fix(tui): desynchronize managed reconnect ensure herd by armancharan · Pull Request #37586 · anomalyco/opencode

armancharan · 2026-07-18T01:54:28Z

Issue for this PR

Related to #36285

Parent epic: #36441

Builds on ○ #37585 (stream ownership / post-ensure jitter) and ○ #37570 (ensure-on-first-failure). This PR is the reconnect-herd slice only — update/replacement UX and cold-location cost remain out of scope.

✧ raised · ○ in review · ● approved · ✓ merged

#36441 epic: scope streams + bound payloads

Merge in order:
v2
└ ○ #37486 location interest
   └ ○ #37487 session interest + fan-out bench
      └ ○ #37585 stream ownership / diagnostics
         └ ○ #37586 reconnect herd ← this PR

Siblings under the epic (not git parents of the stack tip):
└ ○ #37559 session blobs + payloadHash
└ ○ #37570 ensure-on-first-failure → feeds #37586

Stacked on #37585 (event-stream-ownership). GitHub base is v2 because the parent branch is fork-only; herd-only diff: armancharan/opencode@event-stream-ownership...reconnect-herd — retarget base onto that branch after #37585 merges.

Type of change

Bug fix
New feature
Refactor / code improvement
Documentation

What does this PR do?

N TUIs reconnecting after a managed-service restart were calling Service.ensure in lockstep (attempt 1 skipped backoff), launching many serve --service contenders together.

Pre-ensure spread: 0..500ms jitter before every service.reconnect() / ensure so callers desynchronize
Keep ensure-on-first-failure: after the spread, attempt 1 still ensures immediately and skips the post-ensure 1000+jitter backoff when ensure succeeds
Export ServerConnection.managedReconnect + regression tests (aligned with test(cli): lock managed reconnect ensure-on-first-failure #37570) plus a seven-caller concurrent reuse test that must not spawn

How did you verify your code works?

packages/cli test/server-connection.test.ts — ensure-on-first-failure; version reuse; concurrent reconnects reuse one endpoint without spawning
packages/tui test/cli/tui/reconnect-herd.test.ts — ensure spread / backoff bounds
packages/tui test/cli/tui/use-event.test.tsx — attempt-1 ensure still happens after spread; cleanup aborts in-flight ensure
typecheck in packages/cli and packages/tui

Acceptance → test

Claim	Test
Ensure spread desynchronizes reconnecting clients	`ensure spread stays within jitter bounds and desynchronizes samples`
Post-ensure backoff stays in `[1000, 1500)`	`post-ensure backoff stays in [1000, 1500)`
First reconnect still ensures (after spread)	`backs off when a resolved event stream keeps failing` + `managed reconnect ensures the service on the first failure`
N concurrent reconnects reuse healthy endpoint without spawn	`concurrent managed reconnects reuse one healthy endpoint without spawning`
Healthy mismatched version reused under reconnect	`managed reconnect reuses a healthy service from another version`

Screenshots / recordings

N/A — not a UI layout change.

Checklist

I have tested my changes locally
I have not included unrelated changes in this PR

Made with Cursor

Co-authored-by: Cursor <cursoragent@cursor.com>

Narrow unknown event data with object/`in` checks so sessionID is read without an assertion. Co-authored-by: Cursor <cursoragent@cursor.com>

Helpers only need directory; no test supplies a workspace. Co-authored-by: Cursor <cursoragent@cursor.com>

No test supplies a workspace; avoid an unnecessary brand cast. Co-authored-by: Cursor <cursoragent@cursor.com>

Filter interested subscribers once before encode, require directory for location scope (reject workspace-only queries), and cover those cases in tests. Co-authored-by: Cursor <cursoragent@cursor.com>

Collect matching subscribers first so encode/offer share one pass. Co-authored-by: Cursor <cursoragent@cursor.com>

Extract pure interest helpers and cover subscribeInput/equality plus ClientProvider.event.scope reconnecting only when interest changes. Co-authored-by: Cursor <cursoragent@cursor.com>

Serialize TUI subscribe generations, cancel SSE readers on abort, bridge Bun disconnects so EventFeed queues release, and expose content-free feed counters for leak detection. Co-authored-by: Cursor <cursoragent@cursor.com>

Avoid a hard EventFeed.Service requirement on the debug handler so CLI serve Effect inference stays compatible with the runtime handler map. Co-authored-by: Cursor <cursoragent@cursor.com>

Spread ensure calls with pre-ensure jitter, keep attempt-1 ensure-on-first- failure after the spread, and lock concurrent reconnect reuse without spawn. Co-authored-by: Cursor <cursoragent@cursor.com>

github-actions · 2026-07-18T01:54:38Z

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

Open an issue describing the bug/feature (if one doesn't exist)
Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

Replace the hand-rolled unknown guards with decodeUnknownOption so ephemeral sessionID extraction matches host Schema patterns. Co-authored-by: Cursor <cursoragent@cursor.com>

Map HttpApi-decoded ctx.query through interestFromSubscribeQuery, and route deepObject URLSearchParams through the same Schema before building Interest. Co-authored-by: Cursor <cursoragent@cursor.com>

Also derive EventFeed.Diagnostics from the protocol Schema and lock the generation boundary so prior-generation event IDs are not delivered. Co-authored-by: Cursor <cursoragent@cursor.com>

One test was asserting connect, interest change, and no-op scope together. Split into three named cases with a shared mount helper. Co-authored-by: Cursor <cursoragent@cursor.com>

Co-authored-by: Cursor <cursoragent@cursor.com>

…t herd

kitlangton · 2026-07-23T20:18:27Z

Closing this version because the reconnect-herd change is not validated against the missing-service case it is intended to improve. The current tests cover jitter bounds and concurrent reuse of an already-healthy endpoint, but do not measure contender count or recovery latency when several clients ensure a missing service. Recent lifecycle logs also show both single-contender recovery and a slower multi-contender case where a 0-500ms spread would not have covered the observed startup window. We should first add a focused concurrent-ensure benchmark/reproducer, then keep only a policy that measurably reduces launches without delaying normal recovery.

armancharan · 2026-07-23T21:37:11Z

sounds good, thanks for the feedback, @kitlangton!

what do you think about something like #38568, for starters, in that case?

armancharan and others added 15 commits July 17, 2026 21:58

feat(server): opt-in location interest for event subscriptions

3eac0eb

Co-authored-by: Cursor <cursoragent@cursor.com>

feat(server): narrow event subscriptions by session interest

8c92fcc

Co-authored-by: Cursor <cursoragent@cursor.com>

style(tui): alphabetize ClientProvider props

7bfcc92

Co-authored-by: Cursor <cursoragent@cursor.com>

style(tui): alphabetize ClientProvider props

7674f99

Co-authored-by: Cursor <cursoragent@cursor.com>

refactor(schema): drop cast in sessionIDOf

162a75f

Narrow unknown event data with object/`in` checks so sessionID is read without an assertion. Co-authored-by: Cursor <cursoragent@cursor.com>

test(server): drop unused workspaceID from event fixtures

6f87215

Helpers only need directory; no test supplies a workspace. Co-authored-by: Cursor <cursoragent@cursor.com>

test(server): drop unused workspaceID from event fixture

4b54724

No test supplies a workspace; avoid an unnecessary brand cast. Co-authored-by: Cursor <cursoragent@cursor.com>

merge: bring location-interest fixture cleanup into session stack

3c58e70

refactor(server): tighten event subscription interest matching

f637389

Filter interested subscribers once before encode, require directory for location scope (reject workspace-only queries), and cover those cases in tests. Co-authored-by: Cursor <cursoragent@cursor.com>

refactor(server): filter event-feed interest once before encode

51ff5f8

Collect matching subscribers first so encode/offer share one pass. Co-authored-by: Cursor <cursoragent@cursor.com>

merge: bring location-interest encode-once filter into session stack

9e50e73

test(tui): lock event interest query shape and scope reconnect

7444cfa

Extract pure interest helpers and cover subscribeInput/equality plus ClientProvider.event.scope reconnecting only when interest changes. Co-authored-by: Cursor <cursoragent@cursor.com>

fix(server): enforce event stream ownership and diagnostics

0d49a51

Serialize TUI subscribe generations, cancel SSE readers on abort, bridge Bun disconnects so EventFeed queues release, and expose content-free feed counters for leak detection. Co-authored-by: Cursor <cursoragent@cursor.com>

fix(server): read event-feed diagnostics via serviceOption

af08507

Avoid a hard EventFeed.Service requirement on the debug handler so CLI serve Effect inference stays compatible with the runtime handler map. Co-authored-by: Cursor <cursoragent@cursor.com>

fix(tui): desynchronize managed reconnect ensure herd

379be6f

Spread ensure calls with pre-ensure jitter, keep attempt-1 ensure-on-first- failure after the spread, and lock concurrent reconnect reuse without spawn. Co-authored-by: Cursor <cursoragent@cursor.com>

github-actions Bot added the needs:issue label Jul 18, 2026

refactor(schema): decode sessionIDOf data via Schema

d7325fd

Replace the hand-rolled unknown guards with decodeUnknownOption so ephemeral sessionID extraction matches host Schema patterns. Co-authored-by: Cursor <cursoragent@cursor.com>

armancharan marked this pull request as draft July 18, 2026 02:09

armancharan and others added 8 commits July 18, 2026 12:10

merge: bring sessionIDOf Schema decode into ownership stack

2173286

merge: bring sessionIDOf Schema decode into reconnect herd stack

8cefa2f

refactor(server): decode event interest via EventSubscribeQuery

7232360

Map HttpApi-decoded ctx.query through interestFromSubscribeQuery, and route deepObject URLSearchParams through the same Schema before building Interest. Co-authored-by: Cursor <cursoragent@cursor.com>

merge: bring EventSubscribeQuery interest decode into ownership stack

fabaeda

Also derive EventFeed.Diagnostics from the protocol Schema and lock the generation boundary so prior-generation event IDs are not delivered. Co-authored-by: Cursor <cursoragent@cursor.com>

merge: bring interest decode and diagnostics Schema into reconnect herd

3f48633

test(tui): split event.scope reconnect assertions

364c7a5

One test was asserting connect, interest change, and no-op scope together. Split into three named cases with a shared mount helper. Co-authored-by: Cursor <cursoragent@cursor.com>

merge: bring split event.scope tests into ownership stack

9deb064

Co-authored-by: Cursor <cursoragent@cursor.com>

merge: bring split event.scope tests into reconnect herd stack

d8a636b

armancharan marked this pull request as ready for review July 18, 2026 02:44

armancharan and others added 2 commits July 18, 2026 12:55

test(tui): share mount and stream tracker in ownership suite

1a2e1a7

Co-authored-by: Cursor <cursoragent@cursor.com>

merge: bring ownership test mount helpers into reconnect herd stack

7d53546

armancharan and others added 4 commits July 18, 2026 12:56

test(cli): share temp root helpers in managed reconnect suite

a602259

Co-authored-by: Cursor <cursoragent@cursor.com>

refactor(tui): clarify subscribeInput locals and drop test bangs

1963373

Co-authored-by: Cursor <cursoragent@cursor.com>

merge: bring subscribeInput clarity into ownership stack

49e7e02

merge: bring subscribeInput clarity into reconnect herd stack

1a13de7

armancharan and others added 4 commits July 18, 2026 14:00

test(server): prove Node HTTP abort releases EventFeed subscribers

b79e543

Co-authored-by: Cursor <cursoragent@cursor.com>

test(server): add non-gating EventFeed fan-out benchmark

85f9a1f

Co-authored-by: Cursor <cursoragent@cursor.com>

merge: bring EventFeed fan-out benchmark into ownership stack

ffbf5be

merge: bring fan-out benchmark and abort ownership test into reconnec…

6ba2ffa

…t herd

kitlangton added the 2.0 label Jul 19, 2026

kitlangton self-assigned this Jul 19, 2026

github-actions Bot mentioned this pull request Jul 19, 2026

📊 AI CLI 工具社区动态日报 2026-07-20 litang9/big_model_radar#235

Open

jaepo01 mentioned this pull request Jul 21, 2026

MCP transport stays 'failed' after a transient connection close — plugin tools disappear for the rest of the session #38088

Closed

kitlangton closed this Jul 23, 2026

armancharan mentioned this pull request Jul 23, 2026

test(client): add concurrent missing-service ensure benchmark #38568

Open

6 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(tui): desynchronize managed reconnect ensure herd#37586

fix(tui): desynchronize managed reconnect ensure herd#37586
armancharan wants to merge 34 commits into
anomalyco:v2from
armancharan:reconnect-herd

armancharan commented Jul 18, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jul 18, 2026

Uh oh!

kitlangton commented Jul 23, 2026

Uh oh!

armancharan commented Jul 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

armancharan commented Jul 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Issue for this PR

Type of change

What does this PR do?

How did you verify your code works?

Acceptance → test

Screenshots / recordings

Checklist

Uh oh!

github-actions Bot commented Jul 18, 2026

Uh oh!

kitlangton commented Jul 23, 2026

Uh oh!

armancharan commented Jul 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

armancharan commented Jul 18, 2026 •

edited

Loading