[teamleaderleo]

HOLD: Final rebased-head review and paired soak are pending (#27048 has landed).

A peer restart currently marks every session unreachable, including closed chats, then announces the entire list again. This change gives row updates a 10-second disconnect grace and keeps closed rows closed. Reconnects announce only rows whose contents changed. Forwarded requests still fail immediately when the link is down.

Leo: "I do a hell of a lot of routing to remote machines, right? ... Are we covering the edge cases for when, you know, something isn't going right, if there's recovery, whatever, stuff like that."

A single notice loop owns the grace deadlines. A fresh list or ready notice cancels its peer's deadline; repeated failures do not extend it; generation checks reject deadlines for replaced peers. A dedicated persistent stop signal cancels the loop at daemon shutdown without consuming the main daemon's notification. Logging, purge fencing, and transcript persistence still happen at the real disconnect. No daemon update timing, forwarding refusal, or list_guard changes.

Reconnect still saves the cache because a preceding session notice may have changed a row while its save was throttled. Already-stale and closed rows avoid redundant row writes on subsequent failures.

Validation

• Baseline 61028a9: 40770951012, job 123107681551, compiled and executed three regressions; all failed on the expected row/event assertions.
• Fixed core at 54b8597: 40771795001 passed ten tests across grace, cache, purge fencing, and stalled-peer integration.
• Independent Astra high/fast review found a shutdown-notification race. The dedicated stop signal and its regression fixed it; re-review SHIP covers aed5f94. Head 18f5813 differed only by rustfmt module order and had exact-head SHIP confirmation. Current head 372a60f rebases this change onto perf(acpmux): stream session summaries into list snapshots #27048 without conflicts; fresh independent review and remote paired checks are running.
• Final-head focused checks and soak 40772511771 stopped before compiling/measuring: fetching the baseline recursively fetched an unrelated submodule with a missing object. The prepared runner now disables recursive submodule fetch and restores Cargo.lock normalization. Final shutdown-regression execution and the one-hour rerun remain outstanding. No local builds or test suites ran.

Performance

Metric	Before	After	Delta	Harness
Row watcher events per short flap	Not measured	Not measured	—	peer_flap_soak, 30 fake peers × 20 rows
CPU seconds over 1 h	Not measured	Not measured	—	same workload, 17-minute flap interval
Sampled maximum RSS over 1 h	Not measured	Not measured	—	same workload

The Linux-only ignored harness measures the daemon Hub in a test process with fake notices and a memory store. It excludes SSH, app rendering, and disk-cache I/O. The prepared paired runner builds both versions with the identical harness, then runs isolated processes on the same Linux worker.

Remaining integration and landing block

This draft overlaps #27002 in disconnect cache handling. When integrating, preserve that PR's first/hourly outage-save condition and clearing of saved outage state on Connected.

The broker refused a normal merge of companion #26998 because its credential is limited to publishing. The relay also could not reach the authorized landing owner. No merge override was attempted. This PR remains draft pending the final checks, measurements, and an authorized landing path.

Agent session: agent session

[teamleaderleo]

Bead cx-v6macs created for this PR by the cmux coordinator.

[github-actions]

cmux status _{(one comment, edited in place; each line is updated by the job that owns it)}

red: cmux-next checks (god files, concurrency, crash safety, l10n); no finished base run on 538ccd0 to compare; do not merge until green; see lane-rules
red: cmux-next crash ratchet; NEW: green on the base 538ccd0; do not merge until green; see lane-rules

[teamleaderleo]

Owned by an agent session · took over from an earlier session · session

[teamleaderleo]

Carried onto current feat-cmux-next as #27809 (this PR's eight peer-flap commits with their authorship, plus one commit keeping #27214's planned-restart notice settling in the moved notice loop). The paired soak and review continue there. Leaving this PR open for its owner.

[blacksmith-sh]

Found 8 test failures on Blacksmith runners:

Failures

Test	View Logs
--from must install a complete prebuilt tree/ --from must install a complete prebuilt tree	View Logs
/tmp/tmpjxzt3nwa/summary-fail.json (live):/summary-fail.json (live):	View Logs
/tmp/tmpnwehoany/summary-missing.json (live):/summary-missing.json (live):	View Logs
[new_workspaces] debug.new_workspace timed 0 of 50 new workspace rows/ [new_workspaces] debug.new_workspace timed 0 of 50 new workspace rows	View Logs
[new_workspaces] the window draws no frames/[new_workspaces] the window draws no frames	View Logs
[tabs] debug.tab_switch timed 0 of 40 switches/ [tabs] debug.tab_switch timed 0 of 40 switches	View Logs
[tabs] the window draws no frames/[tabs] the window draws no frames	View Logs
test_every_listed_hook_is_one_glaeda_released_with_the_check (main.SigningHooksInSt epWithGlaedaTests.test_every_listed_hook_is_one_glaeda_released_with_the_check)/ test_every_listed_hook_is_one_glaeda_released_with_the_check (main.SigningHooksInSt epWithGlaedaTests.test_every_listed_hook_is_one_glaeda_released_with_the_check)	View Logs

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[teamleaderleo]

Superseded by #27809, which carries its eight peer-flap commits with their authorship. The soak and review continue there (comment at 01:29Z).