Repository navigation
fix(a2a_redis_bridge): resolve lint error in internal/mcp/a2a_redis_bridge.go - #1
Closed
gitwire-hq[bot] wants to merge 41 commits into
Closed
gitwire-hq[bot] wants to merge 41 commits into
gitwire-hq[bot] wants to merge 41 commits into
Conversation
…, 08, 09, 10, 11, 12, 13) v1.0 features: - Unified governance pipeline (ExecuteGoverned) - Provider adapters: OpenAI, Anthropic, Gemini, Azure, Bedrock, Vertex, Groq, Cohere, Mistral - MCP client + server mode + A2A protocol - Admin API with RBAC, OIDC, audit logging - Exact + semantic caching - Guardrails pipeline with webhooks - KMS integration (static, AWS, Vault) v1.1 additions: - BATCH-06: Multi-instance A2A SSE via Redis Pub/Sub - BATCH-08: Config hot-reload with file watcher and SIGHUP - BATCH-09: Admin Dashboard SPA (embed.FS, dark theme) - BATCH-10: Prompt template CRUD management - BATCH-11: Webhook mTLS client certificate support - BATCH-12: Admin health endpoints for providers and models - BATCH-13: Redis Cluster support with UniversalClient Test evidence: 30 packages pass, 0 failures
- Move 4 hanging embeddings tests to integration build tag - Archive AIV documents for BATCH-06, 08, 09, 10, 11, 12, 13, RP1 - Version bumped to v1.1.0
Root cause: bytesReader.Read() had a value receiver, so pos never
advanced when used as io.Reader. io.ReadAll would loop forever reading
the same bytes.
Fix: change Read to pointer receiver, use &bytesReader{} everywhere.
Also merge embeddings_integration_test.go back into embeddings_test.go
since the tests no longer hang.
All 5 embeddings tests now pass in the default test run.
30 packages pass, 0 failures.
TASK-01: ROADMAP.md updated — v1.0 and v1.1 marked complete, stats updated (164 files, 379 tests, 30 packages, 10 providers). Shipped batch descriptions collapsed to single-line summaries. All 9 tech debt items marked resolved. TASK-02: README.md updated — provider list expanded to 10, added dashboard, hot-reload, prompt management, enterprise security features. Architecture diagram updated. Crossed off resolved limitations (#9 A2A SSE, #12 Redis Cluster). TASK-03: .github/workflows/release.yml — multi-arch binary builds (linux/darwin/windows), SHA256 checksums, GitHub Release with auto-generated notes, GHCR Docker push with semver tags. TASK-04: Dockerfile updated with VERSION build-arg and ldflags. docker-compose.yml: postgres upgraded to pgvector/pgvector:pg16, added redis-cluster profile, healthchecks for redis. TASK-05: Makefile — added 'make release' with ldflags, 'make docker-push' for multi-arch GHCR, 'make test-integration', 'make clean'.
Full codebase audit covering: - 164 Go files, 33,862 LOC, 378 tests - Package-by-package analysis of all 29 internal packages - Two critical bugs identified (unwired providers, cluster mode) - 6 design issues, 5 test gaps, 5 documentation gaps - Security assessment, performance characteristics - Prioritized recommendations for v1.1.1, v1.2.0, v2.0
Plans v1.1.1 hotfix through v2.0 platform batches: - BATCH-FX1: Fix unwired providers (BUG-01) + A2A cluster mode (BUG-02) - BATCH-DQ1: Server.go refactor, missing tests, docs, embeddings governance - v1.2.0: A2A multi-turn, TS/Python SDKs, plugin interface - v2.0: gRPC gateway, WASM adapters, multi-region federation
BUG-01: Add bedrock, vertex, groq, cohere, mistral to server.go adapter switch and config validation. Add project/region/publisher config fields for vertex. Default base URLs for groq/cohere/mistral. BUG-02: Change a2a_redis_bridge from *goredis.Client to UniversalClient so the bridge works with both standalone and cluster Redis. Pass redisClient.Universal() instead of .Standalone() in server.go. Tests: 30 packages pass, 0 failures, 0 vet warnings.
BATCH-DQ1: - Refactor server.go: extract buildAdapters, buildMCP, buildTracer, buildGuardrails, buildA2AExecutor, buildA2AStore builder functions - Add tests for tracing (2), usage (2), lifecycle (2) packages - Add rate limiting + budget checks to embeddings handler - Create migration guide v1.0→v1.1 - Update docs/index.md with new migration link - Remove dead code: unused recordUsage() method - 32 packages pass, 0 failures
BATCH-07 + BATCH-DQ1 + BATCH-FX1 combined: - A2A multi-turn: message/send accepts taskId to continue existing task - Agent responses appended to task history for conversation continuity - Rate limiting + budget checks for embeddings handler - Refactor server.go: 6 builder functions extracted (470→331 lines) - Usage Writer nil-DB safety fix - Remove dead code: unused recordUsage() method - Tests: tracing (2), usage (2), lifecycle (2), multi-turn (4) - Migration guide v1.0→v1.1, docs/index.md updated - 33 packages pass, 0 failures
BATCH-14: - @openlimit/sdk TypeScript package in sdks/typescript/ - OpenLimitClient: chat completions (stream + non-stream), embeddings, models, health - SSE streaming parser via async generators - Full TypeScript types for all request/response shapes - APIError + TimeoutError error classes - 7 unit tests (vitest): all passing - Zero runtime dependencies (uses native fetch) - Builds to ESM + CJS via tsup - README with complete usage examples
BATCH-15: - openlimit Python package in sdks/python/ - OpenLimitClient: chat (stream + non-stream), embeddings, models, health - SSE streaming parser via generators - Full dataclass types for all request/response shapes - APIError + TimeoutError exception classes - 6 unit tests (pytest): all passing - Zero third-party runtime dependencies (stdlib http.client) - pyproject.toml, README with usage examples
BATCH-16: - Plugin interfaces: GuardrailPlugin, MiddlewarePlugin, ProviderPlugin - Global PluginRegistry with Register/Lookup/List/ListByType - GuardrailContext for data passing through pipeline - HeaderInjector example middleware plugin - Plugin config section in gateway.yaml - Guardrail pipeline: 'plugin' stage type loads from registry - pluginStageAdapter bridges GuardrailPlugin ↔ guardrails.Stage - 7 plugin tests + guardrail factory test
BATCH-17: - A2APart extended: FilePart (fileUri, mimeType, base64 bytes) - A2APart extended: DataPart (structured JSON key-value data) - extractTextFromHistory handles all part types: text → plain text, file → [file: mime uri], data → JSON - File and data parts round-trip through task history - 5 new tests (FilePart, DataPart, extract helpers)
BATCH-18: - MultiProvider: validates tokens against multiple OIDC issuers - Issuer-based routing: extracts iss claim from JWT to select provider - Pool: convenience wrapper for single/multi-provider modes - OIDCProviderConfig in gateway.yaml for multiple providers - extractIssuer: base64url + JSON decode (no external deps) - 6 new tests: issuer extraction, normalization, pool lifecycle
All v1.2 batches complete: - BATCH-07: A2A Multi-Turn - BATCH-14: TypeScript SDK - BATCH-15: Python SDK - BATCH-16: Plugin Interface - BATCH-17: A2A File/Data Parts - BATCH-18: Multi-Tenant OIDC
Add config toggle that captures full chat completion request messages and response content in the audit_logs table via JSONB metadata. Off by default for privacy. - New audit event types: chat.completion, embedding - auditLog + logBodies wired into openaiapi.Handler - SetAuditLogger/SetLogBodies setter methods - Example config with documentation - 34 packages pass, 0 failures
BATCH-57 — Fix all 5 Critical findings from Stress Test QA Audit: TASK-01: JSON injection in guardrail redaction - Replace string concat into json.RawMessage with json.Marshal() - 4 regression tests for quotes, newlines, backslashes, empty content TASK-02: Double mutex unlock in MCP tryReconnect - Remove defer state.mu.Unlock(), keep explicit unlock only - 4 tests: success, failure, old client close, concurrent no deadlock TASK-03: Expired key auth bypass + readJSON body race - Add expires_at SQL filter to LookupVirtualKeyByToken and ListVirtualKeysForMCP - Simplify quickstart error handling — always 400 on JSON decode error - 6 tests: expired key rejected, MCP excludes expired, malformed JSON 400 TASK-04: Nil LatencyCache for smart routing strategy - Create LatencyCache when strategy is smart OR latency - 4 tests: non-nil cache, different scores, priority no cache, nil reader safe Test results: 677 pass (+18 new), 8 fail (OBL-05 pre-existing), 0 regressions Lint: go vet + go build clean Co-Authored-By: Craft Agent <agents-noreply@craft.do>
…ker, goroutine lifecycle, LRU eviction BATCH-58: Eliminate all race conditions and unprotected goroutine launches. TASK-58-01: OnKeysChanged Panic Guard + Circuit Breaker Atomic Redis Ops - Wrap all 5 go h.OnKeysChanged() calls in keys.go with defer recover() - Replace HGETALL+HSET in breaker.redisRecordFailure with atomic HINCRBY - Add HIncrBy to redis client interface and method - Add 4 tests: panic recovery, panic logged, concurrent atomic increment, threshold exactness TASK-58-02: MCP Goroutine Leak Fix + Notification Lifecycle - Add cancelNotif context.CancelFunc to serverState struct - Cancel old notification listener before launching new one on reconnect - Store cancelNotif per server in Start() with child contexts - Add 3 tests: reconnect spawns 1 goroutine, cancel kills old listener, 10 reconnects no leak TASK-58-03: KeyCache LRU Eviction - Add lastAccess time.Time to KeyCacheEntry - Change Get() from RLock to Lock; update lastAccess on every hit - Replace random map eviction in Set() with oldest-lastAccess eviction - Add 3 tests: MRU survives eviction, oldest evicted first, concurrent Get+Set no race Co-Authored-By: Craft Agent <agents-noreply@craft.do>
…text-aware backoff, default timeout, usage drop metric, ring buffer Co-Authored-By: Craft Agent <agents-noreply@craft.do>
…, breaker LRU eviction TASK-60-01: Add output guardrail check on accumulated streaming content before sending [DONE]. Block sends SSE error event. Redact is audit-only. TASK-60-02: Add 3 smart routing integration tests with mock LatencyReader and HealthChecker. Verifies strict ordering, cost-only fallback, and health score differential. TASK-60-03: Replace random breaker map eviction with LRU tracking. Add breakerEntry wrapper with lastAccess. Evict oldest entry on cap. BAC-01: 10/10 new tests pass BAC-02: 0 regressions (707 pass, 8 OBL-05 pre-existing) BAC-03: go vet + go build = zero warnings/errors Co-Authored-By: Craft Agent <agents-noreply@craft.do>
… parsing, quickstart guard
BATCH-61: Data Integrity
TASK-61-01: SQL-Based Pagination + Budget Context Timeout
- Add offset/limit params to ListVirtualKeys with SQL LIMIT/OFFSET
- Add CountVirtualKeys using SELECT COUNT(*) for X-Total-Count header
- Update listKeys handler to use SQL pagination (remove Go slice pagination)
- Add ctx context.Context param to CheckBudget, pass to GetSpendForCurrentPeriod
- Update all 5 callers (middleware, governed, embeddings, discovery, usage)
- Update 4 existing budget tests with context.Background()
TASK-61-02: parseArrayString Quoted Commas + arrayString Escaping
- Rewrite parseArrayString with state-machine parser for quoted elements
- Handle PostgreSQL quoted commas: {a,"b,c",d} → 3 elements
- Handle escaped quotes (\" → ")
- Rewrite arrayString to quote items containing {, }, or ,
- Escape internal double-quotes as \""
- Handle nil slice → "{}"
TASK-61-03: Quickstart Duplicate Project Guard
- Add GetProjectByName to store/projects.go
- Update handleQuickstart to check for existing project before creating
- Reuse existing quickstart-YYYY-MM-DD project if found
10 new tests, 0 regressions. go vet + go build clean.
Co-Authored-By: Craft Agent <agents-noreply@craft.do>
…ion, guardrail prefill, ARIA landmarks BATCH-62: UX Auth + Navigation - TASK-62-01: Replace prompt() with inline login form (div#login-form, password input, submit button, Enter-key, 401 handling) - TASK-62-02: Consolidate 8 tabs to 5 (Overview, Keys, Usage, Guardrails, MCP); merge Spend/Request Log into Usage, Providers into Overview; add guardrail catalog onclick prefill - TASK-62-03: ARIA landmarks (role=tab/tabpanel, aria-selected, aria-controls, aria-labelledby), arrow key navigation, focus management 8 new tests, 0 regressions, go vet + go build clean. Co-Authored-By: Craft Agent <agents-noreply@craft.do>
…points, focus-visible, colorblind bars BATCH-63: UX Onboarding + Docs TASK-63-01: Add Quick Setup (Recommended) section to getting-started.md - Document openlimit init as one-liner setup command - Existing manual steps preserved as Advanced Setup TASK-63-02: Mobile CSS breakpoints + focus-visible accessibility - @media (max-width: 1024px) reduced padding/font sizes for tablets - @media (max-width: 768px) single-column, vertical tabs, full-width cards - :focus-visible rules for buttons, inputs, tabs with outline accent TASK-63-03: Colorblind-friendly budget bars - .budget-bar-safe (solid green), .budget-bar-warning (yellow+stripes), .budget-bar-danger (red+stripes) with repeating-linear-gradient patterns - Updated computeStatusJS thresholds from 75/95 to 70/90 - Uses CSS custom properties (--green, --yellow, --red) 9 new tests (all passing), 0 regressions, go vet + go build clean. Co-Authored-By: Craft Agent <agents-noreply@craft.do>
…ease certificate BATCH-64 — Release Hardening v1.4.1 GA 7 remediation batches (BATCH-57→63): - 5 Critical security fixes (JSON injection, double-unlock, expired-key bypass, body race, nil cache) - 4 Concurrency fixes (panic guard, atomic breaker, goroutine lifecycle, LRU eviction) - 5 Resource management fixes (admin body limits, context-aware backoff, default timeout, drop metric, ring buffer) - 3 Streaming + routing fixes (output guardrails, smart routing verification, breaker LRU) - 4 Data integrity fixes (SQL pagination, budget context, array parsing, quickstart guard) - 4 UX improvements (inline login, 5-tab consolidation, guardrail prefill, ARIA landmarks) - 4 UX polish (init wizard docs, mobile breakpoints, focus-visible, colorblind bars) Test results: 733 pass (+74 new), 8 fail (OBL-05 pre-existing), 0 regressions Lint: go vet + go build clean Version: v1.4.1 Co-Authored-By: Craft Agent <agents-noreply@craft.do>
…SizeKB in test config Root cause: config.Default() sets MaxBodySizeKB=0. The config loader fixes this to 10240 (10MB), but server tests bypass the loader by calling server.NewRuntime() directly. maxBodySizeMiddleware(0) wraps every request with http.MaxBytesReader(w, r.Body, 0), rejecting all request bodies with 400 'invalid JSON request body'. Fix: Add cfg.Server.MaxBodySizeKB = 10240 to baseConfig() in server_test.go. Result: 741/741 tests pass. Zero failures. OBL-05 CLOSED. Co-Authored-By: Craft Agent <agents-noreply@craft.do>
v1.4.2 GA Release All 741 tests pass. Zero failures. First clean-sheet release in project history. Changes from v1.4.1: - OBL-05 resolved: server integration tests fixed (MaxBodySizeKB in test config) - Version bumped to v1.4.2 - CHANGELOG updated Co-Authored-By: Craft Agent <agents-noreply@craft.do>
…ments
ERR-001: OBL-05 was misdiagnosed as port binding for 3+ weeks.
Actual cause: MaxBodySizeKB=0 in test config. Fixed in v1.4.2.
ERR-002: Test counts (700, 701) in pre-v1.4.2 docs were never
independently verified. Authoritative count: 741.
ERR-003: DG-2 'CONDITIONAL PASS' was overstated. No CI pipeline
exists. -race has never been run. Downgraded to NOT VERIFIED.
ERR-004: All batch certificates are self-signed by Lead (no independent
verifier). Acknowledged as inherent AIV single-agent limitation.
ERR-005: Spawned session audit outputs (667 repos, 50 findings, etc.)
were not independently verified by Lead.
ERR-006: Module map descriptions from BATCH-57→63 are from summaries,
not re-verified against actual source code.
Updated: STATE.md, CERT-V1.4.0, CERT-V1.4.1, CERT-V1.4.2, CERT-BATCH-64
Co-Authored-By: Craft Agent <agents-noreply@craft.do>
- setup-go@v6 supports Go 1.25 with direct-download fallback - Add 'master' to branch triggers so CI works pre-rename - No functional changes to test/build steps
Race conditions found by first -race run on Linux CI: - a2a_types.go: add sync.RWMutex to A2ATask with json:"-" tag and custom MarshalJSON that RLocks before serialization - a2a_handler.go: Lock around all task mutations in executeTask(), blocking mode, and cancel path - keys_panic_test.go: poll log buffer instead of immediate read to avoid race between goroutine slog write and test read Test fixes: - governed_test.go: insert maxBreakers-1 fillers instead of maxBreakers so only stale+mid are evicted (fresh survives unambiguously) - logger_test.go: match PG JSON output with spaces after colons CI improvements: - ci.yml: show gofmt offenders instead of silent exit - .gitattributes: force LF for .go/.yml/.sql/.json to prevent CRLF formatting failures on Linux runners - .gitignore: exclude events.jsonl 741/741 pass locally (Windows, no -race). CI run will verify -race.
… sync.RWMutex in struct li… Applied by GitWire AI self-healing CI. Changed bridgeMessage.Task type from A2ATask to *A2ATask and updated Publish/handleMessage accordingly to prevent copying the lock value during JSON marshaling.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
🔧 GitWire Auto-Heal PR
Workflow: CI
Branch: master
Failure type: lint_error
Confidence: high
Root Cause
Go vet detected copying of lock values due to sync.RWMutex in struct literal and json.Marshal calls
What Changed
Changed bridgeMessage.Task type from A2ATask to *A2ATask and updated Publish/handleMessage accordingly to prevent copying the lock value during JSON marshaling.
This PR was automatically generated by GitWire self-healing CI.
Review the changes carefully before merging.