Levani Kokhreidze created FLINK-39403:
-----------------------------------------
Summary: Postgres snapshotting phase might cause transaction leak
Key: FLINK-39403
URL: https://issues.apache.org/jira/browse/FLINK-39403
Project: Flink
Issue Type: Bug
Components: Flink CDC
Reporter: Levani Kokhreidze
Attachments: Screenshot 2026-04-03 at 14.07.58.png
We observe increase in "idle in transaction" times when using Flink CDC (3.4.0
version) with Postgres database. One possible cause might be
{code:java}
PostgresSnapshotSplitReadTask ->
PostgresQueryUtils.readTableSplitDataStatement ->
PostgresQueryUtils.initStatement
{code}
initStatement in return disables auto commit for the connection due to setting
the fetchSize on the statement:
{code:java}
private static PreparedStatement initStatement(JdbcConnection jdbc, String sql,
int fetchSize)
throws SQLException {
final Connection connection = jdbc.connection();
connection.setAutoCommit(false);
final PreparedStatement statement = connection.prepareStatement(sql);
statement.setFetchSize(fetchSize);
return statement;
} {code}
There's no explicit commit/rollback calls which might cause the transaction
leak.
Attached screenshot shows one of the replica databases showing increase in the
idle in transaction metric. Table in question had evenly split chunks and there
was no other process connecting to that database but Flink CDC.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
