Errors during the releases when there are new GraphQL Executors and still old web servers

[rafaelromcar-parabol]

Issue - Bug

• How serious is the bug? P3
• OS: macOS / Linux / Windows / ?
• Browser: Chrome / Safari / Lynx / ? + version
• Node version: node --version
• NPM version: npm --version

Sentry related issues:

• ACTION-PRODUCTION-K4X
• ACTION-PRODUCTION-JMV

Related with #10453 ?

SEND TO SENTRY Error: Called disconnect without a valid socket                                                                                                                                                     
    at disconnectSocket (/home/node/parabol/dist/gqlExecutor.js:121814:11)                                                                                                                                         
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)                                                                                                                                  
    at async Object.disconnectSocket (/home/node/parabol/dist/gqlExecutor.js:102417:14)                                                                                                                            
    at async Object.MutationdisconnectSocketResolver (/home/node/parabol/dist/gqlExecutor.js:144099:22) {                                                                                                          
  locations: [ { line: 3, column: 3 } ],                                                                                                                                                                           
  path: [ 'disconnectSocket' ]                                                                                                                                                                                     
}                                                                                                                                                                                                                  
Error: Called disconnect without a valid socket                                                                                                                                                                    
    at disconnectSocket (/home/node/parabol/dist/gqlExecutor.js:121814:11)                                                                                                                                         
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)                                                                                                                                  
    at async Object.disconnectSocket (/home/node/parabol/dist/gqlExecutor.js:102417:14)                                                                                                                            
    at async Object.MutationdisconnectSocketResolver (/home/node/parabol/dist/gqlExecutor.js:144099:22)

Acceptance Criteria (optional)

• Users are disconnected as they should and no errors are shown in the logs
  Triage is performed to find the root cause of the bug, timeboxed to ~1 hour.

Estimated effort: 1 hour to triage. More if root cause is already identified.

[rafaelromcar-parabol]

666 times in the last 15 days, mostly during releases. This might impact user experience during releases.

[rafaelromcar-parabol]

Maybe this is just throwing the error because the user presence was still in Redis but the user was not connected to a socket anymore? 🤔 code here.

We also see a lot of "User does not exist" which doesn't seem to me very clear as error message. If I'm getting it correctly, the user exists, but they do not have a presence in Redis, isn't it?

[Dschoordsch]

The "User does not exist" error happens when disconnectSocket was called with an unknown userId, i.e. we don't have a User with such an id in our db.

[jordanh]

Added this to the backlog

[jordanh]

Lowered priority to p3

[rafaelromcar-parabol]

This has happened again during a release that provoked some alerts in Datadog and PagerDuty, and I suspect that it also provoked some bad user experience. IMHO this should be P2 and addressed in Q1 2025. If not, we might want to avoid releasing during peak hours.