php-fpm: fix Solaris port events.mechanism

[psumbera]

Closes GH-9959.

[devnexen]

So I gave a shot at your branch in my illumos instance, looks working to me. cc @bukka

[bukka]

How come this was not noticed during the previous testing? Is that some specific configuration or load that is needed to recreate the issue? Is the issue actually visible when running the tests (without this patch)? If not, we might need to add an extra test for this.

I might need to eventually do some testing myself so will try to setup OpenIndiana sometime soon.

[psumbera]

How come this was not noticed during the previous testing? Is that some specific configuration or load that is needed to recreate the issue? Is the issue actually visible when running the tests (without this patch)? If not, we might need to add an extra test for this.

It was my mistake. I didn't really know where the code was used. I just started php-fpm in background and did some testing. To stop it I simply used 'pkill php-fpm'. This killed all php-fpm processes.

Later I realized that some fpm tests were failing. Then I tried to run php-fpm in foreground and used control-c to stop it. It didn't work. Only first event was delivered. But the event about child exiting never was processed...

Now with the change there is no more problems with fpm tests (due still running main php-fpm process).

[bukka]

Ok makes sense.

I have done some deeper investigation into this and also needed to refresh how the whole thing work in detail. The port is the only one that needs it so I was wondering why. It's actually because it's one-shot only association which is actually confirmed here: https://docs.oracle.com/cd/E86824_01/html/E54766/port-associate-3c.html

When an event for a PORT_SOURCE_FD object is retrieved, the object no longer has an association with the port. The event can be processed without the possibility that another thread can retrieve a subsequent event for the same object. After processing of the file descriptor is completed, the port_associate() function can be called to reassociate the object with the port.

So yeah it needs to be re-associated and this change is correct. Just minor clean up would be great (adding comments for that)

[bukka]

If you want the change in PHP-8.0, please address the requested change and quickly re-test it ASAP as we have only 3 days of support there. It would be nice to merge it there as well as it's broken currently. Otherwise it will have to go only to 8.1+

[bukka]

Merged via 72da2b0 . Thanks!