Firestore: Treat 'None' as EOF in 'Watch.on_snapshot'. #8687
Conversation
|
Something weird is going on, I removed and re-added the force run label, the label is there, but there is no record of that on the PR timeline, and the |
|
I cleaned out firestore to see if that helps. forcing a ci run. |
|
The Kokoro jobs for Bigtable and Natural Language both fail due to a Kokoro-internal glitch:
|
|
@tseaver Is there anything we can do to get around the glitch? |
I'm actually worried that there is something more sinister at play. @crwilcox Can you double-check that the clean-out really worked in the CI project? E.g.: documents = []
for collection in client.collections():
documents.extend(collection.list_documents()
assert len(documents) == 0 |
|
I have added a commit to make the system test output verbose, so that we can figure out which test is hanging. |
|
@crwilcox I'm expecting this Kokoro run to time out again: can you post the hung systest output when it does? Or maybe you have the keys to see the output even before it times out? |
Enable debugging hung tests.
- Use a single unique resource ID across the entire systest run. - Add missing document cleanups in collection_group tests. - Avoid calling 'watch.unsubscribe' in cleanup: it causes hangs, and there is no server-side resource leaked. - Add a 'cleanup_firestore_documents.py' utility script.
See:
#8650 (review)