Details
-
Bug
-
Resolution: Fixed
-
Critical
-
6.6.0
-
Untriaged
-
1
-
Yes
Description
What's the issue?
We are running a test case where we start cbbackupmgr running a backup then asynchronously kill erlang, at which point we wait for cbbackupmgr to timeout. If killed at the right/wrong time then cbbackupmgr will fail to correctly teardown the connection to the cluster.
What's the root cause of the issue?
After looking at the stack trace of the running process/debug logs we can see that gocbcore is receiving an EOF when trying to read from the connect to the cluster. This EOF error doesn't appear to be returned to cbbackupmgr causing a wait group not to be decremented (hence the hang).
If we look at the attached logs we can see that the stream observer 'End' function is not being run with the received EOF error. Gocbcore then continues (in a loop) to indefinitely recreate the connection to the server (using CCCPPOLL).
Attachments
Issue Links
- is caused by
-
GOCBC-929 gocbcore v9 DCP is failing to propagate EOF/socket read failure errors up to the application
- Resolved