Cannot read property 'name' of undefined

Hello,

I start getting this error on my cloud functions logs interface:

TypeError: Cannot read property 'name' of undefined
    at Firestore.snapshot_ (/user_code/node_modules/firebase-admin/node_modules/@google-cloud/firestore/build/src/index.js:422:127)
    at beforeSnapshotConstructor (/user_code/node_modules/firebase-functions/lib/providers/firestore.js:140:30)
    at changeConstructor (/user_code/node_modules/firebase-functions/lib/providers/firestore.js:144:49)
    at cloudFunctionNewSignature (/user_code/node_modules/firebase-functions/lib/cloud-functions.js:104:28)
    at cloudFunction (/user_code/node_modules/firebase-functions/lib/cloud-functions.js:135:20)
    at /var/tmp/worker/worker.js:768:24
    at process._tickDomainCallback (internal/process/next_tick.js:135:7)

This is happening sometimes but frequently. I set a console log on the top of my function, but it doesn’t seem to be logged, so the function does not even started.

Environment:

"dependencies": {
    "cors": "^2.8.5",
    "express": "^4.16.4",
    "firebase-admin": "~6.0.0",
    "firebase-functions": "^2.1.0",
    "lodash": "^4.17.11"
  },

This issue begins in the same time as the issue mentioned here 357. But I cannot confirm it’s related to. Maybe both bugs are due to a change made by Firebase.

[mtzfactory]

Hi, is still happening…

[joshdifabio]

Hey, we’re so sorry about this; just so you know, it’s a known issue and we’re rolling back now to fix it. Thanks for your patience.

Firestore is an amazing product, but this incident is extremely worrying given the explanation on the incident page:

An upgrade of our infrastructure did not support a feature we thought was no longer in use

How was this able to happen? Are you guys not testing the whole stack before you deploy to production? I would appreciate some reassurance here, as would others I’m sure.

Furthermore, I hope you appreciate the severity of this incident. Applications which use Firestore and depend on at-least-once delivery of events will now be in a corrupted state. Actual downtime would have been much less damaging in many cases.