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
Intensive write workload blocks PG instance because the speed of WAL generated is much higher than the amount of WALs that the page server can process.
We do have max_replication_write_lag to avoid having an infinite amount of WAL produced. However, when max_replication_write_lag is set to be high, performing a read will need to wait for all the WALs to be processed, even if none of the writes affected the read, because the system doesn't know that.
Resolves#1889.
This PR adds new tests to measure the WAL backpressure's performance under different workloads.
## Changes
- add new performance tests in `test_wal_backpressure.py`
- allow safekeeper's fsync to be configurable when running tests
Context
Intensive write workload blocks PG instance because the speed of WAL generated is much higher than the amount of WALs that the page server can process.
We do have
max_replication_write_lag
to avoid having an infinite amount of WAL produced. However, whenmax_replication_write_lag
is set to be high, performing a read will need to wait for all the WALs to be processed, even if none of the writes affected the read, because the system doesn't know that.This is the problem we encounter in #1763
Tasks
Related PRs/issues
The text was updated successfully, but these errors were encountered: