parcel build randomly hangs forever

[mzihlmann]

🐛 bug report

When running parcel build as part of our docker build in CI, we have a 50% chance that the build runs through immediately and a 50% chance that the build gets stuck for more than 1 hour. adding --log-level error reduced the likelyhood of getting stuck (1 out of 6) but did not fully solve the issue. --no-cache appears to resolve the issue.

#7585 (comment)
#6473 (comment)

🎛 Configuration (.babelrc, package.json, cli command)

  "scripts": {
    ...
    "build": "rm -fr dist && parcel build index.ts admin.ts",
    ...
  },
  "devDependencies": {
    ...
    "@parcel/transformer-sass": "^2.8.3",
    ...
  }

🤔 Expected Behavior

The build should not get stuck

😯 Current Behavior

with a 50/50 chance the build gets stuck on the build command

💁 Possible Solution

to me it sounds like a race-condition leading to a deadlock somewhere in the caching logic, or at least activated by the caching logic. The log-level might impact timing in the critical section and hence change the probability of a deadlock.

🌍 Your Environment

we are running on node:16-alpine with parcel==2.8.3

[mischnic]

Can you share a complete example (Dockerfile, dummy sourcecode, ...)?

[mzihlmann]

unfortunately not yet, as its company code and i did not yet invest the time to create a minimal reproducible setup. but if you have experiments that i could run i'm eager to help.

[fregante]

I noticed that if I stick a random process.exit() in a transformer, Parcel just silently hangs. I assume that the same happens for real errors that are just never surfaced and that might be what's happening here.

[mischnic]

I noticed that if I stick a random process.exit() in a transformer, Parcel just silently hangs.

I added a console.log and process.exit() and it looks like the worker is restarted and the transformer rerun (repeatedly until you stop Parcel). Should Parcel log something in that case?

[fregante]

I don’t think the alternative is great, so yes:

parcel build randomly hangs forever

[alippai]

I didn't try to create a minimum reproducible case yet, but there is a good chance this appears if upgrading AgGrid from v28 to v29