Details
-
Task
-
Resolution: Done
-
Major
-
None
-
None
-
1
-
SDK14: Tracing/Otel
Description
Specifically, the readWritePermissionsOnOneCollectionNoPermsOnOtherCollection will fail, complaining that the one doc that should have been staged as an insert, then rolled back, does not exist as a tombstone in the collection we can write to. What is happening is the test has a 500ms expiry. When we start executing the transaction, we call cluster->bucket("default"), which actually connects to the bucket since it is the first time it was called on this new connection. That sometimes exceeds 500ms and boom - the txn times out before the first insert happens.
Simple fix is to have the performer cache the bucket. We used to do that, but at some point we stopped since we were not using the exposed bucket from the connection object. Lets add that back in, as it helps for this test (and maybe others?).