Eugene Koifman created HIVE-14427:
-------------------------------------
Summary: CompactionTxnHandler.markCleaned() can delete aborted txns
Key: HIVE-14427
URL: https://issues.apache.org/jira/browse/HIVE-14427
Project: Hive
Issue Type: Improvement
Components: Transactions
Reporter: Eugene Koifman
We can modify
{noformat}
s = "select distinct txn_id from TXNS, TXN_COMPONENTS where txn_id = tc_txnid
and txn_state = '" +
TXN_ABORTED + "' and tc_database = '" + info.dbname + "' and tc_table
= '" +
info.tableName + "'" + (info.highestTxnId == 0 ? "" : " and txn_id <=
" + info.highestTxnId);
{noformat}
to use select txn_id, count(*) ... group by txn_id so that we know the number
of components in a TXN.
Then when running "delete from TXN_COMPONENTS where..." we know how many rows
were deleted.
If the sum of all values from 1st query matched total number of rows deleted,
we know that all Aborted txns in this set are empty and thus can be deleted
here.
This means we clean up aborted txns from TXNS table quicker and avoid a large
join in _cleanEmptyAbortedTxns()_. Also, doing delete on TXNS here will have
PKs in WHERE clause so it should be cheap.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
