[
https://issues.apache.org/jira/browse/HIVE-28772?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Harshal Patel updated HIVE-28772:
---------------------------------
Status: Patch Available (was: In Progress)
> Clear REPL_TXN_MAP table on DR when deleting replication policy
> ---------------------------------------------------------------
>
> Key: HIVE-28772
> URL: https://issues.apache.org/jira/browse/HIVE-28772
> Project: Hive
> Issue Type: Bug
> Components: repl
> Reporter: Harshal Patel
> Assignee: Harshal Patel
> Priority: Major
> Labels: Replication, pull-request-available
>
> Currently, if you create scheduled queries for repl dump and repl load, then
> by design during incremental replication transactions can span across
> multiple replication runs.
> And if OPEN_TXN is replicated, then on the DR side it will add an entry in
> the REPL_TXN_MAP table to keep track of open transactions for that database.
> But before replaying COMMIT_TXN/ABORT_TXN, if user deletes the policy and
> drops the database on the DR to re-bootstrap, then we are not cleaning up the
> REPL_TXN_MAP table
> Which can lead to 2 major issues:
> # It can bring down AcidHouseKeeper cleaner services, which clean up
> transaction-related tables in the HMS, because it reads the database name in
> the REPL_TXN_MAP and tries to get the database information for that and fails
> with a database doesn't exist error. This can bloat up the HMS.
> # If user creates a policy quickly before AcidHouseKeeper kicks in the in
> that case newly created database will be marked as repl incompatible from the
> AcidHouseKeeper while cleaning up the TXNS and REPL_TXN_MAP table after 11
> days
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
