You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
[#26979] docdb: Disable load balancing for PgSharedMemTest.LongRead
Summary:
Issue:
`PgCronTest.CancelJobOnLeaderChange` might fail when new system tablets are added on tserver - This can occur, for example, when:
* Enabling `ysql_yb_enable_advisory_locks`, creates an advisory lock system table
* Enabling `ysql_enable_auto_analyze_infra`, creates a stateful service table.
The test performs a long-running read by sleeping for `FLAGS_TEST_transactional_read_delay_ms` (65 seconds) before executing the actual read. During this period, the tablet leader may move to a different node because the newly added system tablets trigger load balancing. After the sleep completes, the system detects the leader change and retries the read on the new leader. However, since each long read takes over 65 seconds and the read timeout is 120 seconds, two consecutive long reads can exceed the timeout limit, resulting in test failure.
Fix:
Disable load balancing during the test to prevent unexpected leader moves during long reads.
Jira: DB-16439
Test Plan: ./yb_build.sh --cxx-test pgwrapper_pg_shared_mem-test --gtest_filter PgSharedMemTest.LongRead
Reviewers: sergei, rthallam
Reviewed By: sergei, rthallam
Subscribers: rthallam, ybase, yql
Differential Revision: https://phorge.dev.yugabyte.com/D43567
0 commit comments