[BUG, NEED VERIFICATION] $worker->getWorker()->error((string)$e); cause a force kill and lost error msg

[dhopkalo]

When I created my own worker process and use the code example for the push error to the rr server for workers

    } catch (\Throwable $e) {
        $worker->getWorker()->error((string)$e);
    }

I don't see my original error only the next errors:

2021-05-26T16:35:46.711+0300    DEBUG   server          server/plugin.go:225    worker destructed       {"pid": 24906}
2021-05-26T16:35:46.725+0300    ERROR   server          server/plugin.go:238    worker_watcher_wait: signal: killed; process_wait: signal: killed
github.com/spiral/roadrunner/v2/plugins/server.(*Plugin).collectEvents
        github.com/spiral/roadrunner/v2@v2.2.1/plugins/server/plugin.go:238
github.com/spiral/roadrunner/v2/pkg/events.(*HandlerImpl).Push
        github.com/spiral/roadrunner/v2@v2.2.1/pkg/events/general.go:37
github.com/spiral/roadrunner/v2/pkg/worker_watcher.(*workerWatcher).wait
        github.com/spiral/roadrunner/v2@v2.2.1/pkg/worker_watcher/worker_watcher.go:235
github.com/spiral/roadrunner/v2/pkg/worker_watcher.(*workerWatcher).addToWatch.func1
        github.com/spiral/roadrunner/v2@v2.2.1/pkg/worker_watcher/worker_watcher.go:260
2021-05-26T16:35:47.278+0300    DEBUG   server          server/plugin.go:223    worker constructed      {"pid": 24965}

I expected to see my (string)$e in the rr logs

Instead, i see that worker process was force killed

I create a simple fix for my issue: #690

[rustatian]

Hey @drefixs. Thanks for the report.

1. Could you please send the whole .rr.yaml you used and the PHP (your) worker sample to reproduce your problem?
2. Your OS version.

[dhopkalo]

Hi @48d90782 thanks for the quick responce:

server:
   command: "php bin/console baldinof:roadrunner:worker"

http:
  address: 0.0.0.0:80
  middleware: [ "static", "gzip" ]
  pool:
    num_workers: 4
  uploads:
    forbid: [ ".php", ".exe", ".bat" ]
  static:
    dir: "web"
    forbid: [ ".php", ".htaccess" ]

logs:
  mode: production
  encoding: console
  channels:
    http:
      mode: production
      level: debug
      encoding: console
      output: stdout
      err_output: stderr

rpc:
  listen: tcp://127.0.0.1:6001

Also I noticed that if you change http.pool.debug=true it starts to work normally without force killing the process with log

2. OSX(Darwin)(x86) and Docker Alpine linux (x86) on server

[dhopkalo]

@48d90782 May be i found the reason
https://github.com/spiral/roadrunner/blob/master/pkg/pool/static_pool.go#L267
this part of the code first in the else condition tries to do sp.ww.Push(w) with force kill inside for the worker that sent frame ERROR

after that it try again to do graceful termination with w.Stop()

when I just remove this else condition all start work as expected and worker_watcher push stopped worker back to the vector queue also stop command have fallback kill logic inside.

[rustatian]

First of all, please, provide a worker you used and full steps on how to reproduce this issue. I'll definitely take into account your investigation, thanks 👍🏻

[rustatian]

@drefixs Am I right, that you are expecting to see smt like this in the logs output?:

where test_error is:

[rustatian]

Or, I guess, that would be better to see smt like this (without RR specific error wrapping):

[rustatian]

Fixed, will be as on the second screenshot.