[client] Treat answering upstreams as reachable and widen DNS health grace window (#6453)

This commit is contained in:
Viktor Liu
2026-06-19 17:32:49 +02:00
committed by GitHub
parent 883a1a8961
commit 15a0504fb1
4 changed files with 133 additions and 7 deletions
+29 -4
View File
@@ -6,6 +6,7 @@ import (
"fmt"
"net/netip"
"net/url"
"os"
"slices"
"strings"
"sync"
@@ -38,11 +39,15 @@ const (
// defaultWarningDelayBase is the starting grace window before a
// "Nameserver group unreachable" event fires for a group that's
// never been healthy and only has overlay upstreams with no
// Connected peer. Per-server and overridable; see warningDelayFor.
defaultWarningDelayBase = 30 * time.Second
// Connected peer. Per-server and overridable via envWarningDelay;
// see warningDelay.
defaultWarningDelayBase = 60 * time.Second
// warningDelayBonusCap caps the route-count bonus added to the
// base grace window. See warningDelayFor.
// base grace window. See warningDelay.
warningDelayBonusCap = 30 * time.Second
// envWarningDelay overrides defaultWarningDelayBase with a Go duration
// string (e.g. "90s", "2m"). Invalid or non-positive values are ignored.
envWarningDelay = "NB_DNS_HEALTH_WARNING_DELAY"
)
// errNoUsableNameservers signals that a merged-domain group has no usable
@@ -298,7 +303,7 @@ func newDefaultServer(
hostManager: &noopHostConfigurator{},
mgmtCacheResolver: mgmtCacheResolver,
currentConfigHash: ^uint64(0), // Initialize to max uint64 to ensure first config is always applied
warningDelayBase: defaultWarningDelayBase,
warningDelayBase: warningDelayBaseFromEnv(),
healthRefresh: make(chan struct{}, 1),
}
// Wire the local resolver against the peer status recorder so it can
@@ -1154,6 +1159,26 @@ func (s *DefaultServer) projectUnhealthy(p *nsGroupProj, servers []netip.AddrPor
return false
}
// warningDelayBaseFromEnv returns the base grace window, honoring
// envWarningDelay when it holds a valid positive Go duration. Invalid or
// non-positive values fall back to defaultWarningDelayBase.
func warningDelayBaseFromEnv() time.Duration {
val := os.Getenv(envWarningDelay)
if val == "" {
return defaultWarningDelayBase
}
d, err := time.ParseDuration(val)
if err != nil {
log.Warnf("invalid %s value %q, using default %v: %v", envWarningDelay, val, defaultWarningDelayBase, err)
return defaultWarningDelayBase
}
if d <= 0 {
log.Warnf("%s must be positive, got %v, using default %v", envWarningDelay, d, defaultWarningDelayBase)
return defaultWarningDelayBase
}
return d
}
// warningDelay returns the grace window for the given selected-route
// count. Scales gently: +1s per 100 routes, capped by
// warningDelayBonusCap. Parallel handshakes mean handshake time grows