This pull request introduces 1 alert when merging 4fd6c4a into 4e2ac0c - view on LGTM.com

new alerts:

• 1 for Unreachable statement

This pull request introduces 1 alert when merging a327521 into 4e2ac0c - view on LGTM.com

new alerts:

• 1 for Unreachable statement

This pull request introduces 1 alert when merging 0985271 into 4e2ac0c - view on LGTM.com

new alerts:

• 1 for Unreachable statement

This pull request introduces 1 alert when merging 7e05941 into 7959b22 - view on LGTM.com

new alerts:

• 1 for Unreachable statement

@ampproject/wg-infra @samouri @jridgewell

This PR is my attempt to recreate and figure out the reason for an occasional build failure that affects, seemingly, only CircleCI (I haven't had any reports of it happening to anyone locally, although it might be happening and people are just not reporting it)

The error thrown happens on amp dist:

ERROR events.js:377
      throw er; // Unhandled 'error' event
      ^

Error: write EPIPE
    at afterWriteDispatched (internal/stream_base_commons.js:156:25)
    at writeGeneric (internal/stream_base_commons.js:147:3)
    at Socket._writeGeneric (net.js:787:11)
    at Socket._write (net.js:799:8)
    at writeOrBuffer (internal/streams/writable.js:358:12)
    at Socket.Writable.write (internal/streams/writable.js:303:10)
    at Object.writeToStdin (/home/circleci/project/node_modules/esbuild/lib/main.js:1741:19)
    at sendResponse (/home/circleci/project/node_modules/esbuild/lib/main.js:623:14)
    at handleRequest (/home/circleci/project/node_modules/esbuild/lib/main.js:629:11)
    at handleIncomingPacket (/home/circleci/project/node_modules/esbuild/lib/main.js:700:7)
Emitted 'error' event on Socket instance at:
    at emitErrorNT (internal/streams/destroy.js:106:8)
    at emitErrorCloseNT (internal/streams/destroy.js:74:3)
    at processTicksAndRejections (internal/process/task_queues.js:82:21) {
  errno: -32,
  code: 'EPIPE',
  syscall: 'write'
}

What we know so far is:

• It started after 🏗 Parallelize dist steps #35943, which is the last PR that I think might be related (yes, I made this change)
• It's (almost) always the Nomodule build (I found only one Module build fail with the same error)
• it's (almost) always after minifying amp-shadow-dom-polyfill-0.1.js (I found a few cases where this line didn't show up in the logs)
• It happens on around 3-6% of builds (I can pretty easily recreate the error by running the build in ×100 parallel machines, it fails on between 3-6 of them)
• The error is thrown from esbuild but it looks like it's a much lower level issue. The logs aren't that helpful
• Restarting the job usually fixes it

EPIPE errors indicate that an attempt to write (or read, not the case here though) to a pipe failed, usually because it closed, there's not enough memory at the other end, etc.

In this PR I either made changes to the file node_modules/esbuild/lib/main.js to try to figure it out, or made other type of environmental changes. So far none of these attempts resulted in any changes or insights:

• Tried wrapping the write call in a try/catch. This didn't work because the failure happens at the syscall level, and so the process ends abruptly
• Tried to replace direct writes to child.stdin with a PassThrough stream that waits on drain events when needed, based on this
• Tried to increase the memory of the V8 engine in the node process, as suggested by CircleCI
• Tried logging various stream/pipe related fields, such as child.stdin.writable, all of them are as expected
• Tried to replace the call to esbuild.build with esbuild.buildSync, but that's impossible because buildSync does not support plugins (we pass the Babel plugin)
• Updated esbuild from v0.12.24 to v0.13.2
• Updated Node from --lts to v16.10.0

I'm really stumped with this one and could really use help!

This pull request introduces 1 alert when merging 5923b61 into baa2163 - view on LGTM.com

new alerts:

• 1 for Unreachable statement