Todd Lipcon created KUDU-1369:
---------------------------------
Summary: client does not fail over snapshot scans when querying
lagging replicas
Key: KUDU-1369
URL: https://issues.apache.org/jira/browse/KUDU-1369
Project: Kudu
Issue Type: Bug
Components: client
Affects Versions: 0.7.0
Reporter: Todd Lipcon
If the client species SCAN_AT_SNAPSHOT, and then tries to read from a replica,
it's possible the replica won't have recent enough data to service the scan, or
the replica may have some operations that are "stuck" started but not yet
committed because the leader recently crashed. In this case, it responds with
'Timed out: could not wait for desired snapshot timestamp to be consistent:
Timed out waiting for all transactions with ts < P: 1457574158715836 usec, L: 0
to commit'. However, it's possible (likely, even) that another replica does
have this operation committed. The client doesn't handle this error at the
moment and instead propagates it to the caller even if it could otherwise
failover.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
