mirror of
https://github.com/neondatabase/neon.git
synced 2026-05-28 10:30:40 +00:00
979fa0682b
The batching perf test workload is currently read-only sequential scans. However, realistic workloads have concurrent writes (to other pages) going on. This PR simulates concurrent writes to other pages by emitting logical replication messages. These degrade the achieved batching factor, for the reason see - https://github.com/neondatabase/neon/issues/10765 PR - https://github.com/neondatabase/neon/pull/11494 will fix this problem and get batching factor back up. --------- Co-authored-by: Vlad Lazar <vlad@neon.tech>
How to reproduce benchmark results / run these benchmarks interactively.
- Get an EC2 instance with Instance Store. Use the same instance type as used for the benchmark run.
- Mount the Instance Store =>
neon.git/scripts/ps_ec2_setup_instance_store - Use a pytest command line (see other READMEs further up in the pytest hierarchy).
For tests that take a long time to set up / consume a lot of storage space,
we use the test suite's repo_dir snapshotting functionality (from_repo_dir).
It supports mounting snapshots using overlayfs, which improves iteration time.
Here's a full command line.
RUST_BACKTRACE=1 NEON_ENV_BUILDER_USE_OVERLAYFS_FOR_SNAPSHOTS=1 DEFAULT_PG_VERSION=16 BUILD_TYPE=release \
./scripts/pytest test_runner/performance/pageserver/pagebench/test_pageserver_max_throughput_getpage_at_latest_lsn.py