📝 docs+config: ecosystem alignment — integration roadmap, branch-protection-as-code, README cross-links#9
Conversation
…ks; drop misleading Snyk placeholder
…ture-request link
scttbnsn
requested review from
ALARGECOMPANY and
biggest-littlest
as code owners
… parse inside the centered footer div
…Claude/OpenAI/Cursor) + trimmed Built With
…I/Cursor under Built With instead
…cket (embedded SVGs; both dropped from simple-icons)
…t logo in footer (drydock parity)
…m-wide overlap color)
…payload - Thread *config.Config through NewSSEBroadcaster and NewAdapter so the ack payload can read LogLevel and DDPollInterval at runtime - MemoryGB: parse /proc/meminfo (MemTotal) on Linux via parseProcMeminfo; return 0.0 on all other GOOS — no cgo, no extra deps - ackDataBody.MemoryGB promoted from int to float64 (one-decimal GiB) - Wire fields: logLevel (string), pollInterval (string), memoryGb (float64)
Replace the blocking retry loop in HandleInput (up to 500 ms on readPump) with a per-ExecSession buffered channel drained by a single dedicated goroutine. - readPump now NEVER blocks on exec stdin I/O; it enqueues via a non-blocking channel select and returns immediately. - Stdin ordering within a session is strictly preserved: drainInput is the only writer on net.Conn, so concurrent HandleInput callers cannot reorder or interleave bytes. - Backpressure policy (documented in execInputQueueDepth comment): when the queue is full (drain cannot keep up), the incoming frame is dropped and an error frame is sent to the controller. This is "drop-with-error-frame": safer than blocking readPump (which would freeze pings, requests, and all other session I/O) and safer than silent corruption. - drainInput exits on session.Close() (done channel); the buffered channel is fully drained before exit so no enqueued bytes are silently discarded. - No goroutine leaks: Close() closes done, drainInput selects on it.
Stand up an in-process mock drydock WebSocket server (httptest.Server +
gorilla/websocket) that speaks the real lookout/1.0 frame format from
internal/protocol. Covers:
- hello/welcome handshake: happy path (validates protocol, agentId, capabilities)
and rejected-hello path (server returns error frame → connect() returns error)
- request fan-out: maxStreams semaphore saturation immediately returns an error
frame instead of blocking readPump
- exec session concurrency cap: StartExec beyond maxExecSessions sends exec_end
rejection frame
- exec session lifecycle: readLoop emits exec_output frames and exec_end on EOF
- #30 regression — ordered input queue:
TestExecInputQueue_OrderPreserved: N concurrent HandleInput callers; all N
byte values arrive at the Docker-side sink exactly once, none interleaved
TestExecInputQueue_NoBlockOnFull: HandleInput returns in <200 ms even when
queue is saturated (enforces readPump non-blocking requirement)
TestExecInputQueue_DrainOnClose: drainInput goroutine exits promptly on
session.Close() - no goroutine leak
TestExecClose_Idempotent: session.Close() is safe to call multiple times
- ping round-trip: readPump replies TypePing with TypePong and correct timestamp
- TestPing_UnderExecLoad: pong arrives within 500 ms while 200 exec_input frames
are in flight, proving readPump is non-blocking with the ordered queue fix
All 11 tests pass under -race.
HandleResize ran a 10×50 ms blocking retry loop directly on readPump, the same anti-pattern that issue #30 fixed for HandleInput. During a 10-attempt resize burst, readPump could not service pings, exec_input frames, or any other message type for up to 500 ms. Dispatch HandleResize with `go` so readPump returns immediately, exactly as HandleInput was fixed in #30. The retry loop in HandleResize itself is unchanged; resize failures are non-fatal and isolated to their own goroutine.
When the WebSocket tunnel dropped, connect() called wg.Wait() and then closed the WebSocket conn, but never iterated execSessions. Each session's readLoop and drainInput goroutines remained alive, blocked on s.conn.Read / s.conn.Write against the Docker hijacked net.Conn. The Docker exec process kept running and the goroutines were orphaned for the duration of the reconnect window — an unbounded leak on every unclean disconnect. After wg.Wait(), range execSessions and call Close() on every surviving session. The sync.Once in Close() ensures idempotency when readLoop also calls it.
drainInput previously spun indefinitely when conn.Write failed, relying on readLoop to eventually call Close() via a read error. For asymmetric half-closed connections (e.g. a proxy that closes only the write path), conn.Read may never error while conn.Write keeps failing — drainInput would spin forever and the goroutine would never exit. Add a consecutive write-error counter. After maxWriteErrors (3) back- to-back failures, drainInput calls s.Close() directly and returns. The sync.Once in Close() ensures idempotency with any concurrent readLoop Close() call.
- Clarify TestExecInputQueue_OrderPreserved: it does NOT catch the readPump-blocking bug (HandleInput is called directly, bypassing readPump). Note that TestPing_UnderExecLoad is the primary regression test for that pattern. - Label TestPing_UnderExecLoad as the PRIMARY regression test for the readPump-blocking anti-pattern from #30 and explain why it is the only test that exercises the full readPump dispatch path. - Add TestHandleResize_NonBlocking: asserts that a burst of exec_resize frames for an unknown session does not stall a concurrent ping for more than 500 ms (regression test for the HandleResize fix). - Add TestExecSessions_TornDownOnDisconnect: injects two fake sessions, drops the server-side WS conn, and verifies both sessions' Docker net.Conns are closed after connect() returns (regression test for the exec-session teardown fix). - Add TestDrainInput_ExitsOnWriteErrors: closes the exec conn and feeds frames into inputQ; asserts drainInput exits within 2 s and that session.done is closed (regression test for the write-error exit fix).
- close dial response body (bodyclose) - check c.connect/conn.Read returns in goroutines (errcheck) - remove unused newTestClientWithWS helper (unused)
✨ feat(adapter): populate logLevel/pollInterval/memoryGb in dd:ack (M4)
🐛 fix(edge): per-session ordered exec input queue + edge test harness (#30/#38)
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
1 participant
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.
Ecosystem-alignment batch for lookout (v0.2.2 line). Derived from two background investigation workflows over lookout · drydock · sockguard (integration analysis + presentation/governance normalization), with every finding adversarially verified before it landed here.
Changes
ROADMAP.md— adds a The ecosystem section (sockguard → lookout → drydock topology, standard vs edge mode) and folds the verified integration gaps into Now/Next/Later. The headline: edge mode is fully built agent-side but blocked on drydock's missing/api/lookout/wscontroller endpoint.scripts/apply-branch-protection.sh— codifies the OpenSSF-top-tier ruleset now active onmain(idempotent create/update), so branch protection is reproducible-as-code.README.md— adds the Part of the CodesWhat ecosystem cross-link block to the footer (drydock + sockguard), adds the missing Discussions badge, and removes the misleading Snyk placeholder (it rendered as "monitored" but Snyk was never wired up)..github/ISSUE_TEMPLATE/config.yml— routes Question/Help to Discussions Q&A and adds a Feature-request link, matching the drydock/sockguard convention.No code or behavior changes — docs, governance, and CI/issue config only. The
.roadmap/internal.mdpunch-list (gitignored) captures the full M1–M7 integration sequence and risks.Sibling CODEOWNERS normalization is in separate PRs (drydock #427, sockguard #97); their README/community normalization will follow in their own PRs.