fix(cua-driver): retry daemon socket writes on EAGAIN instead of failing fatally (write-side mirror of #1997)

[LaZzyMan]

Problem

list_apps (and other tools) intermittently fail with:

MCP error -32603: daemon transport error forwarding `list_apps`: Resource temporarily unavailable (os error 35)

even though the request is tiny. Seen across a large fraction of trajectories; the heavy hitters are slow system queries (list_apps, get_accessibility_tree, launch_app).

Root cause

serve.rs::send_request handles EAGAIN asymmetrically:

• read side (the response loop) already treats WouldBlock/TimedOut as "daemon still working, keep waiting" up to a 120 s deadline — the fix(cua-driver): don't treat socket read-timeout (EAGAIN) as fatal in daemon proxy (#1864) #1997 fix for cua-driver MCP daemon proxy returns os error 35 forwarding get_window_state #1864.
• write side had a 5 s SO_SNDTIMEO (set_write_timeout) but no retry: w.write_all(...)? propagates the timeout as a bare os error 35.

When the daemon (a machine-wide singleton serving every proxy) is momentarily too busy to read — backpressure from concurrent slow system tools — the proxy's write times out and the bare EAGAIN becomes a fatal transport error, wrapped by proxy.rs into the message above.

That os error 35 has no connect to …: prefix (so not the connect path) and isn't the timed out after 120s read message — by elimination it's write_all, the only bare-EAGAIN site left after #1997 fixed the read side. A daemon momentarily not reading is the exact write-side mirror of a daemon still computing a response: not a transport failure.

Fix

cua-driver-core::socket_io::write_all_with_retry: write with offset bookkeeping, treat WouldBlock/TimedOut as "keep waiting" until an overall deadline (the same 120 s budget as the read loop). send_request uses it instead of write_all. Extracted to core so the retry logic is unit-testable without linking the platform crates (and their Swift/Metal interop) the cua-driver binary pulls in.

Tests

New socket_io tests:

• retries_through_transient_eagain — N transient WouldBlocks then success.
• times_out_when_daemon_never_drains — a never-draining writer surfaces TimedOut (not a bare EAGAIN) after the deadline.
• accumulates_partial_writes — offset bookkeeping for short writes.

Verification status

• ✅ cargo test -p cua-driver-core green (3 socket_io tests + all core).
• ✅ cargo check -p cua-driver green.
• ⚠️ Not reproduced against a live backpressured daemon: the cua-driver binary doesn't link on my machine (unrelated platform-macos Swift/Metal interop vs the local SDK) and I can't re-run the eval that surfaced this. The retry logic is unit-tested and the send_request change is a drop-in for the existing write_all.

Summary by CodeRabbit

• Bug Fixes
  • Improved socket writes to better handle temporary backpressure and partial writes.
  • Reduced failures caused by transient “resource temporarily unavailable” errors when communicating with the daemon.
  • Added a longer overall timeout so requests can keep retrying before failing.
  • Included validation to ensure writes complete reliably across retries and short writes.

[vercel]

@LaZzyMan is attempting to deploy a commit to the Cua Team on Vercel.

A member of the Team first needs to authorize it.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[coderabbitai]

📝 Walkthrough

Walkthrough

The crate now exports a socket write helper that retries partial, interrupted, and backpressured writes until a deadline. Unix send_request now uses that helper with a 120-second deadline instead of direct write_all/flush.

Changes

Socket Write Retries

Layer / File(s)	Summary
Core helper and export libs/cua-driver/rust/crates/cua-driver-core/src/lib.rs, libs/cua-driver/rust/crates/cua-driver-core/src/socket_io.rs	cua-driver-core exports socket_io, which defines write_all_with_retry and unit tests for transient WouldBlock, partial writes, and deadline timeout.
Unix send_request integration libs/cua-driver/rust/crates/cua-driver/src/serve.rs	Unix send_request uses write_all_with_retry with a 120-second deadline and drops the unused Write import.

Sequence Diagram(s)

sequenceDiagram
  participant unix_send_request as "Unix send_request"
  participant write_all_with_retry as "write_all_with_retry"
  participant daemon_socket as "daemon socket"

  unix_send_request->>write_all_with_retry: bytes + 120s deadline
  loop until all bytes are written or deadline expires
    write_all_with_retry->>daemon_socket: Write::write
    daemon_socket-->>write_all_with_retry: partial write / Interrupted / WouldBlock / TimedOut
  end
  write_all_with_retry-->>unix_send_request: Ok(()) or TimedOut(bytes_written)

Loading

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Poem

I hopped through bytes with twitchy nose,
Past EAGAIN winds and sleepy throes.
A deadline moon kept me on track,
I wrote it all and flushed on back,
Hop-hop—*thump*—the socket answered back. 🐇

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: daemon socket writes now retry on EAGAIN instead of failing immediately.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}

[coderabbitai]

🧹 Nitpick comments (1)

libs/cua-driver/rust/crates/cua-driver-core/src/socket_io.rs (1)

21-62: 🩺 Stability & Availability | 🔵 Trivial | 💤 Low value

Document the blocking-write precondition for this public helper.

The retry loop has no sleep/backoff: on WouldBlock/TimedOut it immediately re-issues w.write(...). This is safe only because the sole caller sets SO_SNDTIMEO (5s) so each write call blocks before returning TimedOut. A future caller passing a genuinely non-blocking writer (no send timeout) would turn this into a CPU-bound hot loop until the deadline. Since this is now exported from core, consider documenting that the writer must be blocking-with-timeout (the doc comment mentions SO_SNDTIMEO but doesn't state it as a requirement).

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@libs/cua-driver/rust/crates/cua-driver-core/src/socket_io.rs` around lines 21
- 62, Document write_all_with_retry as requiring a blocking writer with a send
timeout, since the retry path immediately loops on WouldBlock/TimedOut with no
backoff. Update the public doc comment on write_all_with_retry in socket_io.rs
to state the precondition clearly, referencing the existing SO_SNDTIMEO behavior
and warning that non-blocking writers are unsupported and may spin until the
deadline.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@libs/cua-driver/rust/crates/cua-driver-core/src/socket_io.rs`:
- Around line 21-62: Document write_all_with_retry as requiring a blocking
writer with a send timeout, since the retry path immediately loops on
WouldBlock/TimedOut with no backoff. Update the public doc comment on
write_all_with_retry in socket_io.rs to state the precondition clearly,
referencing the existing SO_SNDTIMEO behavior and warning that non-blocking
writers are unsupported and may spin until the deadline.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 2e3822f7-afba-4b14-9116-2dbb32ecc995

📥 Commits

Reviewing files that changed from the base of the PR and between e08574c and df22db8.

📒 Files selected for processing (3)

• libs/cua-driver/rust/crates/cua-driver-core/src/lib.rs
• libs/cua-driver/rust/crates/cua-driver-core/src/socket_io.rs
• libs/cua-driver/rust/crates/cua-driver/src/serve.rs