Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
-
ghx-label-2
Description
We've seen queries hanging in the planner. The concrete hang was in the Kudu client, for that I opened KUDU-3019.
But we should investigate the possibilities of adding an overall timeout on the Frontend activities (analysis/authorization/planning).
It'd be useful because the Frontend reaches out to other services – e.g. Catalogd, HDFS Namenode, authorization provider, Kudu, etc. --. and an overall timeout would be a robust solution for these kind of hangs in broken API calls.
Of course, in the meantime we should still use timeouts on each API call where possible.
Attachments
Issue Links
- is related to
-
IMPALA-2568 ExecuteStatement RPC (and beeswax query() RPC) should not block
- Open