Mryange opened a new pull request, #69245:
URL: https://github.com/apache/doris/pull/69245
## Proposed changes
Enable jemalloc's dedicated oversize arena for allocations at or above 64 MiB
by default. The setting is applied consistently to the normal BE launcher,
the sample BE configuration, Cloud startup, pipeline configurations, and the
related Cloud regression setup.
This keeps large transient allocations separate from per-CPU arenas, reducing
large-allocation fragmentation and the amount of allocator memory retained by
the BE process. The setting takes effect after a BE restart.
## Validation
- Verified all production and pipeline configuration entries use
`oversize_threshold:67108864`.
- Passed `bash -n bin/start_be.sh cloud/script/start.sh`.
- Passed `git diff --check`.
- On an ARM Neoverse-N2 host, compared disabled, 64 MiB, and 128 MiB settings
with a 30-minute four-client Doris workload.
- With 64 MiB, the sustained `cbq33` execute/fetch p50 improved by 7.4%, mean
sampled RSS decreased by 16.8%, minor faults per query decreased by 24.8%,
and system CPU per query decreased by 24.2% relative to the disabled
setting.
- Query result invariants, allocator configuration restoration, and service
health checks passed after the experiment.
## User impact
The default BE allocator configuration changes after upgrade. Existing
explicit `JEMALLOC_CONF` values continue to take precedence; a BE restart is
required for the default to take effect.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
