Is your feature request related to a problem? Please describe.
In a customer case, we saw almost 10 billion intents created over multiple ranges over a 5 hour window from a single INSERT INTO ... SELECT FROM statement that was ultimately aborted. We should create a perturbation test that simulates this behavior and validates that whether the intents are cleaned up by either a conflicting higher priority transaction or the mvcc gc queue.
Describe the solution you'd like
We have seen large availability outages due to LSM inversion when the mvcc gc queue cleaned up these intents.
Describe alternatives you've considered
There is an existing test registerIntentResolutionOverload which tests that the LSM does not get too overloaded. We can replace this test with a perturbation test.
kvoli
commented
14 hours ago
Is your feature request related to a problem? Please describe. In a customer case, we saw almost 10 billion intents created over multiple ranges over a 5 hour window from a single INSERT INTO ... SELECT FROM statement that was ultimately aborted. We should create a perturbation test that simulates this behavior and validates that whether the intents are cleaned up by either a conflicting higher priority transaction or the mvcc gc queue.
Describe the solution you'd like We have seen large availability outages due to LSM inversion when the mvcc gc queue cleaned up these intents.
Describe alternatives you've considered There is an existing test
registerIntentResolutionOverloadwhich tests that the LSM does not get too overloaded. We can replace this test with a perturbation test.Jira issue: CRDB-44796