[management,proxy] Agent network: per-account LLM gateway (policy, metering, multi-provider) (#6555)

* [agent-network] Shared proto, OpenAPI schema, and generated types

* [agent-network] Management: store, manager, synthesizer, policy engine, provider catalog, HTTP/gRPC API

Adds the account-scoped agent-network module: provider/policy/budget CRUD and
store, the reverse-proxy service synthesizer, policy selection + limit
enforcement, the provider catalog (incl. Vertex AI and AWS Bedrock entries),
and the management HTTP + proxy gRPC surfaces.

* [management] Fix agent-network proxy-peer fan-out on affected-peer recompute

The affected-peers resolver loaded only persisted reverse-proxy services, but
agent-network services are synthesized on demand and never persisted. As a
result the embedded proxy peer was never folded into the affected set when a
client's group changed, so the proxy received no network-map update for a newly
authorised client and rejected its handshake until a full resync (restart).

loadProxyServices now merges the synthesized agent-network services (injected
via a registration hook to avoid an import cycle), so proxy peers learn newly
authorised clients immediately.

* [proxy] Reverse-proxy middleware framework, chain, and request plumbing

The per-target middleware chain (slots, dispatcher, mutation gate, metadata
merger), body capture, access-log terminal sink, and the proxy wiring that
builds + runs chains for synthesized agent-network services.

* [proxy] LLM parsers, pricing, and builtin middlewares (OpenAI, Anthropic, Vertex AI, AWS Bedrock)

Request/response parsers and SSE/event-stream metering, the embedded pricing
table, and the builtin middleware set: request parser, router, policy
limit-check/record, cost meter, guardrail, identity inject, response parser.
Includes the path-routed providers — Google Vertex AI (keyfile:: service-account
OAuth minting) and AWS Bedrock (bearer auth, invoke/converse/streaming, optional
/bedrock prefix) — plus the Models allowlist and unmeterable-publisher deny.

* [proxy] IPv6 in-place apply and TCP accept-loop hardening on netstack listeners

* [agent-network] End-to-end test suite, module docs, and deployment preset

* [agent-network] Fix codespell typos and exclude false positives

- labelgen word pool: vermillion -> vermilion, racoon -> raccoon.
- codespell ignore list: add flate (Go compress/flate package), recordin
  (a test-local identifier), and unparseable (a valid alternative spelling used
  consistently across identifiers + a metadata-value constant).

* [management] Set LastSeen on injected proxy peer in realstack test (MySQL strict-mode)

The injected embedded proxy peer had a PeerStatus with a zero LastSeen, which
serializes to '0000-00-00' and is rejected by MySQL in strict mode (SQLite
tolerates it). Set LastSeen to a valid time so SaveAccount succeeds on both
engines.

* [agent-network] Remove e2e shell-script suite from this branch

The end-to-end shell scripts under scripts/e2e/ are maintained in a separate
testing suite and are not part of this change set.

* [agent-network] Polish module docs: remove internal review scaffolding, fix links, verify diagrams

Strip PR-review framing, commit references, absolute paths, and stale internal
references from the agent-network module docs; fix broken relative links; verify
all diagrams against the current architecture. Remove the internal AI-reviewer
prompt file.

* [management] Refine session expiration handling to support 3-state encoding for SSO deadlines

* [agent-network] Relocate agentnetwork package to internals/modules

Move management/server/agentnetwork (and its catalog/, labelgen/, types/
subpackages) to management/internals/modules/agentnetwork, alongside the
reverse-proxy module, and rewrite all importers. Pure relocation: package names,
the synthesizer + affectedpeers registration hook, and store access (shared
store.Store) are unchanged, so no import cycle is introduced (affectedpeers
still depends only on the agentnetwork/types leaf).

* [agent-network] Co-locate HTTP handlers in the module (RegisterEndpoints)

Move the agent-network HTTP handlers from server/http/handlers/agentnetwork into
the module at internals/modules/agentnetwork/handlers (package handlers) and
rename the entrypoint AddEndpoints -> RegisterEndpoints, matching the
reverse-proxy module convention. Wiring in http/handler.go updated accordingly.
This commit is contained in:
Maycon Santos
2026-06-27 13:41:00 +02:00
committed by GitHub
parent 615631567a
commit b416063bcc
187 changed files with 36835 additions and 660 deletions
+122
View File
@@ -533,3 +533,125 @@ MHcCAQEEIIrYSSNQFaA2Hwf1duRSxKtLYX5CB04fSeQ6tF1aY/PuoAoGCCqGSM49
AwEHoUQDQgAEPR3tU2Fta9ktY+6P9G0cWO+0kETA6SFs38GecTyudlHz6xvCdz8q
EKTcWGekdmdDPsHloRNtsiCa697B2O9IFA==
-----END EC PRIVATE KEY-----`)
// scriptedAcceptListener returns pre-scripted errors from Accept(). Used
// to drive the feedRouterFromListener tests without binding a real
// socket — the production code path is a netstack-backed listener that
// returns gVisor's "endpoint is in invalid state" forever after its
// endpoint is destroyed.
type scriptedAcceptListener struct {
errs chan error
closed chan struct{}
}
func newScriptedAcceptListener(errs ...error) *scriptedAcceptListener {
s := &scriptedAcceptListener{
errs: make(chan error, len(errs)+1),
closed: make(chan struct{}),
}
for _, e := range errs {
s.errs <- e
}
return s
}
func (s *scriptedAcceptListener) Accept() (net.Conn, error) {
select {
case <-s.closed:
return nil, net.ErrClosed
case err := <-s.errs:
return nil, err
}
}
func (s *scriptedAcceptListener) Close() error {
select {
case <-s.closed:
default:
close(s.closed)
}
return nil
}
func (s *scriptedAcceptListener) Addr() net.Addr {
return &net.TCPAddr{IP: net.IPv4(127, 0, 0, 1), Port: 0}
}
// errSentinel carries a literal error message so tests can synthesise
// the exact gVisor text without importing the netstack package.
type errSentinel string
func (e errSentinel) Error() string { return string(e) }
// TestFeedRouterFromListener_ExitsOnGVisorInvalidEndpoint is the
// regression guard for the inbound side of the tight-loop bug. The
// per-account plain-HTTP feeder must recognise gVisor's "endpoint is in
// invalid state" and exit, otherwise it pegs a CPU core and floods the
// account-scoped log with the same accept error every iteration.
func TestFeedRouterFromListener_ExitsOnGVisorInvalidEndpoint(t *testing.T) {
logger := log.StandardLogger()
addr := &net.TCPAddr{IP: net.IPv4(127, 0, 0, 1), Port: 80}
router := nbtcp.NewRouter(logger, nil, addr)
gvisorErr := &net.OpError{
Op: "accept",
Net: "tcp",
Addr: addr,
Err: errSentinel("endpoint is in invalid state"),
}
ln := newScriptedAcceptListener(gvisorErr)
defer ln.Close()
done := make(chan struct{})
go func() {
defer close(done)
feedRouterFromListener(context.Background(), ln, router, logger, "acct-1")
}()
select {
case <-done:
// Expected: loop recognised the gVisor error and returned.
case <-time.After(2 * time.Second):
t.Fatal("feedRouterFromListener did not exit on gVisor 'endpoint is in invalid state' — accept loop is spinning")
}
}
// TestFeedRouterFromListener_BacksOffOnTransientError asserts the
// defence-in-depth path: an unknown sticky Accept error must NOT cause
// CPU spin. The loop backs off and exits cleanly when ctx is cancelled.
func TestFeedRouterFromListener_BacksOffOnTransientError(t *testing.T) {
logger := log.StandardLogger()
addr := &net.TCPAddr{IP: net.IPv4(127, 0, 0, 1), Port: 80}
router := nbtcp.NewRouter(logger, nil, addr)
const transientCount = 5
errs := make([]error, transientCount)
for i := range errs {
errs[i] = errSentinel("transient: temporary network error")
}
ln := newScriptedAcceptListener(errs...)
defer ln.Close()
ctx, cancel := context.WithCancel(context.Background())
start := time.Now()
done := make(chan struct{})
go func() {
defer close(done)
feedRouterFromListener(ctx, ln, router, logger, "acct-1")
}()
time.AfterFunc(150*time.Millisecond, cancel)
select {
case <-done:
// Expected.
case <-time.After(2 * time.Second):
t.Fatal("feedRouterFromListener did not exit on ctx cancellation — backoff or exit path broken")
}
// Without backoff the 5 scripted errors would burn in microseconds.
// With backoff the first delay alone is 5ms, so the loop must take
// at least that long even though ctx fires at 150ms.
elapsed := time.Since(start)
assert.GreaterOrEqual(t, elapsed, 5*time.Millisecond,
"loop ran without backing off — would burn CPU in production")
}