Fix stale SSH PTY resize: retry pty_resize delivery with bounded backoff

[austinywang]

Fixes #6306.

Problem

In SSH workspaces, a terminal TUI pane could keep a stale resize state: after a pane-width change the remote TUI no longer reconciled to the visible size, and only Reconnect Workspace restored the self-healing behavior. cmux refresh-surfaces did not reliably repair it.

cmux ssh-pty-attach forwards size changes to the remote PTY via workspace.remote.pty_resize over a bounded 50 ms RPC (SSHPTYResizeMonitor). When that send failed — e.g. it raced a stale or blocked remote-session control path and the bounded RPC timed out — drainPendingResizes recorded the pending size, resumed input waiters, and returned without scheduling any further delivery. The size was only re-attempted on the next SIGWINCH or input edge, so a settled pane (a TUI the user is just viewing) could keep a stale size indefinitely. Output kept flowing, so the terminal looked alive, but the remote PTY/TUI never reconciled until a manual reconnect rebuilt the session path — exactly the report in the issue.

Fix

Make resize delivery self-healing in CLI/SSHPTYResizeMonitor.swift:

• On delivery failure, keep the latest desired size pending and schedule a backoff retry that re-drains it without waiting for an external event.
• Backoff starts at 100 ms and doubles to a 2 s cap; it resets on the first successful send.
• Input waiters still resume immediately on failure, so typing is never blocked behind a wedged resize (input-edge ordering stays bounded).
• The retry task is cancelled on teardown, so retries are bounded by the attach connection's lifetime — the output read loop already cancels the monitor on bridge EOF.

This matches the issue's "small first fix … add coalesced retry around the resize send." The heavier "trigger a remote-session reconnect / PTY reattach after repeated failures" path is intentionally out of scope — it's riskier (can disrupt the live session) and the bounded-backoff retry already lets the common transient-failure case converge on its own.

Test

Two-commit red/green. The regression test (added to the already-wired CLISSHPTYResizeInputTests suite) runs the real ssh-pty-attach CLI against a mock daemon + bridge. It opens the PTY at the target size so the attach-time reconcile is the single forced resize, makes the daemon reject the first pty_resize, and asserts the helper re-delivers the same size automatically with no further SIGWINCH or input edge.

• Commit 1 (test only) is red: without a self-driven retry the second delivery never arrives.
• Commit 2 (fix) is green.

Localization

No user-facing strings added or changed (CLI internal control path only); no Localizable.xcstrings / web message updates required.

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag /codesmith with what you need. Autofix is disabled.}

---

Summary by cubic

Adds a bounded backoff retry for SSH PTY resize delivery so TUI panes don’t stick at a stale size in SSH workspaces. Fixes #6306 and removes the need to reconnect after a resize RPC timeout.

• Bug Fixes

  • On failed workspace.remote.pty_resize, keep the latest size pending and retry with backoff (100ms doubling to 2s) via a single cancellable DispatchWorkItem timer; reset on success and cancel on teardown in CLI/SSHPTYResizeMonitor.swift.
  • Resume input waiters immediately on failure to keep typing responsive and preserve input-edge ordering.
  • Added a regression test in cmuxTests/CLISSHPTYResizeInputTests.swift that forces a failed first resize and asserts automatic re-delivery without a new SIGWINCH or input.

• CI

  • Regenerated .github/swift-file-length-budget.tsv for the larger test file and tightened a few budgets.

^{Written for commit 105ab95. Summary will update on new commits.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
cmux	Ready	Preview, Comment	Jul 4, 2026 11:05pm
cmux-staging	Building	Preview, Comment	Jul 4, 2026 11:05pm

[coderabbitai]

Warning

Review limit reached

@austinywang, you've reached your PR review limit, so we couldn't start this review.

Next review available in: 47 minutes

Enable usage-based reviews in Billing to review now. Otherwise, wait until the next included review is available.
You're only billed for reviews past your plan's rate limits ($0.25/file).

How can I continue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

To avoid repeated limits, reduce automatic review volume by pausing incremental auto-reviews earlier, using label-based review opt-in, excluding WIP or generated PR titles, or requesting reviews manually when the PR is ready. If your team needs uninterrupted high-volume reviews, an organization admin can enable usage-based reviews.

How do review limits work?

CodeRabbit enforces per-developer PR review limits for each organization. Most developers receive the normal plan review availability.

For paid Pro and Pro+ PR reviews, CodeRabbit uses adaptive limits for sustained high-volume activity. When a developer's recent PR review activity reaches the 95th percentile or higher among CodeRabbit users, additional reviews become available more gradually as earlier reviews age out of the rolling window.

Please refer docs for additional details.

Review details

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 5b721486-66e9-4954-94e4-3d81c989b06d

📥 Commits

Reviewing files that changed from the base of the PR and between 9c91710 and 105ab95.

⛔ Files ignored due to path filters (1)

• .github/swift-file-length-budget.tsv is excluded by !**/*.tsv

📒 Files selected for processing (2)

• CLI/SSHPTYResizeMonitor.swift
• cmuxTests/CLISSHPTYResizeInputTests.swift

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch issue-6306-ssh-workspace-tui-resize-can-remain

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[greptile-apps]

Greptile Summary

This PR fixes stale SSH PTY resize state in SSHPTYResizeMonitor by adding a bounded-backoff retry for failed workspace.remote.pty_resize deliveries. Previously, a timed-out RPC would leave the remote PTY at a stale size until a SIGWINCH or input edge arrived; now a cancellable DispatchWorkItem timer re-drives the drain after 100 ms (doubling to a 2 s cap), input waiters are still resumed immediately on failure, and the retry is torn down cleanly on actor cancellation.

• Production fix (CLI/SSHPTYResizeMonitor.swift): adds scheduleResizeRetry() / resetResizeRetry() / retryPendingResize() to the actor, using a one-shot cancellable DispatchWorkItem timer identical to the existing websocket-keepalive idiom; the retry is properly cancelled in markCancelled() and reset on any successful send.
• Regression test (cmuxTests/CLISSHPTYResizeInputTests.swift): exercises the full ssh-pty-attach CLI against a mock daemon that rejects the first pty_resize and asserts the helper re-delivers the same size automatically, no SIGWINCH or input needed.
• CI budget (.github/swift-file-length-budget.tsv): updated row for the now-larger test file.

Confidence Score: 5/5

Safe to merge — the change is actor-isolated, the retry timer is properly cancellable, and the regression test verifies end-to-end re-delivery without any external event.

The backoff state machine is straightforward: scheduleResizeRetry / resetResizeRetry / retryPendingResize are all actor-isolated, the DispatchWorkItem is stored and cancelled on every success and on teardown, and isCancelled guards the post-cancellation Task hop. Edge cases (timer fires during an active drain, SIGWINCH racing the retry window, multiple consecutive failures) are all handled correctly by the actor's serial execution and the isDraining guard in startDrainIfNeeded. The regression test provides real process-level coverage of the failure → retry → re-delivery path.

No files require special attention.

Important Files Changed

Filename	Overview
CLI/SSHPTYResizeMonitor.swift	Adds bounded-backoff retry logic via a cancellable DispatchWorkItem; actor isolation is correct, backoff state management and cancellation paths are sound.
cmuxTests/CLISSHPTYResizeInputTests.swift	Adds a red/green regression test that forces a failed first resize and verifies automatic re-delivery; test-only NSLock usage is within established scaffolding norms.
.github/swift-file-length-budget.tsv	Mechanically updated to reflect the enlarged test file; no logic changes.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[SIGWINCH / input edge\nor retry timer fires] --> B[recordPendingResize /\nretryPendingResize]
    B --> C{isDraining?}
    C -- Yes --> D[Return — drain loop\nwill pick up new size]
    C -- No --> E[startDrainIfNeeded\nisDraining = true]
    E --> F[drainPendingResizes loop]
    F --> G{pendingSize set?}
    G -- No --> H[resumeInputWaiters\nisDraining = false\nreturn]
    G -- Yes --> I[sendResize via GCD\nbounded 50 ms RPC]
    I --> J{sent OK?}
    J -- Yes --> K[lastSentSize = size\nresetResizeRetry\nbackoff reset\ncancel retryWorkItem]
    K --> G
    J -- No --> L[pendingSize = size\nresumeInputWaiters immediately]
    L --> M[scheduleResizeRetry\nDispatchWorkItem asyncAfter backoff]
    M --> N[retryWorkItem stored\nbackoff doubled up to 2 s]
    N --> O[isDraining = false\nreturn]
    O -.->|timer fires| A
    P[markCancelled / cancel] --> Q[retryWorkItem.cancel\nretryWorkItem = nil\nisCancelled = true]

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A[SIGWINCH / input edge\nor retry timer fires] --> B[recordPendingResize /\nretryPendingResize]
    B --> C{isDraining?}
    C -- Yes --> D[Return — drain loop\nwill pick up new size]
    C -- No --> E[startDrainIfNeeded\nisDraining = true]
    E --> F[drainPendingResizes loop]
    F --> G{pendingSize set?}
    G -- No --> H[resumeInputWaiters\nisDraining = false\nreturn]
    G -- Yes --> I[sendResize via GCD\nbounded 50 ms RPC]
    I --> J{sent OK?}
    J -- Yes --> K[lastSentSize = size\nresetResizeRetry\nbackoff reset\ncancel retryWorkItem]
    K --> G
    J -- No --> L[pendingSize = size\nresumeInputWaiters immediately]
    L --> M[scheduleResizeRetry\nDispatchWorkItem asyncAfter backoff]
    M --> N[retryWorkItem stored\nbackoff doubled up to 2 s]
    N --> O[isDraining = false\nreturn]
    O -.->|timer fires| A
    P[markCancelled / cancel] --> Q[retryWorkItem.cancel\nretryWorkItem = nil\nisCancelled = true]

Loading

_{Reviews (9): Last reviewed commit: "Merge remote-tracking branch 'origin/mai..." | Re-trigger Greptile}