Summary
The PREBUILD phase ends by awaiting a second workingdir package-manager install. That spawn resolves on the child's close event and has no inactivity watchdog. When the package manager wedges after printing its final line, the build produces zero further output and runs silently until the max-duration cap kills it. The neighboring phases both protect against exactly this: INSTALL_DEPENDENCIES and INSTALL_PODS carry 15/30-minute inactivity watchdogs. The PREBUILD trailing install is the unprotected sibling.
What happened
Four consecutive iOS builds of our SDK 56 pnpm monorepo hung with the same signature:
- Last log line:
Done in Xs using pnpm v11.5.2 (the PREBUILD workingdir install finishing).
- No
End phase: PREBUILD marker. INSTALL_PODS never logs a single line.
- Silence until the max-duration cap cancels the build (45 min free tier, 120 min paid).
--clear-cache changed nothing. Verbose env changed nothing. Android built the same commits fine.
Build IDs (project 5e6bf560-6110-4ca1-8268-b4076dc2fc7b, account groundtruth):
398a904d-be61-487d-8924-0a18fc611e5f (killed at 45 min)b7f0505f-6630-4089-b252-085627b2fe44 (killed at 120 min)8133efff-d281-44b1-8caf-42cd6ee5975c (killed at 120 min, --clear-cache)d9e54ed1-25ef-4233-b658-15f2bd1d6b74 (self-canceled at 20 min during diagnosis)
Root cause
pnpm 11.5.2 on the darwin worker completes the install, prints Done, and never exits. This matches a known pnpm bug family (for example pnpm/pnpm#9530: install hangs indefinitely after showing done). Because the spawn resolves on close and nothing watches it, the build waits forever with no output.
Pinning packageManager: "pnpm@11.21.0" fixed it immediately. The identical build now finishes in under 20 minutes.
Request
Add the same inactivity watchdog that INSTALL_DEPENDENCIES and INSTALL_PODS already have to the PREBUILD trailing install spawn. A wedged package manager should fail fast with a clear error instead of burning the full build window in silence. A stuck build at this boundary is very hard to diagnose from the outside: the log looks like a successful install with nothing after it.
Environment
- eas-cli 12.x, iOS,
production profile, Expo SDK 56, RN 0.85.3
- pnpm workspace monorepo,
nodeLinker: hoisted, app in apps/mobile
- Failing pnpm: 11.5.2. Working pnpm: 11.21.0. Same commit otherwise.
Summary
The PREBUILD phase ends by awaiting a second workingdir package-manager install. That spawn resolves on the child's
closeevent and has no inactivity watchdog. When the package manager wedges after printing its final line, the build produces zero further output and runs silently until the max-duration cap kills it. The neighboring phases both protect against exactly this: INSTALL_DEPENDENCIES and INSTALL_PODS carry 15/30-minute inactivity watchdogs. The PREBUILD trailing install is the unprotected sibling.What happened
Four consecutive iOS builds of our SDK 56 pnpm monorepo hung with the same signature:
Done in Xs using pnpm v11.5.2(the PREBUILD workingdir install finishing).End phase: PREBUILDmarker. INSTALL_PODS never logs a single line.--clear-cachechanged nothing. Verbose env changed nothing. Android built the same commits fine.Build IDs (project
5e6bf560-6110-4ca1-8268-b4076dc2fc7b, accountgroundtruth):398a904d-be61-487d-8924-0a18fc611e5f(killed at 45 min)b7f0505f-6630-4089-b252-085627b2fe44(killed at 120 min)8133efff-d281-44b1-8caf-42cd6ee5975c(killed at 120 min,--clear-cache)d9e54ed1-25ef-4233-b658-15f2bd1d6b74(self-canceled at 20 min during diagnosis)Root cause
pnpm 11.5.2 on the darwin worker completes the install, prints
Done, and never exits. This matches a known pnpm bug family (for example pnpm/pnpm#9530: install hangs indefinitely after showing done). Because the spawn resolves oncloseand nothing watches it, the build waits forever with no output.Pinning
packageManager: "pnpm@11.21.0"fixed it immediately. The identical build now finishes in under 20 minutes.Request
Add the same inactivity watchdog that INSTALL_DEPENDENCIES and INSTALL_PODS already have to the PREBUILD trailing install spawn. A wedged package manager should fail fast with a clear error instead of burning the full build window in silence. A stuck build at this boundary is very hard to diagnose from the outside: the log looks like a successful install with nothing after it.
Environment
productionprofile, Expo SDK 56, RN 0.85.3nodeLinker: hoisted, app inapps/mobile