Open nvanbenschoten opened 2 years ago
I'm a bit hesitant to remove batching entirely... We can do that by changing cluster setting. But in this PR https://github.com/cockroachdb/cockroach/pull/78564, perhaps we can keep batching w/out all of the negatives we observed.
Moving back to triage; I don't think we should do anything on this front right now. Feel free to close.
The job scheduler currently attempts to schedule up to 10 jobs in the same transaction. This is unnecessary from a throughput perspective, as job scheduling throughput is very low. Meanwhile, it increases the size of transactions, increasing the change of transaction retries. It also makes the job scheduler queries more complex and potentially susceptible to suboptimal query plans. Unless there's a strong reason to keep this batching, we should remove it.
cc. @miretskiy
Jira issue: CRDB-14206