13,10 @@ import (
 
 // newExporter creates the appropriate OTLP exporter based on cfg.Protocol.
 func newExporter(ctx context.Context, cfg Config) (sdkmetric.Exporter, error) {
+	if strings.TrimSpace(cfg.Endpoint) == "" {
+		return nil, fmt.Errorf("otel: cfg.Endpoint is empty")
+	}
+
 	switch cfg.Protocol {
 	case "grpc", "":
 		return newGRPCExporter(ctx, cfg)
@@ -36,8 +42,20 @@ func newGRPCExporter(ctx context.Context, cfg Config) (sdkmetric.Exporter, error
 }
 
 func newHTTPExporter(ctx context.Context, cfg Config) (sdkmetric.Exporter, error) {
-	opts := []otlpmetrichttp.Option{
+	endpoint := strings.TrimSpace(cfg.Endpoint)
-		otlpmetrichttp.WithEndpoint(cfg.Endpoint),
+
+	opts := make([]otlpmetrichttp.Option, 0, 3)
+	if strings.Contains(endpoint, "://") {
+		parsed, err := url.Parse(endpoint)
+		if err != nil {
+			return nil, fmt.Errorf("otlp http exporter: parse endpoint URL %q: %w", endpoint, err)
+		}
+		opts = append(opts, otlpmetrichttp.WithEndpointURL(parsed.String()))
+	} else {
+		opts = append(opts,
+			otlpmetrichttp.WithEndpoint(endpoint),
+			otlpmetrichttp.WithURLPath("/v1/metrics"),
+		)
 	}
 	if cfg.Insecure {
 		opts = append(opts, otlpmetrichttp.WithInsecure())

internal/observability/otel/resource.go

@@ -3,7 +3,7 @@ package otel
 import (
 	"go.opentelemetry.io/otel/attribute"
 	"go.opentelemetry.io/otel/sdk/resource"
-	semconv "go.opentelemetry.io/otel/semconv/v1.40.0"
+	semconv "go.opentelemetry.io/otel/semconv/v1.18.0"
 )
 
 // newResource builds an OTel resource for the Gerbil service.
@@ -15,11 +15,11 @@ func newResource(serviceName, serviceVersion, deploymentEnv string) (*resource.R
 		attrs = append(attrs, semconv.ServiceVersion(serviceVersion))
 	}
 	if deploymentEnv != "" {
-		attrs = append(attrs, semconv.DeploymentEnvironmentName(deploymentEnv))
+		attrs = append(attrs, semconv.DeploymentEnvironment(deploymentEnv))
 	}
 
 	return resource.Merge(
 		resource.Default(),
-		resource.NewWithAttributes(semconv.SchemaURL, attrs...),
+		resource.NewSchemaless(attrs...),
 	)
 }

internal/observability/prometheus/backend.go

@@ -7,6 +7,7 @@ package prometheus
 
 import (
 	"context"
+	"log"
 	"net/http"
 
 	"github.com/prometheus/client_golang/prometheus"
@@ -30,9 +31,10 @@ type Config struct {
 // in the backend-specific instrument types that implement the observability
 // instrument interfaces.
 type Backend struct {
-	cfg      Config
+	cfg                   Config
-	registry *prometheus.Registry
+	registry              *prometheus.Registry
-	handler  http.Handler
+	handler               http.Handler
+	droppedSamplesCounter prometheus.Counter
 }
 
 // New creates and initialises a Prometheus backend.
@@ -48,6 +50,11 @@ func New(cfg Config) (*Backend, error) {
 	}
 
 	registry := prometheus.NewRegistry()
+	droppedSamplesCounter := prometheus.NewCounter(prometheus.CounterOpts{
+		Name: "gerbil_dropped_metric_samples_total",
+		Help: "Total number of metric samples dropped due to invalid labels or unsupported label sets",
+	})
+	registry.MustRegister(droppedSamplesCounter)
 
 	// Include Go and process metrics by default.
 	includeGo := cfg.IncludeGoMetrics == nil || *cfg.IncludeGoMetrics
@@ -62,7 +69,7 @@ func New(cfg Config) (*Backend, error) {
 		EnableOpenMetrics: false,
 	})
 
-	return &Backend{cfg: cfg, registry: registry, handler: handler}, nil
+	return &Backend{cfg: cfg, registry: registry, handler: handler, droppedSamplesCounter: droppedSamplesCounter}, nil
 }
 
 // HTTPHandler returns the Prometheus /metrics HTTP handler.
@@ -78,60 +85,107 @@ func (b *Backend) Shutdown(_ context.Context) error {
 }
 
 // NewCounter creates a Prometheus CounterVec registered on the backend's registry.
-func (b *Backend) NewCounter(name, desc string, labelNames ...string) *Counter {
+func (b *Backend) NewCounter(name, desc string, labelNames ...string) (*Counter, error) {
 	vec := prometheus.NewCounterVec(prometheus.CounterOpts{
 		Name: name,
 		Help: desc,
 	}, labelNames)
-	b.registry.MustRegister(vec)
+	if err := b.registry.Register(vec); err != nil {
-	return &Counter{vec: vec}
+		if are, ok := err.(prometheus.AlreadyRegisteredError); ok {
+			existing, ok := are.ExistingCollector.(*prometheus.CounterVec)
+			if !ok {
+				return nil, err
+			}
+			return &Counter{vec: existing, labelNames: append([]string(nil), labelNames...), droppedSamplesCounter: b.droppedSamplesCounter}, nil
+		}
+		return nil, err
+	}
+	return &Counter{vec: vec, labelNames: append([]string(nil), labelNames...), droppedSamplesCounter: b.droppedSamplesCounter}, nil
 }
 
 // NewUpDownCounter creates a Prometheus GaugeVec (Prometheus gauges are
 // bidirectional) registered on the backend's registry.
-func (b *Backend) NewUpDownCounter(name, desc string, labelNames ...string) *UpDownCounter {
+func (b *Backend) NewUpDownCounter(name, desc string, labelNames ...string) (*UpDownCounter, error) {
 	vec := prometheus.NewGaugeVec(prometheus.GaugeOpts{
 		Name: name,
 		Help: desc,
 	}, labelNames)
-	b.registry.MustRegister(vec)
+	if err := b.registry.Register(vec); err != nil {
-	return &UpDownCounter{vec: vec}
+		if are, ok := err.(prometheus.AlreadyRegisteredError); ok {
+			existing, ok := are.ExistingCollector.(*prometheus.GaugeVec)
+			if !ok {
+				return nil, err
+			}
+			return &UpDownCounter{vec: existing, labelNames: append([]string(nil), labelNames...), droppedSamplesCounter: b.droppedSamplesCounter}, nil
+		}
+		return nil, err
+	}
+	return &UpDownCounter{vec: vec, labelNames: append([]string(nil), labelNames...), droppedSamplesCounter: b.droppedSamplesCounter}, nil
 }
 
 // NewInt64Gauge creates a Prometheus GaugeVec registered on the backend's registry.
-func (b *Backend) NewInt64Gauge(name, desc string, labelNames ...string) *Int64Gauge {
+func (b *Backend) NewInt64Gauge(name, desc string, labelNames ...string) (*Int64Gauge, error) {
 	vec := prometheus.NewGaugeVec(prometheus.GaugeOpts{
 		Name: name,
 		Help: desc,
 	}, labelNames)
-	b.registry.MustRegister(vec)
+	if err := b.registry.Register(vec); err != nil {
-	return &Int64Gauge{vec: vec}
+		if are, ok := err.(prometheus.AlreadyRegisteredError); ok {
+			existing, ok := are.ExistingCollector.(*prometheus.GaugeVec)
+			if !ok {
+				return nil, err
+			}
+			return &Int64Gauge{vec: existing, labelNames: append([]string(nil), labelNames...), droppedSamplesCounter: b.droppedSamplesCounter}, nil
+		}
+		return nil, err
+	}
+	return &Int64Gauge{vec: vec, labelNames: append([]string(nil), labelNames...), droppedSamplesCounter: b.droppedSamplesCounter}, nil
 }
 
 // NewFloat64Gauge creates a Prometheus GaugeVec registered on the backend's registry.
-func (b *Backend) NewFloat64Gauge(name, desc string, labelNames ...string) *Float64Gauge {
+func (b *Backend) NewFloat64Gauge(name, desc string, labelNames ...string) (*Float64Gauge, error) {
 	vec := prometheus.NewGaugeVec(prometheus.GaugeOpts{
 		Name: name,
 		Help: desc,
 	}, labelNames)
-	b.registry.MustRegister(vec)
+	if err := b.registry.Register(vec); err != nil {
-	return &Float64Gauge{vec: vec}
+		if are, ok := err.(prometheus.AlreadyRegisteredError); ok {
+			existing, ok := are.ExistingCollector.(*prometheus.GaugeVec)
+			if !ok {
+				return nil, err
+			}
+			return &Float64Gauge{vec: existing, labelNames: append([]string(nil), labelNames...), droppedSamplesCounter: b.droppedSamplesCounter}, nil
+		}
+		return nil, err
+	}
+	return &Float64Gauge{vec: vec, labelNames: append([]string(nil), labelNames...), droppedSamplesCounter: b.droppedSamplesCounter}, nil
 }
 
 // NewHistogram creates a Prometheus HistogramVec registered on the backend's registry.
-func (b *Backend) NewHistogram(name, desc string, buckets []float64, labelNames ...string) *Histogram {
+func (b *Backend) NewHistogram(name, desc string, buckets []float64, labelNames ...string) (*Histogram, error) {
 	vec := prometheus.NewHistogramVec(prometheus.HistogramOpts{
 		Name:    name,
 		Help:    desc,
 		Buckets: buckets,
 	}, labelNames)
-	b.registry.MustRegister(vec)
+	if err := b.registry.Register(vec); err != nil {
-	return &Histogram{vec: vec}
+		if are, ok := err.(prometheus.AlreadyRegisteredError); ok {
+			existing, ok := are.ExistingCollector.(*prometheus.HistogramVec)
+			if !ok {
+				return nil, err
+			}
+			return &Histogram{vec: existing, labelNames: append([]string(nil), labelNames...), droppedSamplesCounter: b.droppedSamplesCounter}, nil
+		}
+		return nil, err
+	}
+	return &Histogram{vec: vec, labelNames: append([]string(nil), labelNames...), droppedSamplesCounter: b.droppedSamplesCounter}, nil
 }
 
 // Counter is a native Prometheus counter instrument.
 type Counter struct {
-	vec *prometheus.CounterVec
+	vec                   *prometheus.CounterVec
+	labelNames            []string
+	droppedSamplesCounter prometheus.Counter
 }
 
 // Add increments the counter by value for the given labels.
@@ -139,47 +193,118 @@ type Counter struct {
 // value must be non-negative. Negative values are ignored.
 func (c *Counter) Add(_ context.Context, value int64, labels map[string]string) {
 	if value < 0 {
+		log.Printf("WARN: counter add called with negative value=%d labels=%v expected_labels=%v", value, labels, c.labelNames)
 		return
 	}
-	c.vec.With(prometheus.Labels(labels)).Add(float64(value))
+	normalized, ok := normalizeLabels(c.labelNames, labels, c.droppedSamplesCounter)
+	if !ok {
+		return
+	}
+	defer guardMetricPanic("counter", c.labelNames, labels)
+	c.vec.With(normalized).Add(float64(value))
 }
 
 // UpDownCounter is a native Prometheus gauge used as a bidirectional counter.
 type UpDownCounter struct {
-	vec *prometheus.GaugeVec
+	vec                   *prometheus.GaugeVec
+	labelNames            []string
+	droppedSamplesCounter prometheus.Counter
 }
 
 // Add adjusts the gauge by value for the given labels.
 func (u *UpDownCounter) Add(_ context.Context, value int64, labels map[string]string) {
-	u.vec.With(prometheus.Labels(labels)).Add(float64(value))
+	normalized, ok := normalizeLabels(u.labelNames, labels, u.droppedSamplesCounter)
+	if !ok {
+		return
+	}
+	defer guardMetricPanic("updown", u.labelNames, labels)
+	u.vec.With(normalized).Add(float64(value))
 }
 
 // Int64Gauge is a native Prometheus gauge recording integer snapshot values.
 type Int64Gauge struct {
-	vec *prometheus.GaugeVec
+	vec                   *prometheus.GaugeVec
+	labelNames            []string
+	droppedSamplesCounter prometheus.Counter
 }
 
 // Record sets the gauge to value for the given labels.
 func (g *Int64Gauge) Record(_ context.Context, value int64, labels map[string]string) {
-	g.vec.With(prometheus.Labels(labels)).Set(float64(value))
+	normalized, ok := normalizeLabels(g.labelNames, labels, g.droppedSamplesCounter)
+	if !ok {
+		return
+	}
+	defer guardMetricPanic("int64-gauge", g.labelNames, labels)
+	g.vec.With(normalized).Set(float64(value))
 }
 
 // Float64Gauge is a native Prometheus gauge recording float snapshot values.
 type Float64Gauge struct {
-	vec *prometheus.GaugeVec
+	vec                   *prometheus.GaugeVec
+	labelNames            []string
+	droppedSamplesCounter prometheus.Counter
 }
 
 // Record sets the gauge to value for the given labels.
 func (g *Float64Gauge) Record(_ context.Context, value float64, labels map[string]string) {
-	g.vec.With(prometheus.Labels(labels)).Set(value)
+	normalized, ok := normalizeLabels(g.labelNames, labels, g.droppedSamplesCounter)
+	if !ok {
+		return
+	}
+	defer guardMetricPanic("float64-gauge", g.labelNames, labels)
+	g.vec.With(normalized).Set(value)
 }
 
 // Histogram is a native Prometheus histogram instrument.
 type Histogram struct {
-	vec *prometheus.HistogramVec
+	vec                   *prometheus.HistogramVec
+	labelNames            []string
+	droppedSamplesCounter prometheus.Counter
 }
 
 // Record observes value for the given labels.
 func (h *Histogram) Record(_ context.Context, value float64, labels map[string]string) {
-	h.vec.With(prometheus.Labels(labels)).Observe(value)
+	normalized, ok := normalizeLabels(h.labelNames, labels, h.droppedSamplesCounter)
+	if !ok {
+		return
+	}
+	defer guardMetricPanic("histogram", h.labelNames, labels)
+	h.vec.With(normalized).Observe(value)
+}
+
+func normalizeLabels(labelNames []string, labels map[string]string, droppedSamplesCounter prometheus.Counter) (prometheus.Labels, bool) {
+	if len(labelNames) == 0 {
+		if len(labels) > 0 {
+			if droppedSamplesCounter != nil {
+				droppedSamplesCounter.Inc()
+			}
+			log.Printf("WARN: dropping metric sample due to unexpected labels: got=%v expected=none", labels)
+			return nil, false
+		}
+		return nil, true
+	}
+
+	normalized := make(prometheus.Labels, len(labelNames))
+	for _, name := range labelNames {
+		normalized[name] = ""
+	}
+
+	for k, v := range labels {
+		if _, ok := normalized[k]; !ok {
+			if droppedSamplesCounter != nil {
+				droppedSamplesCounter.Inc()
+			}
+			log.Printf("WARN: dropping metric sample due to unexpected label key %q (expected=%v)", k, labelNames)
+			return nil, false
+		}
+		normalized[k] = v
+	}
+
+	return normalized, true
+}
+
+func guardMetricPanic(kind string, expected []string, labels map[string]string) {
+	if recovered := recover(); recovered != nil {
+		log.Printf("WARN: dropped %s metric sample due to label panic: expected=%v got=%v err=%v", kind, expected, labels, recovered)
+	}
 }

internal/observability/prometheus/backend_test.go

@@ -36,7 +36,10 @@ func TestPrometheusBackendShutdown(t *testing.T) {
 
 func TestPrometheusBackendCounter(t *testing.T) {
 	b := newTestBackend(t)
-	c := b.NewCounter("test_counter_total", "A test counter", "result")
+	c, err := b.NewCounter("test_counter_total", "A test counter", "result")
+	if err != nil {
+		t.Fatalf("NewCounter returned error: %v", err)
+	}
 	c.Add(context.Background(), 3, map[string]string{"result": "ok"})
 
 	body := scrapeMetrics(t, b)
@@ -45,7 +48,10 @@ func TestPrometheusBackendCounter(t *testing.T) {
 
 func TestPrometheusBackendUpDownCounter(t *testing.T) {
 	b := newTestBackend(t)
-	u := b.NewUpDownCounter("test_gauge_total", "A test up-down counter", "state")
+	u, err := b.NewUpDownCounter("test_gauge_total", "A test up-down counter", "state")
+	if err != nil {
+		t.Fatalf("NewUpDownCounter returned error: %v", err)
+	}
 	u.Add(context.Background(), 5, map[string]string{"state": "active"})
 	u.Add(context.Background(), -2, map[string]string{"state": "active"})
 
@@ -55,7 +61,10 @@ func TestPrometheusBackendUpDownCounter(t *testing.T) {
 
 func TestPrometheusBackendInt64Gauge(t *testing.T) {
 	b := newTestBackend(t)
-	g := b.NewInt64Gauge("test_int_gauge", "An integer gauge", "ifname")
+	g, err := b.NewInt64Gauge("test_int_gauge", "An integer gauge", "ifname")
+	if err != nil {
+		t.Fatalf("NewInt64Gauge returned error: %v", err)
+	}
 	g.Record(context.Background(), 42, map[string]string{"ifname": "wg0"})
 
 	body := scrapeMetrics(t, b)
@@ -64,7 +73,10 @@ func TestPrometheusBackendInt64Gauge(t *testing.T) {
 
 func TestPrometheusBackendFloat64Gauge(t *testing.T) {
 	b := newTestBackend(t)
-	g := b.NewFloat64Gauge("test_float_gauge", "A float gauge", "cert")
+	g, err := b.NewFloat64Gauge("test_float_gauge", "A float gauge", "cert")
+	if err != nil {
+		t.Fatalf("NewFloat64Gauge returned error: %v", err)
+	}
 	g.Record(context.Background(), 7.5, map[string]string{"cert": "example.com"})
 
 	body := scrapeMetrics(t, b)
@@ -74,7 +86,10 @@ func TestPrometheusBackendFloat64Gauge(t *testing.T) {
 func TestPrometheusBackendHistogram(t *testing.T) {
 	b := newTestBackend(t)
 	buckets := []float64{0.1, 0.5, 1.0, 5.0}
-	h := b.NewHistogram("test_duration_seconds", "A test histogram", buckets, "method")
+	h, err := b.NewHistogram("test_duration_seconds", "A test histogram", buckets, "method")
+	if err != nil {
+		t.Fatalf("NewHistogram returned error: %v", err)
+	}
 	h.Record(context.Background(), 0.3, map[string]string{"method": "GET"})
 
 	body := scrapeMetrics(t, b)
@@ -85,7 +100,10 @@ func TestPrometheusBackendHistogram(t *testing.T) {
 
 func TestPrometheusBackendMultipleLabels(t *testing.T) {
 	b := newTestBackend(t)
-	c := b.NewCounter("multi_label_total", "Multi-label counter", "method", "route", "status_code")
+	c, err := b.NewCounter("multi_label_total", "Multi-label counter", "method", "route", "status_code")
+	if err != nil {
+		t.Fatalf("NewCounter returned error: %v", err)
+	}
 	c.Add(context.Background(), 1, map[string]string{
 		"method":      "POST",
 		"route":       "/api/peers",
@@ -122,23 +140,29 @@ func TestPrometheusBackendNoGoMetrics(t *testing.T) {
 func TestPrometheusBackendNilLabels(t *testing.T) {
 	// Adding with nil labels should not panic (treated as empty map).
 	b := newTestBackend(t)
-	c := b.NewCounter("nil_labels_total", "counter with no labels")
+	c, err := b.NewCounter("nil_labels_total", "counter with no labels")
+	if err != nil {
+		t.Fatalf("NewCounter returned error: %v", err)
+	}
 	// nil labels with no label names declared should be safe
 	c.Add(context.Background(), 1, nil)
 }
 
 func TestPrometheusBackendConcurrentAdd(t *testing.T) {
 	b := newTestBackend(t)
-	c := b.NewCounter("concurrent_total", "concurrent counter", "worker")
+	c, err := b.NewCounter("concurrent_total", "concurrent counter", "worker")
+	if err != nil {
+		t.Fatalf("NewCounter returned error: %v", err)
+	}
 
 	done := make(chan struct{})
 	for i := 0; i < 10; i++ {
-		go func(_ int) {
+		go func() {
 			for j := 0; j < 100; j++ {
 				c.Add(context.Background(), 1, map[string]string{"worker": "w"})
 			}
 			done <- struct{}{}
-		}(i)
+		}()
 	}
 	for i := 0; i < 10; i++ {
 		<-done
@@ -148,6 +172,40 @@ func TestPrometheusBackendConcurrentAdd(t *testing.T) {
 	assertMetricPresent(t, body, `concurrent_total{worker="w"} 1000`)
 }
 
+func TestPrometheusBackendAlreadyRegisteredCounter(t *testing.T) {
+	b := newTestBackend(t)
+	c1, err := b.NewCounter("dupe_counter_total", "duplicate counter", "result")
+	if err != nil {
+		t.Fatalf("first NewCounter returned error: %v", err)
+	}
+	c2, err := b.NewCounter("dupe_counter_total", "duplicate counter", "result")
+	if err != nil {
+		t.Fatalf("second NewCounter returned error: %v", err)
+	}
+
+	c1.Add(context.Background(), 1, map[string]string{"result": "ok"})
+	c2.Add(context.Background(), 2, map[string]string{"result": "ok"})
+
+	body := scrapeMetrics(t, b)
+	assertMetricPresent(t, body, `dupe_counter_total{result="ok"} 3`)
+}
+
+func TestPrometheusBackendInvalidLabelsNoPanic(t *testing.T) {
+	b := newTestBackend(t)
+	c, err := b.NewCounter("invalid_labels_total", "invalid labels test", "result")
+	if err != nil {
+		t.Fatalf("NewCounter returned error: %v", err)
+	}
+
+	// Extra label key should be dropped and must not panic.
+	c.Add(context.Background(), 5, map[string]string{"result": "ok", "unexpected": "x"})
+
+	body := scrapeMetrics(t, b)
+	if strings.Contains(body, `invalid_labels_total{result="ok"}`) {
+		t.Error("invalid label sample should have been dropped")
+	}
+}
+
 // --- helpers ---
 
 func scrapeMetrics(t *testing.T, b *obsprom.Backend) string {

main.go

@@ -3,6 +3,7 @@ package main
 import (
 	"bytes"
 	"context"
+	"crypto/tls"
 	"encoding/json"
 	"errors"
 	"flag"
@@ -11,10 +12,12 @@ import (
 	"log"
 	"net"
 	"net/http"
+	"net/http/httputil"
 	_ "net/http/pprof"
 	"os"
 	"os/exec"
 	"os/signal"
+	"path/filepath"
 	"runtime"
 	"runtime/pprof"
 	"strconv"
@@ -46,6 +49,7 @@ var (
 	doTrafficShaping bool
 	bandwidthLimit   string
 	ifbName          string // IFB device name for ingress traffic shaping
+	disableFirewall  bool
 )
 
 type WgConfig struct {
@@ -102,6 +106,144 @@ type UpdateDestinationsRequest struct {
 	Destinations []relay.PeerDestination `json:"destinations"`
 }
 
+// pangolinDestHeader carries the downstream host:port (reachable over the
+// WireGuard interface) that a /router request should be rewritten to,
+// optionally prefixed with a scheme (e.g. "https://100.96.128.1:443"). It is
+// stripped before the request is forwarded.
+const pangolinDestHeader = "p-dest-header"
+
+// pangolinHostHeader optionally carries the Host header value that should be
+// sent to the destination, when it differs from pangolinDestHeader (e.g. the
+// target's configured IP/hostname rather than the WireGuard routing
+// address). It is stripped before the request is forwarded. When absent,
+// the destination from pangolinDestHeader is used as the Host header, same
+// as before.
+const pangolinHostHeader = "p-dest-host-header"
+
+// splitDestHeader separates an optional "scheme://" prefix from a
+// pangolinDestHeader value, defaulting to "http" when none is present.
+func splitDestHeader(dest string) (scheme, host string) {
+	if s, rest, ok := strings.Cut(dest, "://"); ok {
+		return s, rest
+	}
+	return "http", dest
+}
+
+// hostname strips an optional ":port" suffix from a host header value.
+func hostname(hostport string) string {
+	if h, _, err := net.SplitHostPort(hostport); err == nil {
+		return h
+	}
+	return hostport
+}
+
+// routerSNIContextKey carries the TLS ServerName (SNI) that
+// routerTransport's DialTLSContext should present, since we dial the
+// WireGuard destination IP but the remote end typically terminates TLS
+// based on the original hostname (pangolinHostHeader), not that IP.
+type routerSNIContextKey struct{}
+
+// routerTransport is routerProxy's RoundTripper. It mirrors
+// http.DefaultTransport except for TLS dials, where it sets the SNI from
+// routerSNIContextKey instead of letting it default to the dial address
+// (the WireGuard IP), which the destination's TLS termination won't have a
+// matching certificate/route for.
+var routerTransport = &http.Transport{
+	Proxy: http.ProxyFromEnvironment,
+	DialTLSContext: func(ctx context.Context, network, addr string) (net.Conn, error) {
+		serverName := hostname(addr)
+		if sni, ok := ctx.Value(routerSNIContextKey{}).(string); ok && sni != "" {
+			serverName = sni
+		}
+		dialer := &tls.Dialer{Config: &tls.Config{ServerName: serverName}}
+		return dialer.DialContext(ctx, network, addr)
+	},
+}
+
+// logDebugRequest dumps a request's destination, headers, and body at debug
+// level for troubleshooting (e.g. verifying an auth header made it through
+// the proxy chain intact). It reads and restores req.Body so the request can
+// still be sent afterward. Header values (including secrets like API keys)
+// are logged as-is - only intended to be enabled for local troubleshooting.
+func logDebugRequest(label string, req *http.Request) {
+	var headerLines strings.Builder
+	for name, values := range req.Header {
+		for _, v := range values {
+			fmt.Fprintf(&headerLines, "\n  %s: %s", name, v)
+		}
+	}
+
+	body := []byte("<empty>")
+	if req.Body != nil {
+		data, err := io.ReadAll(req.Body)
+		req.Body.Close()
+		if err != nil {
+			logger.Error("%s: failed to read body for logging: %v", label, err)
+		} else {
+			body = data
+		}
+		req.Body = io.NopCloser(bytes.NewReader(data))
+	}
+
+	logger.Debug("%s: %s %s://%s%s headers:%s\nbody: %s", label, req.Method, req.URL.Scheme, req.Host, req.URL.RequestURI(), headerLines.String(), body)
+}
+
+// routerProxy forwards /router/* requests from the Pangolin AI gateway to a
+// destination on the WireGuard network, as named by pangolinDestHeader.
+// Used for proxying AI chat completion requests (incl. streaming) to
+// providers reachable only from a site.
+var routerProxy = &httputil.ReverseProxy{
+	Rewrite: func(pr *httputil.ProxyRequest) {
+		scheme, dest := splitDestHeader(pr.In.Header.Get(pangolinDestHeader))
+
+		pr.Out.URL.Scheme = scheme
+		pr.Out.URL.Host = dest
+		pr.Out.URL.Path = strings.TrimPrefix(pr.In.URL.Path, "/router")
+		if !strings.HasPrefix(pr.Out.URL.Path, "/") {
+			pr.Out.URL.Path = "/" + pr.Out.URL.Path
+		}
+		pr.Out.URL.RawPath = ""
+		pr.Out.Host = dest
+		if hostOverride := pr.In.Header.Get(pangolinHostHeader); hostOverride != "" {
+			pr.Out.Host = hostOverride
+			ctx := context.WithValue(pr.Out.Context(), routerSNIContextKey{}, hostname(hostOverride))
+			pr.Out = pr.Out.WithContext(ctx)
+		}
+
+		pr.Out.Header.Del(pangolinDestHeader)
+		pr.Out.Header.Del(pangolinHostHeader)
+
+		logger.Debug("Router proxy: %s %s -> %s (Host: %s)", pr.In.Method, pr.In.URL.Path, pr.Out.URL.String(), pr.Out.Host)
+		logDebugRequest("Router outbound request", pr.Out)
+	},
+	Transport: routerTransport,
+	// Flush written bytes to the client immediately rather than buffering,
+	// which is required for SSE-based streaming chat completions.
+	FlushInterval: -1,
+	ErrorHandler: func(w http.ResponseWriter, r *http.Request, err error) {
+		logger.Error("Router proxy error for %s: %v", r.URL.Path, err)
+		http.Error(w, "Bad gateway", http.StatusBadGateway)
+	},
+}
+
+func handleRouter(w http.ResponseWriter, r *http.Request) {
+	dest := r.Header.Get(pangolinDestHeader)
+	hostOverride := r.Header.Get(pangolinHostHeader)
+	logger.Debug("Router request received: %s %s dest=%s host=%s remote=%s", r.Method, r.URL.Path, dest, hostOverride, r.RemoteAddr)
+
+	if dest == "" {
+		http.Error(w, fmt.Sprintf("Missing %s header", pangolinDestHeader), http.StatusBadRequest)
+		return
+	}
+	_, host := splitDestHeader(dest)
+	if _, _, err := net.SplitHostPort(host); err != nil {
+		http.Error(w, "Invalid destination", http.StatusBadRequest)
+		return
+	}
+
+	routerProxy.ServeHTTP(w, r)
+}
+
 // httpMetricsMiddleware wraps HTTP handlers with metrics tracking
 func httpMetricsMiddleware(endpoint string, handler http.HandlerFunc) http.HandlerFunc {
 	return func(w http.ResponseWriter, r *http.Request) {
@@ -131,6 +273,16 @@ func (w *responseWriterWrapper) WriteHeader(statusCode int) {
 	w.ResponseWriter.WriteHeader(statusCode)
 }
 
+// Unwrap exposes the underlying ResponseWriter so http.ResponseController
+// (used by httputil.ReverseProxy's streaming Flush, and by Hijack/Push
+// callers) can see through this wrapper to the real http.Flusher etc.
+// Without this, ReverseProxy's flushes on /router/* silently no-op and
+// streamed responses (e.g. SSE) get buffered until the response completes
+// instead of being forwarded incrementally.
+func (w *responseWriterWrapper) Unwrap() http.ResponseWriter {
+	return w.ResponseWriter
+}
+
 func parseLogLevel(level string) logger.LogLevel {
 	switch strings.ToUpper(level) {
 	case "DEBUG":
@@ -149,8 +301,6 @@ func parseLogLevel(level string) logger.LogLevel {
 }
 
 func main() {
-	go monitorMemory(1024 * 1024 * 512) // trigger if memory usage exceeds 512MB
-
 	var (
 		err                  error
 		wgconfig             WgConfig
@@ -175,6 +325,7 @@ func main() {
 		otelMetricsEndpoint       string
 		otelMetricsInsecure       bool
 		otelMetricsExportInterval time.Duration
+		otelMetricsTimeout        time.Duration
 	)
 
 	interfaceName = os.Getenv("INTERFACE")
@@ -195,6 +346,7 @@ func main() {
 	proxyProtocolStr := os.Getenv("PROXY_PROTOCOL")
 	doTrafficShapingStr := os.Getenv("DO_TRAFFIC_SHAPING")
 	bandwidthLimitStr := os.Getenv("BANDWIDTH_LIMIT")
+	disableFirewallStr := os.Getenv("DISABLE_FIREWALL")
 
 	// Read metrics env vars (defaults applied by DefaultMetricsConfig; these override defaults).
 	metricsEnabled = true // default
@@ -229,6 +381,14 @@ func main() {
 			log.Printf("WARN: invalid OTEL_METRICS_EXPORT_INTERVAL=%q: %v", v, err2)
 		}
 	}
+	otelMetricsTimeout = 10 * time.Second // default
+	if v := os.Getenv("OTEL_METRICS_TIMEOUT"); v != "" {
+		if d, err2 := time.ParseDuration(v); err2 == nil {
+			otelMetricsTimeout = d
+		} else {
+			log.Printf("WARN: invalid OTEL_METRICS_TIMEOUT=%q: %v", v, err2)
+		}
+	}
 
 	if interfaceName == "" {
 		flag.StringVar(&interfaceName, "interface", "wg0", "Name of the WireGuard interface")
@@ -307,6 +467,13 @@ func main() {
 		flag.BoolVar(&doTrafficShaping, "do-traffic-shaping", false, "Whether to set up traffic shaping rules for peers (requires tc command and root privileges)")
 	}
 
+	if disableFirewallStr != "" {
+		disableFirewall = strings.ToLower(disableFirewallStr) == "true"
+	}
+	if disableFirewallStr == "" {
+		flag.BoolVar(&disableFirewall, "disable-firewall", false, "Disable WireGuard firewall rules to allow all inbound traffic on the interface")
+	}
+
 	if bandwidthLimitStr != "" {
 		bandwidthLimit = bandwidthLimitStr
 	}
@@ -322,9 +489,19 @@ func main() {
 	flag.StringVar(&otelMetricsEndpoint, "otel-metrics-endpoint", otelMetricsEndpoint, "OTLP collector endpoint (e.g. localhost:4317)")
 	flag.BoolVar(&otelMetricsInsecure, "otel-metrics-insecure", otelMetricsInsecure, "Disable TLS for OTLP connection")
 	flag.DurationVar(&otelMetricsExportInterval, "otel-metrics-export-interval", otelMetricsExportInterval, "Interval between OTLP metric pushes")
+	flag.DurationVar(&otelMetricsTimeout, "otel-metrics-timeout", otelMetricsTimeout, "Timeout for OTLP exporter setup")
 
 	flag.Parse()
 
+	// Heap profiles are dumped into the configured config directory (the same
+	// directory as the config file, or /var/config as a fallback for remote-config
+	// setups) so they land on the volume the operator actually has mounted.
+	heapDir := "/var/config/heap"
+	if configFile != "" {
+		heapDir = filepath.Join(filepath.Dir(configFile), "heap")
+	}
+	go monitorMemory(1024*1024*512, heapDir) // trigger if memory usage exceeds 512MB
+
 	// Derive IFB device name from the WireGuard interface name (Linux limit: 15 chars)
 	ifbName = "ifb_" + interfaceName
 	if len(ifbName) > 15 {
@@ -347,6 +524,7 @@ func main() {
 			Endpoint:       otelMetricsEndpoint,
 			Insecure:       otelMetricsInsecure,
 			ExportInterval: otelMetricsExportInterval,
+			Timeout:        otelMetricsTimeout,
 		},
 		ServiceName:           "gerbil",
 		ServiceVersion:        "1.0.0",
@@ -540,9 +718,12 @@ func main() {
 	http.HandleFunc("/update-destinations", httpMetricsMiddleware("update_destinations", handleUpdateDestinations))
 	http.HandleFunc("/update-local-snis", httpMetricsMiddleware("update_local_snis", handleUpdateLocalSNIs))
 	http.HandleFunc("/healthz", httpMetricsMiddleware("healthz", handleHealthz))
+	http.HandleFunc("/router/", httpMetricsMiddleware("router", handleRouter))
 
 	// Register metrics endpoint only for Prometheus backend.
 	// OTel backend pushes to a collector; no /metrics endpoint needed.
+	// Note: metricsPath is registered directly without httpMetricsMiddleware to prevent infinite recursion.
+	// The metricsHandler must not be wrapped by the middleware, as it would observe its own observation calls.
 	if metricsHandler != nil {
 		http.Handle(metricsPath, metricsHandler)
 		logger.Info("Metrics endpoint enabled at %s", metricsPath)
@@ -719,7 +900,9 @@ func ensureWireguardInterface(wgconfig WgConfig) error {
 		logger.Warn("Failed to ensure MSS clamping: %v", err)
 	}
 
-	if err := ensureWireguardFirewall(); err != nil {
+	if disableFirewall {
+		logger.Warn("Firewall disabled: all inbound traffic on %s will be allowed", interfaceName)
+	} else if err := ensureWireguardFirewall(wgconfig.IpAddress); err != nil {
 		logger.Warn("Failed to ensure WireGuard firewall rules: %v", err)
 	}
 
@@ -895,11 +1078,18 @@ func ensureMSSClamping() error {
 	return nil
 }
 
-func ensureWireguardFirewall() error {
+func ensureWireguardFirewall(localIpAddress string) error {
 	// Rules to enforce:
 	// 1. Allow established/related connections (responses to our outbound traffic)
 	// 2. Allow ICMP ping packets
-	// 3. Drop all other inbound traffic from peers
+	// 3. Allow inbound traffic to ports 80/443 on the local IP only (for Traefik)
+	// 4. Drop all other inbound traffic from peers
+
+	// Strip any CIDR suffix so we're left with just the host IP
+	localIp := localIpAddress
+	if ip, _, err := net.ParseCIDR(localIpAddress); err == nil {
+		localIp = ip.String()
+	}
 
 	// Define the rules we want to ensure exist
 	rules := [][]string{
@@ -919,6 +1109,24 @@ func ensureWireguardFirewall() error {
 			"--icmp-type", "8",
 			"-j", "ACCEPT",
 		},
+		// Allow inbound HTTP to the local IP only (for Traefik)
+		{
+			"-A", "INPUT",
+			"-i", interfaceName,
+			"-p", "tcp",
+			"--dport", "80",
+			"-d", localIp,
+			"-j", "ACCEPT",
+		},
+		// Allow inbound HTTPS to the local IP only (for Traefik)
+		{
+			"-A", "INPUT",
+			"-i", interfaceName,
+			"-p", "tcp",
+			"--dport", "443",
+			"-d", localIp,
+			"-j", "ACCEPT",
+		},
 		// Drop all other inbound traffic from WireGuard interface
 		{
 			"-A", "INPUT",
@@ -1162,10 +1370,12 @@ func removePeerInternal(publicKey string) error {
 
 	// Get current peer info before removing to clear relay connections and bandwidth limits
 	var wgIPs []string
+	allowedIPsCount := 0
 	device, err := wgClient.Device(interfaceName)
 	if err == nil {
 		for _, peer := range device.Peers {
 			if peer.PublicKey.String() == publicKey {
+				allowedIPsCount = len(peer.AllowedIPs)
 				// Extract WireGuard IPs from this peer's allowed IPs
 				for _, allowedIP := range peer.AllowedIPs {
 					wgIPs = append(wgIPs, allowedIP.IP.String())
@@ -1208,7 +1418,7 @@ func removePeerInternal(publicKey string) error {
 
 	// Record metrics
 	metrics.RecordPeersTotal(interfaceName, -1)
-	metrics.RecordAllowedIPsCount(interfaceName, publicKey, -int64(len(wgIPs)))
+	metrics.RecordAllowedIPsCount(interfaceName, publicKey, -int64(allowedIPsCount))
 
 	return nil
 }
@@ -1478,6 +1688,24 @@ func calculatePeerBandwidth() ([]PeerBandwidth, error) {
 	return peerBandwidths, nil
 }
 
+// defaultBandwidthReportBatchSize caps how many peer bandwidth readings are
+// sent in a single POST. Reporting every peer in one request grows unbounded
+// with fleet size and can exceed the remote server's request body limit,
+// which surfaces as "413 Payload Too Large" and silently drops that whole
+// report cycle. Overridable via GERBIL_BANDWIDTH_BATCH_SIZE.
+const defaultBandwidthReportBatchSize = 250
+
+var bandwidthReportBatchSize = loadBandwidthReportBatchSize()
+
+func loadBandwidthReportBatchSize() int {
+	if v := os.Getenv("GERBIL_BANDWIDTH_BATCH_SIZE"); v != "" {
+		if n, err := strconv.Atoi(v); err == nil && n > 0 {
+			return n
+		}
+	}
+	return defaultBandwidthReportBatchSize
+}
+
 func reportPeerBandwidth(apiURL string) error {
 	bandwidths, err := calculatePeerBandwidth()
 	if err != nil {
@@ -1486,26 +1714,47 @@ func reportPeerBandwidth(apiURL string) error {
 		return fmt.Errorf("failed to calculate peer bandwidth: %v", err)
 	}
 
-	jsonData, err := json.Marshal(bandwidths)
+	if len(bandwidths) == 0 {
+		return nil
+	}
+
+	var batchErrs []error
+	for start := 0; start < len(bandwidths); start += bandwidthReportBatchSize {
+		end := start + bandwidthReportBatchSize
+		if end > len(bandwidths) {
+			end = len(bandwidths)
+		}
+
+		if err := sendPeerBandwidthBatch(apiURL, bandwidths[start:end]); err != nil {
+			metrics.RecordBandwidthReport("error")
+			batchErrs = append(batchErrs, err)
+			continue
+		}
+		metrics.RecordBandwidthReport("success")
+	}
+
+	if len(batchErrs) > 0 {
+		return fmt.Errorf("failed to report %d bandwidth batch(es): %w", len(batchErrs), errors.Join(batchErrs...))
+	}
+	return nil
+}
+
+func sendPeerBandwidthBatch(apiURL string, batch []PeerBandwidth) error {
+	jsonData, err := json.Marshal(batch)
 	if err != nil {
-		metrics.RecordBandwidthReport("error")
 		return fmt.Errorf("failed to marshal bandwidth data: %v", err)
 	}
 
 	resp, err := http.Post(apiURL, "application/json", bytes.NewBuffer(jsonData))
 	if err != nil {
-		metrics.RecordBandwidthReport("error")
 		return fmt.Errorf("failed to send bandwidth data: %v", err)
 	}
 	defer resp.Body.Close()
 
 	if resp.StatusCode != http.StatusOK {
-		metrics.RecordBandwidthReport("error")
 		return fmt.Errorf("API returned non-OK status: %s", resp.Status)
 	}
 
-	// Record successful bandwidth report
-	metrics.RecordBandwidthReport("success")
 	return nil
 }
 
@@ -1534,7 +1783,7 @@ func notifyPeerChange(action, publicKey string) {
 	}
 }
 
-func monitorMemory(limit uint64) {
+func monitorMemory(limit uint64, heapDir string) {
 	var m runtime.MemStats
 	for {
 		runtime.ReadMemStats(&m)
@@ -1550,8 +1799,9 @@ func monitorMemory(limit uint64) {
 			// Record memory spike metric
 			metrics.RecordMemorySpike(severity)
 
-			f, err := os.Create(fmt.Sprintf("/var/config/heap/heap-spike-%d.pprof", time.Now().Unix()))
+			if err := os.MkdirAll(heapDir, 0o755); err != nil {
-			if err != nil {
+				log.Println("could not create heap profile directory:", err)
+			} else if f, err := os.Create(filepath.Join(heapDir, fmt.Sprintf("heap-spike-%d.pprof", time.Now().Unix()))); err != nil {
 				log.Println("could not create profile:", err)
 			} else {
 				pprof.WriteHeapProfile(f)

proxy/proxy.go

@@ -548,7 +548,7 @@ func (p *SNIProxy) handleConnection(clientConn net.Conn) {
 		logger.Debug("SNI extraction failed: %v", err)
 		return
 	}
-	metrics.RecordProxyTLSHandshake(hostname, time.Since(clientHelloStart).Seconds())
+	metrics.RecordProxyTLSHandshake(time.Since(clientHelloStart).Seconds())
 
 	if hostname == "" {
 		log.Println("No SNI hostname found")
@@ -596,8 +596,8 @@ func (p *SNIProxy) handleConnection(clientConn net.Conn) {
 	defer targetConn.Close()
 
 	logger.Debug("Connected to target: %s:%d", route.TargetHost, route.TargetPort)
-	metrics.RecordActiveProxyConnection(hostname, 1)
+	metrics.RecordActiveProxyConnection(1)
-	defer metrics.RecordActiveProxyConnection(hostname, -1)
+	defer metrics.RecordActiveProxyConnection(-1)
 
 	// Send PROXY protocol header if enabled
 	if p.proxyProtocol {
@@ -655,7 +655,7 @@ func (p *SNIProxy) getRoute(hostname, clientAddr string) (*RouteRecord, error) {
 	// Check local overrides first
 	if _, isOverride := p.localOverrides[hostname]; isOverride {
 		logger.Debug("Local override matched for hostname: %s", hostname)
-		metrics.RecordProxyRouteLookup("local_override", hostname)
+		metrics.RecordProxyRouteLookup("local_override")
 		return &RouteRecord{
 			Hostname:   hostname,
 			TargetHost: p.localProxyAddr,
@@ -668,7 +668,7 @@ func (p *SNIProxy) getRoute(hostname, clientAddr string) (*RouteRecord, error) {
 	_, isLocal := p.localSNIs[hostname]
 	p.localSNIsLock.RUnlock()
 	if isLocal {
-		metrics.RecordProxyRouteLookup("local", hostname)
+		metrics.RecordProxyRouteLookup("local")
 		return &RouteRecord{
 			Hostname:   hostname,
 			TargetHost: p.localProxyAddr,
@@ -679,16 +679,16 @@ func (p *SNIProxy) getRoute(hostname, clientAddr string) (*RouteRecord, error) {
 	// Check cache first
 	if cached, found := p.cache.Get(hostname); found {
 		if cached == nil {
-			metrics.RecordProxyRouteLookup("cached_not_found", hostname)
+			metrics.RecordProxyRouteLookup("cached_not_found")
 			return nil, nil // Cached negative result
 		}
 		logger.Debug("Cache hit for hostname: %s", hostname)
-		metrics.RecordProxyRouteLookup("cache_hit", hostname)
+		metrics.RecordProxyRouteLookup("cache_hit")
 		return cached.(*RouteRecord), nil
 	}
 
 	logger.Debug("Cache miss for hostname: %s, querying API", hostname)
-	metrics.RecordProxyRouteLookup("cache_miss", hostname)
+	metrics.RecordProxyRouteLookup("cache_miss")
 
 	// Query API with timeout
 	ctx, cancel := context.WithTimeout(p.ctx, 5*time.Second)
@@ -822,7 +822,7 @@ func (p *SNIProxy) pipe(hostname string, clientConn, targetConn net.Conn, client
 		}()
 
 		bytesCopied, err := io.CopyBuffer(targetConn, clientReader, *bufPtr)
-		metrics.RecordProxyBytesTransmitted(hostname, "client_to_target", bytesCopied)
+		metrics.RecordProxyBytesTransmitted("client_to_target", bytesCopied)
 		if err != nil && err != io.EOF {
 			logger.Debug("Copy client->target error: %v", err)
 		}
@@ -842,7 +842,7 @@ func (p *SNIProxy) pipe(hostname string, clientConn, targetConn net.Conn, client
 		}()
 
 		bytesCopied, err := io.CopyBuffer(clientConn, targetConn, *bufPtr)
-		metrics.RecordProxyBytesTransmitted(hostname, "target_to_client", bytesCopied)
+		metrics.RecordProxyBytesTransmitted("target_to_client", bytesCopied)
 		if err != nil && err != io.EOF {
 			logger.Debug("Copy target->client error: %v", err)
 		}

relay/relay.go

@@ -1,6 +1,7 @@
 package relay
 
 import (
+	"bufio"
 	"bytes"
 	"context"
 	"encoding/binary"
@@ -9,8 +10,11 @@ import (
 	"io"
 	"net"
 	"net/http"
+	"os"
 	"runtime"
+	"strconv"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/fosrl/gerbil/internal/metrics"
@@ -59,8 +63,31 @@ type PeerDestination struct {
 }
 
 type DestinationConn struct {
-	conn     *net.UDPConn
+	conn *net.UDPConn
-	lastUsed time.Time
+	// lastUsed is unix nanoseconds, read/written via atomic ops since it's
+	// touched from packet workers and the response goroutine concurrently
+	// with no lock, and is also scanned for LRU eviction below.
+	lastUsed atomic.Int64
+}
+
+// defaultMaxUDPConnections caps the number of concurrent per-peer outbound
+// UDP sockets the relay will keep open in s.connections. Without a cap, a
+// burst of peer churn creates sockets faster than the 5-minute idle cleanup
+// can reap them, exhausting the host's ephemeral port range or fd ulimit.
+// That surfaces as "dial udp ...: resource temporarily unavailable" on every
+// subsequent packet and pegs the CPU logging the flood (outage 2026-07-03,
+// recurrence 2026-07-05). Overridable via GERBIL_MAX_UDP_CONNECTIONS.
+const defaultMaxUDPConnections = 8192
+
+var maxUDPConnections = loadMaxUDPConnections()
+
+func loadMaxUDPConnections() int64 {
+	if v := os.Getenv("GERBIL_MAX_UDP_CONNECTIONS"); v != "" {
+		if n, err := strconv.ParseInt(v, 10, 64); err == nil && n > 0 {
+			return n
+		}
+	}
+	return defaultMaxUDPConnections
 }
 
 // Type for storing WireGuard handshake information
@@ -137,6 +164,24 @@ const (
 	WireGuardMessageTypeTransportData       = 4
 )
 
+// cachedEndpointState holds the last-known endpoint fields used for change detection.
+// Timestamp is intentionally excluded since it always changes.
+type cachedEndpointState struct {
+	OlmID     string
+	NewtID    string
+	Token     string
+	IP        string
+	Port      int
+	PublicKey string
+}
+
+// cachedEndpointEntry wraps cachedEndpointState with a timestamp so the cache
+// can be expired after a short TTL even when the endpoint fields are unchanged.
+type cachedEndpointEntry struct {
+	state    cachedEndpointState
+	cachedAt time.Time
+}
+
 // --- End Types ---
 
 // bufferPool allows reusing buffers to reduce allocations.
@@ -153,10 +198,13 @@ type UDPProxyServer struct {
 	conn          *net.UDPConn
 	proxyMappings sync.Map // map[string]ProxyMapping where key is "ip:port"
 	connections   sync.Map // map[string]*DestinationConn where key is destination "ip:port"
-	privateKey    wgtypes.Key
+	// connectionCount mirrors len(connections) without an O(n) sync.Map walk,
-	packetChan    chan Packet
+	// so the cap check in getOrCreateConnection is cheap on the hot path.
-	ctx           context.Context
+	connectionCount atomic.Int64
-	cancel        context.CancelFunc
+	privateKey      wgtypes.Key
+	packetChan      chan Packet
+	ctx             context.Context
+	cancel          context.CancelFunc
 
 	// Session tracking for WireGuard peers
 	// Key format: "senderIndex:receiverIndex"
@@ -172,6 +220,12 @@ type UDPProxyServer struct {
 	// Cache for resolved UDP addresses to avoid per-packet DNS lookups
 	// Key: "ip:port" string, Value: *net.UDPAddr
 	addrCache sync.Map
+	// lastEndpointCache stores the last-known endpoint state per client (key: olmId:newtId)
+	// used to skip redundant HTTP notifications when nothing has changed.
+	lastEndpointCache sync.Map
+	// notifyChan is the async queue for hole-punch endpoint notifications.
+	// Dedicated notifier workers drain this channel and perform the HTTP call.
+	notifyChan chan ClientEndpoint
 	// ReachableAt is the URL where this server can be reached
 	ReachableAt string
 }
@@ -184,6 +238,7 @@ func NewUDPProxyServer(parentCtx context.Context, addr, serverURL string, privat
 		serverURL:   serverURL,
 		privateKey:  privateKey,
 		packetChan:  make(chan Packet, 50000), // Increased from 1000 to handle high throughput
+		notifyChan:  make(chan ClientEndpoint, 1000),
 		ReachableAt: reachableAt,
 		ctx:         ctx,
 		cancel:      cancel,
@@ -192,10 +247,15 @@ func NewUDPProxyServer(parentCtx context.Context, addr, serverURL string, privat
 
 // Start sets up the UDP listener, worker pool, and begins reading packets.
 func (s *UDPProxyServer) Start() error {
-	// Fetch initial mappings.
+	// Fetch initial mappings asynchronously so a large (potentially 100MB+)
-	if err := s.fetchInitialMappings(); err != nil {
+	// response does not block the UDP listener from coming up. Any packets
-		return fmt.Errorf("failed to fetch initial mappings: %v", err)
+	// arriving for unknown mappings before the load completes will simply
-	}
+	// log and be repopulated via the hole-punch path.
+	go func() {
+		if err := s.fetchInitialMappings(); err != nil {
+			logger.Error("Failed to fetch initial mappings: %v", err)
+		}
+	}()
 
 	udpAddr, err := net.ResolveUDPAddr("udp", s.addr)
 	if err != nil {
@@ -237,6 +297,11 @@ func (s *UDPProxyServer) Start() error {
 	// Start the hole punch rate limiter cleanup routine
 	go s.cleanupHolePunchRateLimiter()
 
+	// Start async endpoint notifier workers (HTTP calls off the hot path)
+	for i := 0; i < 5; i++ {
+		go s.endpointNotifierWorker()
+	}
+
 	return nil
 }
 
@@ -375,7 +440,43 @@ func (s *UDPProxyServer) packetWorker() {
 				ClientPublicKey:   msg.PublicKey,
 			}
 			logger.Debug("Created endpoint from packet remoteAddr %s: IP=%s, Port=%d", packet.remoteAddr.String(), endpoint.IP, endpoint.Port)
-			s.notifyServer(endpoint)
+
+			// Check if anything meaningful changed before queuing an HTTP notification.
+			// The cache expires after 2.5 s so the server always receives a fresh
+			// timestamp within its 5-second staleness window.
+			const endpointCacheTTL = 2500 * time.Millisecond
+			cacheKey := endpoint.OlmID + ":" + endpoint.NewtID
+			newState := cachedEndpointState{
+				OlmID:     endpoint.OlmID,
+				NewtID:    endpoint.NewtID,
+				Token:     endpoint.Token,
+				IP:        endpoint.IP,
+				Port:      endpoint.Port,
+				PublicKey: endpoint.ClientPublicKey,
+			}
+			if cached, ok := s.lastEndpointCache.Load(cacheKey); ok {
+				entry := cached.(cachedEndpointEntry)
+				if entry.state == newState && time.Since(entry.cachedAt) < endpointCacheTTL {
+					// Endpoint unchanged and cache still fresh - skip the HTTP call.
+					logger.Debug("Endpoint unchanged for %s, skipping notification", cacheKey)
+					metrics.RecordHolePunchEvent(relayIfname, "deduplicated")
+					s.clearSessionsForIP(endpoint.IP)
+					metrics.RecordHolePunchEvent(relayIfname, "success")
+					bufferPool.Put(packet.data[:1500])
+					continue
+				}
+			}
+			s.lastEndpointCache.Store(cacheKey, cachedEndpointEntry{state: newState, cachedAt: time.Now()})
+
+			// Queue the notification asynchronously so the hot path is not blocked by HTTP.
+			select {
+			case s.notifyChan <- endpoint:
+			case <-s.ctx.Done():
+				// shutting down
+			default:
+				logger.Debug("Notification queue full, dropping hole punch notification for %s:%d", endpoint.IP, endpoint.Port)
+				metrics.RecordHolePunchEvent(relayIfname, "queue_full")
+			}
 			s.clearSessionsForIP(endpoint.IP) // Clear sessions for this IP to allow re-establishment
 			metrics.RecordHolePunchEvent(relayIfname, "success")
 		}
@@ -384,6 +485,22 @@ func (s *UDPProxyServer) packetWorker() {
 	}
 }
 
+// endpointNotifierWorker drains the notifyChan and performs the HTTP notification for each
+// hole-punch endpoint. Running several of these keeps latency low even when the server is slow.
+func (s *UDPProxyServer) endpointNotifierWorker() {
+	for {
+		select {
+		case endpoint, ok := <-s.notifyChan:
+			if !ok {
+				return
+			}
+			s.notifyServer(endpoint)
+		case <-s.ctx.Done():
+			return
+		}
+	}
+}
+
 // decryptMessage decrypts the message using the server's private key
 func (s *UDPProxyServer) decryptMessage(encMsg EncryptedHolePunchMessage) ([]byte, error) {
 	// Parse the ephemeral public key
@@ -419,6 +536,7 @@ func (s *UDPProxyServer) decryptMessage(encMsg EncryptedHolePunchMessage) ([]byt
 }
 
 func (s *UDPProxyServer) fetchInitialMappings() error {
+	logger.Info("Requesting initial proxy mappings")
 	body := bytes.NewBuffer([]byte(fmt.Sprintf(`{"publicKey": "%s"}`, s.privateKey.PublicKey().String())))
 	resp, err := http.Post(s.serverURL+"/gerbil/get-all-relays", "application/json", body)
 	if err != nil {
@@ -430,24 +548,82 @@ func (s *UDPProxyServer) fetchInitialMappings() error {
 		return fmt.Errorf("server returned non-OK status: %d, body: %s",
 			resp.StatusCode, string(body))
 	}
-	data, err := io.ReadAll(resp.Body)
+	logger.Info("Received initial mappings, streaming decode")
+
+	// Stream-decode the response instead of buffering the entire body
+	// (which can be 100MB+) and then re-walking it with json.Unmarshal.
+	// This both lowers peak memory and lets us start populating the
+	// sync.Map as entries arrive.
+	dec := json.NewDecoder(bufio.NewReaderSize(resp.Body, 1<<20))
+
+	// Expect opening '{' of the top-level object.
+	tok, err := dec.Token()
 	if err != nil {
-		return fmt.Errorf("failed to read response body: %v", err)
+		return fmt.Errorf("failed to read opening token: %v", err)
 	}
-	logger.Info("Received initial mappings: %s", string(data))
+	if d, ok := tok.(json.Delim); !ok || d != '{' {
-	var initialMappings InitialMappings
+		return fmt.Errorf("expected '{' at top level, got %v", tok)
-	if err := json.Unmarshal(data, &initialMappings); err != nil {
-		return fmt.Errorf("failed to unmarshal initial mappings: %v", err)
 	}
-	// Store mappings in our sync.Map.
+
-	for key, mapping := range initialMappings.Mappings {
+	count := 0
-		// Initialize LastUsed timestamp for initial mappings
+	now := time.Now()
-		mapping.LastUsed = time.Now()
+
-		s.proxyMappings.Store(key, mapping)
+	for dec.More() {
+		keyTok, err := dec.Token()
+		if err != nil {
+			return fmt.Errorf("failed to read top-level key: %v", err)
+		}
+		key, ok := keyTok.(string)
+		if !ok {
+			return fmt.Errorf("expected string key at top level, got %T", keyTok)
+		}
+
+		if key != "mappings" {
+			// Skip unknown top-level fields without materializing them.
+			var skip json.RawMessage
+			if err := dec.Decode(&skip); err != nil {
+				return fmt.Errorf("failed to skip field %q: %v", key, err)
+			}
+			continue
+		}
+
+		// Expect opening '{' of the mappings object.
+		tok, err := dec.Token()
+		if err != nil {
+			return fmt.Errorf("failed to read mappings open: %v", err)
+		}
+		if d, ok := tok.(json.Delim); !ok || d != '{' {
+			return fmt.Errorf("expected '{' for mappings, got %v", tok)
+		}
+
+		for dec.More() {
+			mapKeyTok, err := dec.Token()
+			if err != nil {
+				return fmt.Errorf("failed to read mapping key: %v", err)
+			}
+			mapKey, ok := mapKeyTok.(string)
+			if !ok {
+				return fmt.Errorf("expected string mapping key, got %T", mapKeyTok)
+			}
+
+			var mapping ProxyMapping
+			if err := dec.Decode(&mapping); err != nil {
+				return fmt.Errorf("failed to decode mapping %q: %v", mapKey, err)
+			}
+			mapping.LastUsed = now
+			s.proxyMappings.Store(mapKey, mapping)
+			count++
+		}
+
+		// Consume closing '}' of mappings object.
+		if _, err := dec.Token(); err != nil {
+			return fmt.Errorf("failed to read mappings close: %v", err)
+		}
 	}
-	metrics.RecordProxyInitialMappings(relayIfname, int64(len(initialMappings.Mappings)))
+
-	metrics.RecordProxyMapping(relayIfname, int64(len(initialMappings.Mappings)))
+	metrics.RecordProxyInitialMappings(relayIfname, int64(count))
-	logger.Info("Loaded %d initial proxy mappings", len(initialMappings.Mappings))
+	metrics.RecordProxyMapping(relayIfname, int64(count))
+	logger.Info("Loaded %d initial proxy mappings", count)
 	return nil
 }
 
@@ -715,10 +891,17 @@ func (s *UDPProxyServer) getOrCreateConnection(destAddr *net.UDPAddr, remoteAddr
 	// Check if we have an existing connection
 	if conn, ok := s.connections.Load(key); ok {
 		destConn := conn.(*DestinationConn)
-		destConn.lastUsed = time.Now()
+		destConn.lastUsed.Store(time.Now().UnixNano())
 		return destConn.conn, nil
 	}
 
+	// Enforce a hard cap on concurrent sockets so a burst of peer churn can't
+	// exhaust the host's ephemeral ports/fds. Evict the least-recently-used
+	// connection to make room instead of growing unbounded.
+	if s.connectionCount.Load() >= maxUDPConnections {
+		s.evictLRUConnection()
+	}
+
 	// Create new connection
 	newConn, err := net.DialUDP("udp", nil, destAddr)
 	if err != nil {
@@ -726,11 +909,17 @@ func (s *UDPProxyServer) getOrCreateConnection(destAddr *net.UDPAddr, remoteAddr
 		return nil, fmt.Errorf("failed to create UDP connection: %v", err)
 	}
 
-	// Store the new connection
+	destConn := &DestinationConn{conn: newConn}
-	s.connections.Store(key, &DestinationConn{
+	destConn.lastUsed.Store(time.Now().UnixNano())
-		conn:     newConn,
+
-		lastUsed: time.Now(),
+	// Store the new connection. If another goroutine raced us and already
-	})
+	// created one for this key, close ours and use theirs instead.
+	if existing, loaded := s.connections.LoadOrStore(key, destConn); loaded {
+		newConn.Close()
+		return existing.(*DestinationConn).conn, nil
+	}
+	s.connectionCount.Add(1)
+	metrics.RecordUDPConnection(relayIfname, 1)
 
 	// Start a goroutine to handle responses
 	go s.handleResponses(newConn, destAddr, remoteAddr)
@@ -738,6 +927,33 @@ func (s *UDPProxyServer) getOrCreateConnection(destAddr *net.UDPAddr, remoteAddr
 	return newConn, nil
 }
 
+// evictLRUConnection closes and removes the least-recently-used destination
+// connection so a new one can be created under the concurrent connection cap.
+func (s *UDPProxyServer) evictLRUConnection() {
+	var oldestKey interface{}
+	var oldestConn *DestinationConn
+	var oldestTime int64
+
+	s.connections.Range(func(key, value interface{}) bool {
+		destConn := value.(*DestinationConn)
+		lu := destConn.lastUsed.Load()
+		if oldestKey == nil || lu < oldestTime {
+			oldestKey = key
+			oldestConn = destConn
+			oldestTime = lu
+		}
+		return true
+	})
+
+	if oldestKey != nil {
+		s.connections.Delete(oldestKey)
+		oldestConn.conn.Close()
+		s.connectionCount.Add(-1)
+		metrics.RecordUDPConnection(relayIfname, -1)
+		metrics.RecordProxyCleanupRemoved(relayIfname, "conn_evicted", 1)
+	}
+}
+
 func (s *UDPProxyServer) handleResponses(conn *net.UDPConn, destAddr *net.UDPAddr, remoteAddr *net.UDPAddr) {
 	buffer := make([]byte, 1500)
 	for {
@@ -789,22 +1005,32 @@ func (s *UDPProxyServer) handleResponses(conn *net.UDPConn, destAddr *net.UDPAdd
 
 // Add a cleanup method to periodically remove idle connections
 func (s *UDPProxyServer) cleanupIdleConnections() {
-	ticker := time.NewTicker(5 * time.Minute)
+	// Ticker interval and idle threshold were previously 5min/10min, meaning
+	// a socket could sit open for up to 15 minutes after going idle. Under a
+	// reconnect/churn burst that lag is enough to exhaust ephemeral ports
+	// before cleanup catches up, so both are tightened here.
+	ticker := time.NewTicker(1 * time.Minute)
 	defer ticker.Stop()
 	for {
 		select {
 		case <-ticker.C:
 			cleanupStart := time.Now()
-			now := time.Now()
+			now := time.Now().UnixNano()
+			removed := int64(0)
 			s.connections.Range(func(key, value interface{}) bool {
 				destConn := value.(*DestinationConn)
-				if now.Sub(destConn.lastUsed) > 10*time.Minute {
+				if now-destConn.lastUsed.Load() > int64(5*time.Minute) {
 					destConn.conn.Close()
 					s.connections.Delete(key)
-					metrics.RecordProxyCleanupRemoved(relayIfname, "conn", 1)
+					removed++
 				}
 				return true
 			})
+			if removed > 0 {
+				s.connectionCount.Add(-removed)
+				metrics.RecordUDPConnection(relayIfname, -removed)
+				metrics.RecordProxyCleanupRemoved(relayIfname, "conn", removed)
+			}
 			metrics.RecordProxyIdleCleanupDuration(relayIfname, "conn", time.Since(cleanupStart).Seconds())
 		case <-s.ctx.Done():
 			return
@@ -966,6 +1192,10 @@ func (s *UDPProxyServer) clearConnectionsForWGIP(wgIP string) {
 	for _, key := range keysToDelete {
 		s.connections.Delete(key)
 	}
+	if len(keysToDelete) > 0 {
+		s.connectionCount.Add(-int64(len(keysToDelete)))
+		metrics.RecordUDPConnection(relayIfname, -int64(len(keysToDelete)))
+	}
 
 	logger.Info("Cleared %d connections for WG IP: %s", len(keysToDelete), wgIP)
 }
@@ -1163,33 +1393,56 @@ func (s *UDPProxyServer) trackCommunicationPattern(fromAddr, toAddr *net.UDPAddr
 
 // tryRebuildSession attempts to rebuild a WireGuard session from communication patterns
 func (s *UDPProxyServer) tryRebuildSession(pattern *CommunicationPattern) {
+	// Require both indices and a minimum amount of bidirectional traffic
+	if pattern.ClientIndex == 0 || pattern.DestIndex == 0 || pattern.PacketCount < 4 {
+		return
+	}
+
 	// Check if we have bidirectional communication within a reasonable time window
 	timeDiff := pattern.LastFromClient.Sub(pattern.LastFromDest)
 	if timeDiff < 0 {
 		timeDiff = -timeDiff
 	}
-
+	if timeDiff >= 30*time.Second {
-	// Only rebuild if we have recent bidirectional communication and both indices
+		return
-	if timeDiff < 30*time.Second && pattern.ClientIndex != 0 && pattern.DestIndex != 0 && pattern.PacketCount >= 4 {
-		// Create session mapping: client's index maps to destination
-		sessionKey := fmt.Sprintf("%d:%d", pattern.DestIndex, pattern.ClientIndex)
-
-		// Check if we already have this session
-		session := &WireGuardSession{
-			ReceiverIndex: pattern.DestIndex,
-			SenderIndex:   pattern.ClientIndex,
-			DestAddr:      pattern.ToDestination,
-			LastSeen:      time.Now(),
-		}
-		if _, loaded := s.wgSessions.LoadOrStore(sessionKey, session); loaded {
-			s.wgSessions.Store(sessionKey, session)
-		} else {
-			metrics.RecordSession(relayIfname, 1)
-			metrics.RecordSessionRebuilt(relayIfname)
-		}
-		logger.Info("Rebuilt WireGuard session from communication pattern: %s -> %s (packets: %d)",
-			sessionKey, pattern.ToDestination.String(), pattern.PacketCount)
 	}
+
+	sessionKey := fmt.Sprintf("%d:%d", pattern.DestIndex, pattern.ClientIndex)
+	destStr := pattern.ToDestination.String()
+
+	// Fast path: if a matching session already exists, just refresh LastSeen and bail out.
+	// This prevents log spam and repeated work for every packet of an established flow.
+	if existing, ok := s.wgSessions.Load(sessionKey); ok {
+		sess := existing.(*WireGuardSession)
+		if da := sess.GetDestAddr(); da != nil && da.String() == destStr {
+			sess.UpdateLastSeen()
+			// Make sure the receiver-index fast-path is populated so future packets
+			// don't keep falling back to broadcast + pattern tracking.
+			if _, indexed := s.sessionsByReceiverIndex.Load(pattern.ClientIndex); !indexed {
+				s.sessionsByReceiverIndex.Store(pattern.ClientIndex, sess)
+			}
+			return
+		}
+	}
+
+	// Create or replace the session mapping
+	session := &WireGuardSession{
+		ReceiverIndex: pattern.DestIndex,
+		SenderIndex:   pattern.ClientIndex,
+		DestAddr:      pattern.ToDestination,
+		LastSeen:      time.Now(),
+	}
+	if _, loaded := s.wgSessions.LoadOrStore(sessionKey, session); loaded {
+		s.wgSessions.Store(sessionKey, session)
+	} else {
+		metrics.RecordSession(relayIfname, 1)
+		metrics.RecordSessionRebuilt(relayIfname)
+	}
+	// Index by client receiver index so the transport-data fast path can find it.
+	s.sessionsByReceiverIndex.Store(pattern.ClientIndex, session)
+
+	logger.Info("Rebuilt WireGuard session from communication pattern: %s -> %s (packets: %d)",
+		sessionKey, destStr, pattern.PacketCount)
 }
 
 // cleanupIdleCommunicationPatterns periodically removes idle communication patterns