About using mg_mgr_poll cause FD_SETSIZE overflow

[BaronStack]

os : centos 7.8
purpose: mongoose mg_mgr_poll for http_server

Our http server process implemented with mg_mgr_poll function.
We can see the mg_mgr_poll function is using select system call to deal the new connection.
But our process have more than 1024 files when running, so the select FD_SET is overflow.

Is there any reason for you to use the select system call rather than poll / epoll?
Do you have the plan to support it ?

[cpq]

There are no plans to use poll/epoll.
Search how to increase FD_SETSIZE in your environment.

[vkoskiv]

Hey! I started using mongoose for a hobby project a while back, and it's been working great so far. I'm a bit bummed about this FD_SETSIZE limit that I didn't anticipate when I started. I've considered making a downstream fork that uses poll(), but before I start investigating that, I'd love to hear more about what drove this decision on your end - some technical limitation or just a desire to keep the code simple?

I saw #439 from a long time ago where using poll() seemed to be on the table, and PR #508 introduced it, but I couldn't find the discussion/rationale for removing it later.
In any case, thanks for this project! Excellent documentation and really nice API design.

[vkoskiv]

@cpq just in case my reply didn't ping you.