Details
-
Bug
-
Status: Resolved
-
Blocker
-
Resolution: Fixed
-
Impala 3.0, Impala 2.12.0
-
None
-
ghx-label-1
Description
Upon catalog restart the coordinators detect this event and request a full topic update from the statestore. In certain cases, the topic update protocol executed between the statestore and the catalog fails to detect catalog objects that were deleted from the Metastore externally (e.g. via HIVE), thus causing these objects to show up again in each coordinator's catalog cache. The end result is that the catalog server and the coordinator's cache are out of sync and in some cases the only solution is to restart both the catalog and the statestore.
The following sequence can reproduce this issue:
impala> create table lala(int a);
bash> kill -9 `pidof catalogd`
hive> drop table lala;
bash> restart catalogd
impala> show tables;
--- lala shows up in the list of tables;
Attachments
Issue Links
- breaks
-
IMPALA-7315 tests_statestore.py fails at assert len(args.topic_deltas[topic_name].topic_entries) == 1
- Resolved
- is broken by
-
IMPALA-5058 Improve concurrency of DDL/DML operations during catalog updates
- Resolved