Intermittent Postinstall error for node 18.20.3

[ahernd2]

Current Behavior

When installing node_modules with pnpm we get this error intermittently.

.../.pnpm/nx@19.5.1/node_modules/nx postinstall: /builds/business/services/products/thunder/thunder-ui/node_modules/.pnpm/nx@19.5.1/node_modules/nx/src/project-graph/plugins/isolation/plugin-pool.js:146
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:                 const { resolver, rejector } = pending.get(tx);
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:                         ^
.../.pnpm/nx@19.5.1/node_modules/nx postinstall: TypeError: Cannot destructure property 'resolver' of 'pending.get(...)' as it is undefined.
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at createNodesResult (/builds/business/services/products/thunder/thunder-ui/node_modules/.pnpm/nx@19.5.1/node_modules/nx/src/project-graph/plugins/isolation/plugin-pool.js:146:25)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at consumeMessage (/builds/business/services/products/thunder/thunder-ui/node_modules/.pnpm/nx@19.5.1/node_modules/nx/src/project-graph/plugins/isolation/messaging.js:37:32)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at /builds/business/services/products/thunder/thunder-ui/node_modules/.pnpm/nx@19.5.1/node_modules/nx/src/project-graph/plugins/isolation/plugin-pool.js:73:47
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at Socket.<anonymous> (/builds/business/services/products/thunder/thunder-ui/node_modules/.pnpm/nx@19.5.1/node_modules/nx/src/utils/consume-messages-from-socket.js:13:17)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at Socket.emit (node:events:517:28)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at addChunk (node:internal/streams/readable:368:12)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at readableAddChunk (node:internal/streams/readable:341:9)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at Readable.push (node:internal/streams/readable:278:10)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at Pipe.onStreamRead (node:internal/stream_base_commons:190:23)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall: Node.js v18.20.3
.../.pnpm/nx@19.5.1/node_modules/nx postinstall: Failed

Expected Behavior

Post install should not throw an error randomly

GitHub Repo

No response

Steps to Reproduce

Nx Report

Node   : 18.20.3
OS     : linux-x64
pnpm   : 7.33.4

nx                 : 19.5.1
@nx/js             : 19.5.1
@nx/jest           : 19.5.1
@nx/linter         : 19.5.1
@nx/eslint         : 19.5.1
@nx/workspace      : 19.5.1
@nx/angular        : 19.5.1
@nx/cypress        : 19.5.1
@nx/devkit         : 19.5.1
@nx/esbuild        : 19.5.1
@nx/eslint-plugin  : 19.5.1
@nx/express        : 19.5.1
@nx/node           : 19.5.1
@nx/playwright     : 19.5.1
@nx/plugin         : 19.5.1
@nx/storybook      : 19.5.1
@nrwl/tao          : 19.5.1
@nx/web            : 19.5.1
@nx/webpack        : 19.5.1
typescript         : 5.5.3
---------------------------------------
Community plugins:
@ngneat/transloco    : 4.3.0
@ngrx/effects        : 18.0.1
@ngrx/operators      : 18.0.1
@ngrx/router-store   : 18.0.1
@ngrx/store          : 18.0.1
@ngrx/store-devtools : 18.0.1
@storybook/angular   : 7.5.3
ngx-build-plus       : 13.0.1
ngx-toastr           : 16.2.0
---------------------------------------
Local workspace plugins:
         workspace-plugin

Failure Logs

.../.pnpm/nx@19.5.1/node_modules/nx postinstall: /builds/business/services/products/thunder/thunder-ui/node_modules/.pnpm/nx@19.5.1/node_modules/nx/src/project-graph/plugins/isolation/plugin-pool.js:146
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:                 const { resolver, rejector } = pending.get(tx);
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:                         ^
.../.pnpm/nx@19.5.1/node_modules/nx postinstall: TypeError: Cannot destructure property 'resolver' of 'pending.get(...)' as it is undefined.
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at createNodesResult (/builds/business/services/products/thunder/thunder-ui/node_modules/.pnpm/nx@19.5.1/node_modules/nx/src/project-graph/plugins/isolation/plugin-pool.js:146:25)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at consumeMessage (/builds/business/services/products/thunder/thunder-ui/node_modules/.pnpm/nx@19.5.1/node_modules/nx/src/project-graph/plugins/isolation/messaging.js:37:32)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at /builds/business/services/products/thunder/thunder-ui/node_modules/.pnpm/nx@19.5.1/node_modules/nx/src/project-graph/plugins/isolation/plugin-pool.js:73:47
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at Socket.<anonymous> (/builds/business/services/products/thunder/thunder-ui/node_modules/.pnpm/nx@19.5.1/node_modules/nx/src/utils/consume-messages-from-socket.js:13:17)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at Socket.emit (node:events:517:28)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at addChunk (node:internal/streams/readable:368:12)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at readableAddChunk (node:internal/streams/readable:341:9)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at Readable.push (node:internal/streams/readable:278:10)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall:     at Pipe.onStreamRead (node:internal/stream_base_commons:190:23)
.../.pnpm/nx@19.5.1/node_modules/nx postinstall: Node.js v18.20.3
.../.pnpm/nx@19.5.1/node_modules/nx postinstall: Failed

### Package Manager Version

pnpm 7.33.4

### Operating System

- [ ] macOS
- [X] Linux
- [ ] Windows
- [ ] Other (Please specify)

### Additional Information

_No response_

[ms-tng]

Hi, I'm seeing the same error in our CI (Bitbucket pipelines), so I investigated a bit. It is due to plugin isolation now being enabled by default (#26888). As it turns out, the names of the sockets used to communicate between the main process and the isolated plugin workers become longer than the maximum length allowed by the OS and hence get truncated, causing the part that makes them unique to be cut off. Consequently, multiple workers try to listen on the same socket and fail, causing them to exit prematurely. The error you see in the log is when Nx tries to access a pending promise that has already been disposed of because the worker is no longer there.

Regarding the long names: This specifically happens with pnpm because pnpm tends to produce folder structures with very long names under node_modules. The maximum allowed length should be 108 characters, though in our Bitbucket pipelines setup it seems to be 119. I'm not sure why it doesn't happen every time - maybe it has to do with the way the workers are scheduled.

As a workaround, you can set NX_ISOLATE_PLUGINS=false to disable the isolation as used to be the default before 19.5.

[ahernd2]

@ms-tng Thanks for the workaround, it does seem to work, haven't seen any failures since

[Squixx]

Also running into this, will test using NX_ISOLATE_PLUGINS=false What's interesting though is that i'm only getting this on linux. MacOs seems fine

[AgentEnder]

@ms-tng Wow! Thanks for the in-depth investigation. We've been pulling our hair out a bit trying to figure out what could cause this, and didn't think to look into the socket path but that makes sense.

I'll look into how we could better produce a shorter path. Thanks again 🙏

[AgentEnder]

Also, @ms-tng do you have an example of the long path names? It sounds like you've likely captured them already. I want to make sure the linked fix is good enough here.

[ms-tng]

@AgentEnder Glad I could help!

Here is an example socket path:

/opt/atlassian/pipelines/agent/build/node_modules/.pnpm/nx@19.5.1_patch_hash=v6gmoqjqk3tlj6q6xv6dk5mfxa_@swc-node+register@1.10.9_@swc+core@1.7.0_@sw_vw6ll7aabdl4qlr3pnaeetdsym/node_modules/nx/ddbea83aee87088f0551/plugin134-1.sock

Note that it contains a patch hash because I had patched Nx to increase logging, so it's a bit on the longer side.

[ms-tng]

@FrozenPandaz After your fix in #27073, the socket path is still too long, e.g.

/opt/atlassian/pipelines/agent/build/node_modules/.pnpm/nx@19.5.3_patch_hash=kaneqlcv6bufy26krrufvbfnhm_@swc-node+register@1.10.9_@swc+core@1.7.1_@sw_xjzfe2hqhqzzioudxymq7jg5sy/node_modules/nx/plugin181-1.sock

Looking at your change, I saw that one can define the socket directory by setting NX_DAEMON_SOCKET_DIR, so I'm now able to use plugin isolation by setting NX_DAEMON_SOCKET_DIR=$(mktemp -d). I didn't check the code and don't know all the implications, but maybe you could also use the default OS temp dir instead of a subdirectory of node_modules.

[viceice]

shouldn't those files land in <project-dir/.nx ?

[csi-lk]

We're now seeing this issue on node 20.15.1, specifically in CI environments and not locally after updating