Revision b8189466
Added by Evgeni Golov about 1 year ago
| templates/pulpcore-api.service.erb | ||
|---|---|---|
|
WorkingDirectory=<%= scope['pulpcore::user_home'] %>
|
||
|
RuntimeDirectory=pulpcore-api
|
||
|
ExecStart=/usr/libexec/pulpcore/gunicorn pulpcore.app.wsgi:application \
|
||
|
--preload \
|
||
|
--timeout <%= scope['pulpcore::api_service_worker_timeout'] %> \
|
||
|
--workers <%= scope['pulpcore::api_service_worker_count'] %> \
|
||
|
--access-logfile - \
|
||
| templates/pulpcore-content.service.erb | ||
|---|---|---|
|
WorkingDirectory=<%= scope['pulpcore::user_home'] %>
|
||
|
RuntimeDirectory=pulpcore-content
|
||
|
ExecStart=/usr/libexec/pulpcore/gunicorn pulpcore.content:server \
|
||
|
--preload \
|
||
|
--timeout <%= scope['pulpcore::content_service_worker_timeout'] %> \
|
||
|
--worker-class 'aiohttp.GunicornWebWorker' \
|
||
|
--workers <%= scope['pulpcore::content_service_worker_count'] %> \
|
||
Also available in: Unified diff
Fixes #36437 - preload pulpcore API and content code
this reduces the overall memory usage by keeping only one copy of the
code, instead of N (= number of workers).
unscientific benchmark says that with 5 API and 17 content workers, this
reduces the memory footprint from 500MiB to 125MiB and from 1.5GiB to
500MiB respectively.
those numbers are obviously before any actual work is executed and
memory consumed by the app itself.