I think we're leaking sockets

[athornton]

I was getting quite a lot of

[pid: 7|app: 0|req: 6255/12909] 10.165.4.11 () {50 vars in 850 bytes} [Fri Feb 21 15:05:16 2025] POST /lsst-dm/testdata_image_cutouts/objects/batch => generated 3334 bytes in 832 msecs (HTTP/1.1 200) 2 headers in 85 bytes (1 switches on core 0)
Fri Feb 21 15:05:18 2025 - *** uWSGI listen queue of socket "127.0.0.1:39397" (fd: 3) full !!! (1025/1024) ***
Fri Feb 21 15:05:19 2025 - *** uWSGI listen queue of socket "127.0.0.1:39397" (fd: 3) full !!! (1025/1024) ***
Fri Feb 21 15:05:20 2025 - *** uWSGI listen queue of socket "127.0.0.1:39397" (fd: 3) full !!! (1025/1024) ***
Fri Feb 21 15:05:21 2025 - *** uWSGI listen queue of socket "127.0.0.1:39397" (fd: 3) full !!! (1025/1024) ***
Fri Feb 21 15:05:22 2025 - *** uWSGI listen queue of socket "127.0.0.1:39397" (fd: 3) full !!! (1025/1024) ***
Fri Feb 21 15:05:23 2025 - *** uWSGI listen queue of socket "127.0.0.1:39397" (fd: 3) full !!! (1025/1024) ***

at a listen queue of depth 100. Raising it to 1024 helped for a while, but it still ran out (see above). So I think there's something going on with not reclaiming sockets if connections time out, or something like that.

I realize this is very vague, but I wanted to open the issue so I don't forget about it.