cache: move open to thread pool

Noam Cvikel noamc at qwilt.com
Thu Sep 2 16:16:55 UTC 2021


Hello,

Didn't see a way to get into the thread after joining the list, but
hopefully this still replies correctly.

First of all thank you for the prompt response and updated patch.
We implemented the updated async open patch, and started testing it in our
lab.
The static module part of the patch isn't updated yet, so we didn't apply
it.
While we're almost not using the static module, we'd still gladly test it
and share a feedback, once the patch is updated.

The patch itself is looking good. We've not encountered any errors or
issues, and we see a significant performance improvement.
We're working with spinning disks (HDD), and before applying the patch we
would sometimes see the main thread get stuck on open IO anywhere between
2-15 seconds.
After applying the patch, we see that an iteration of the worker's main
loop doesn't take more than 10ms! Significantly reducing the penalty for
other transactions.

We will keep monitoring the changes and report. We would really like to see
this patch integrated into the official branch. If we can do anything else
to help make this happen - please let us know.

Best,



*Noam Cvikel*
Qwilt | | noamc at qwilt.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.nginx.org/pipermail/nginx-devel/attachments/20210902/93f62dff/attachment.htm>


More information about the nginx-devel mailing list