[teamleaderleo]

Leo-ask: ask-e09ceb4656

Leo, 2026-11-04: "why is it taking so long to start a chat "in cloud"?"

Why it takes long

Every first send from a Mac workspace that has no Cloud machine yet makes a new machine (one workspace per small Cloud machine, #30020), so the wait is mostly that machine starting. On top of that, the client wasted time:

• While a new machine provisioned, CloudMove.connected slept 3 s before every read, the first included, then did a full list read. A machine that came up just after a read was seen up to 3 s late.

This change

• The first 20 reads are 1 s apart, then 3 s (80 reads, about 3 min, up from 60 × 3 s).
• Each pause ends as soon as any other read (the change tracker's refresh, a list read) shows the machine running, and then no extra list read is made.
• Cancelling the start still ends the wait at once.

Expected saving: about 1–3 s per first send that makes a machine (average lag after the machine runs goes from ~1.5 s + the 3 s first sleep to ~0.5 s). Not measured on a real start yet; the larger fixes for this ask are the prewarm PRs already open (#34974 prewarm while typing, #34487 warm on composer select), which start the machine before Send.

Risk: low. One wait loop's pacing; a provisioning machine now costs up to 20 extra list reads in its first 20 s.

Test: CloudMachineChangeStateTests.aStartWaitsForANewMachineToRun gains a case: a machine seen running during a pause is not read again.

Agent session: agent session

🤖 Generated with Claude Code

---

^{Open in [code]smith · Need help on this PR? Tag @codesmith-bot with what you need.}

---

Summary by cubic

Speeds up the first send from a Mac workspace that provisions a new Cloud machine. Previously CloudMove slept 3 s before every state read (the first included), so a machine that came up just after a read was seen up to 3 s late.

• Reads 1 s apart for the first 20 reads, then 3 s (80 reads, about 3 min) instead of 60 reads at 3 s.
• Stops waiting as soon as any read shows the machine running, with no extra list read; cancelling still ends the wait immediately.

^{Written for commit d26c8e7. Summary will update on new commits.}

[github-actions]

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

Review d26c8e7: HQ feed card · gallery before/after runs · checks

App of this head (app.zip, kept 1 day)

gh workflow run reload-build.yml -R manaflow-ai/cmux --ref feat-cmux-next -f tag=pr35023 -f ref=lane/cloud-start-faster-wait -f platform=macos -f backend_mode=local

Shipped in:

• cmux NEXT nightly 1.0.0-nightly.4182499138501 (e0aa712b3b)
• cmux NEXT nightly 1.0.0-nightly.4184087390301 (08052d331e)
• cmux NEXT nightly 1.0.0-nightly.4184224384601 (a3abfc4160)
• cmux NEXT nightly 1.0.0-nightly.4184409870901 (78ee0e108e)

[teamleaderleo]

Review (claude-opus-5-5): SHIP at d26c8e7

Review (subagent claude-opus-5-5 medium): SHIP at d26c8e7

Checked Swift 6 isolation (the condition closure captures session like CloudHandlers.swift:363), that machine.status only becomes .running from a server record applied through listed/applyRecord (so skipping the forced read loses nothing), cancellation (CloudWait.holds returns, checkCancellation throws, untilRunning returns), the wakeup-allow lint (the only direct sleep is gone), and the new test case (old code reads once, new code zero times). check-concurrency.sh passes.

Follow-ups: [follow-up] the longest wait is about 200 s, up from 180 s, with up to 20 reads in the first 20 s of a new machine (not blocking).

[blacksmith-sh]

Found 6 test failures on Blacksmith runners:

Failures

Test	View Logs
/tmp/tmpc5ig2s6c/summary-missing.json (live):/summary-missing.json (live):	View Logs
/tmp/tmpdwk6thlu/summary-fail.json (live):/summary-fail.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

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