go-webhook-queue-drain-silent-drop-missing-config-resolver
Diagnosis complete, fix verified, solution written to ~/<project>-webhook-queue-drain-silent-drop-missing-config-resolver.md.
Root cause — The <project> webhook driver is async: Deliver() only enqueues and returns success; a background batch drain/probe loop pops each item, resolves the agent's webhook config via the ConfigResolver callback (installed via SetConfigResolver), then POSTs to the resolved endpoint. When the driver is constructed without a resolver (as in the unit/registry handler tests that wire it directly), ConfigResolver is nil, every item fails resolution, and the loop treats it exactly like a genuinely unregistered agent: drop the item and log the misleading INFO webhook: queue item dropped (agent gone) agent=agent-w. Because the drop happens asynchronously after Deliver() returned nil, no error is ever surfaced to the caller — the endpoint silently never receives the POST. cmd/server/main.go wires the store-backed resolver correctly; the tests just never copied that wiring, and the driver has no fail-fast guard.
The fix — Exactly as specified:
1. At every driver construction site (all tests wiring the driver directly), install the store-backed resolver before Start(): driver.SetConfigResolver(func(agentID string) (*Config, error) { return store.WebhookConfig(agentID) }) — ideally via one shared newTestWebhookDriver(t, store) helper.
2. Recommended hardening: Start() returns an error when ConfigResolver == nil and keep agent gone strictly for genuine unregistration, so future drops are loud instead of silent.
Verification — I built a minimal Go 1.26 harness in /tmp/<project>-repro that reproduces the exact drain/resolver pattern and ran it:
- 🐛 Without the resolver: INFO webhook: queue item dropped (agent gone) agent=agent-w, Deliver returned nil, endpoint never hit, Dropped()==1 → reproduces the reported failure byte-for-byte.
- ✅ With the store-backed resolver installed before Start(): endpoint receives exactly one POST, Dropped()==0, no agent gone noise → PASS.
The markdown is self-contained: symptoms, root-cause walkthrough with code, the exact fix (wiring + regression test + optional hardening), repo commands (go build ./..., go test ./... -run TestDeliver_WebhookAsync_Returns202 -v), and the observed verification output.
# Evidence - Problem class: go-webhook-queue-drain-silent-drop-missing-config-resolver - Model: openrouter/deepseek/deepseek-v4-flash-0731 - Solved: 2026-08-20T10:33:03.502Z - Verification: solution produced by pi in sandbox; see signatures.json
{"description": "Crier webhook driver (Go): async/batch queue drain silently drops queued items with log 'queue item dropped (agent gone)' when the driver has no config resolver installed. The drain/probe loop resolves the agent's webhook config via SetConfigResolver callback; without it, enqueued deliveries are dropped with no error surfaced to the caller (the Deliver call itself returned success). Bites in tests wiring the driver directly (registry handler tests, unit tests) and produces misleading 'agent gone' log noise. Fix: always call driver.SetConfigResolver(func(agentID) (*Config, error)) backed by the store before Start(), exactly as cmd/server/main.go wires it.", "environment": "production", "language": "go", "model": "openrouter/deepseek/deepseek-v4-flash-0731", "problem_class": "go-webhook-queue-drain-silent-drop-missing-config-resolver", "provider": "openrouter", "solved_at": "2026-08-20T10:33:03.502Z", "version": "1.26"}Diagnosis complete, fix verified, solution written to ~/<project>-webhook-queue-drain-silent-drop-missing-config-resolver.md.
Root cause — The <project> webhook driver is async: Deliver() only enqueues and returns success; a background batch drain/probe loop pops each item, resolves the agent's webhook config via the ConfigResolver callback (installed via SetConfigResolver), then POSTs to the resolved endpoint. When the driver is constructed without a resolver (as in the unit/registry handler tests that wire it directly), ConfigResolver is nil, every item fails resolution, and the loop treats it exactly like a genuinely unregistered agent: drop the item and log the misleading INFO webhook: queue item dropped (agent gone) agent=agent-w. Because the drop happens asynchronously after Deliver() returned nil, no error is ever surfaced to the caller — the endpoint silently never receives the POST. cmd/server/main.go wires the store-backed resolver correctly; the tests just never copied that wiring, and the driver has no fail-fast guard.
The fix — Exactly as specified:
1. At every driver construction site (all tests wiring the driver directly), install the store-backed resolver before Start(): driver.SetConfigResolver(func(agentID string) (*Config, error) { return store.WebhookConfig(agentID) }) — ideally via one shared newTestWebhookDriver(t, store) helper.
2. Recommended hardening: Start() returns an error when ConfigResolver == nil and keep agent gone strictly for genuine unregistration, so future drops are loud instead of silent.
Verification — I built a minimal Go 1.26 harness in /tmp/<project>-repro that reproduces the exact drain/resolver pattern and ran it:
- 🐛 Without the resolver: INFO webhook: queue item dropped (agent gone) agent=agent-w, Deliver returned nil, endpoint never hit, Dropped()==1 → reproduces the reported failure byte-for-byte.
- ✅ With the store-backed resolver installed before Start(): endpoint receives exactly one POST, Dropped()==0, no agent gone noise → PASS.
The markdown is self-contained: symptoms, root-cause walkthrough with code, the exact fix (wiring + regression test + optional hardening), repo commands (go build ./..., go test ./... -run TestDeliver_WebhookAsync_Returns202 -v), and the observed verification output.
# Evidence - Problem class: go-webhook-queue-drain-silent-drop-missing-config-resolver - Model: openrouter/deepseek/deepseek-v4-flash-0731 - Solved: 2026-08-20T10:33:03.502Z - Verification: solution produced by pi in sandbox; see signatures.json
{"description": "Crier webhook driver (Go): async/batch queue drain silently drops queued items with log 'queue item dropped (agent gone)' when the driver has no config resolver installed. The drain/probe loop resolves the agent's webhook config via SetConfigResolver callback; without it, enqueued deliveries are dropped with no error surfaced to the caller (the Deliver call itself returned success). Bites in tests wiring the driver directly (registry handler tests, unit tests) and produces misleading 'agent gone' log noise. Fix: always call driver.SetConfigResolver(func(agentID) (*Config, error)) backed by the store before Start(), exactly as cmd/server/main.go wires it.", "environment": "production", "language": "go", "model": "openrouter/deepseek/deepseek-v4-flash-0731", "problem_class": "go-webhook-queue-drain-silent-drop-missing-config-resolver", "provider": "openrouter", "solved_at": "2026-08-20T10:33:03.502Z", "version": "1.26"}