Relax timeouts on MOVING rebind without buffered commands

[ggivo]

Relax timeouts on MOVING rebind without buffered commands

Summary

MaintenanceAwareConnectionWatchdog only notified onRebindStarted when a command was in flight at the moment the MOVING push arrived. On an idle connection the relaxed-timeout window was never opened, so commands issued during the ~rebind window timed out with the normal (unrelaxed) timeout. This surfaced as the recurring RelaxedTimeoutConfigurationTest.timeoutUnrelaxedOnMovingTest scenario failure. Rebind-started is now reported for every MOVING notification, regardless of the command stack.

Description

• Commands issued after the MOVING notification are buffered until the new connection is up, so timeouts must be relaxed for the whole rebind window even when nothing was in flight when the push arrived.
• In the empty-stack path the channel is closed immediately and channelReadComplete() never fires again, so onRebindCompleted is reported inline there; the non-empty path keeps reporting completion from channelReadComplete() once the stack drains.

Changes

• onRebindStarted now fires on every MOVING rebind, not only when the command stack is non-empty.
• The empty-stack path now also reports onRebindCompleted, so relaxed timeouts are lifted after the grace period instead of never being opened/closed.

Testing

MaintenanceAwareConnectionWatchdogUnitTests updated: the empty-stack test previously asserted onRebindStarted was not called — that expectation encoded the bug and now asserts both started and completed notifications; the non-empty-stack tests additionally verify completion is deferred until the stack drains. The scenario test RelaxedTimeoutConfigurationTest#timeoutUnrelaxedOnMovingTest reproduces the bug end-to-end.

---

Note

Medium Risk
Touches maintenance/MOVING rebind lifecycle and command timeout behavior during cluster endpoint moves; scope is narrow but affects timing-sensitive client paths.

Overview
MOVING rebind handling in MaintenanceAwareConnectionWatchdog now always opens the relaxed-timeout window, including when the connection is idle and no commands were in flight when the push arrived.

notifyRebindStarted runs for every rebind (not only when the command stack is non-empty), so commands buffered while the endpoint is disconnected during rebind get the same relaxed timeouts as in-flight work. On the empty-stack path the channel closes immediately and channelReadComplete() never runs again, so onRebindCompleted is invoked there to end the relaxed window after the grace period; the non-empty path still completes via channelReadComplete() after the stack drains.

Unit tests were updated to match: the idle-connection case now expects both started and completed notifications, and non-empty-stack cases assert completion is not reported until later.

^{Reviewed by Cursor Bugbot for commit 8132ccc. Bugbot is set up for automated code reviews on this repo. Configure here.}

[ggivo]

run integration tests

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8132ccc3e9

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[uglide]

run scenario tests