fix: Keep working when EMFILE is reached

[steve-chavez]

Closes #2042.

TODO

• Admin docs about timeouts and EMFILE Add admin section on file descriptors postgrest-docs#504

[wolfgangwalther]

Do you think we can craft an io test that runs into the EMFILE problem without the patch?

[steve-chavez]

We'd be testing Warp if we do that, so I think we shouldn't. However, basically we need to dynamically change the file descriptor limit(can be done with prlimit) to something like 1, make requests and see postgrest fail.

[steve-chavez]

make requests and see postgrest fail.

Doing a prlimit --pid <pid> --nofile=10:1024 and making a new request:

• old version: postgrest dies and prints the Network.Socket.accept: resource exhausted (Too many open files) error.
• new version: new requests wait and when increasing the fds(prlimit --pid <pid> --nofile=1024:1024) they get a response.

(Ideally the EMFILE error would bubble up to the client instead of waiting but at least we keep running)

[steve-chavez]

The above looks like a recovery test, so I've listed it on #1766 to add it later if it's worth it.

[wolfgangwalther]

(Ideally the EMFILE error would bubble up to the client instead of waiting but at least we keep running)

Agree with both!