Hi,

here's a minimal configuration where I can reproduce this:

error_log debug.log debug;

events {
worker_connections 1024;
}

http {
uwsgi_buffers 64 8k;

upstream nginx-test.uwsgi {
server 10.0.0.7:13003;
least_conn;
}

server {
listen 8080;
server_name nginx-test.com;

location /api/ {
include uwsgi_params;
uwsgi_pass nginx-test.uwsgi;
}
}
}

Here's a debug log covering server start, a single request that exhibits the problem, and server shutdown:

http://filebin.ca/1UClE4zzhfZe/debug.log.gz

Everything goes OK for a while, just a few stray mallocs, and then maybe half a minute into the request (the time varies), after maybe 20-25MB have been transferred, the flood of mallocs starts:

$ grep malloc: debug.log | cut -d ' ' -f 2 | uniq -c
4 14:34:51
1 14:34:52
3 14:34:56
1 14:34:59
1 14:35:03
2 14:35:12
1216 14:35:27
1135 14:35:28
2144 14:35:29
1996 14:35:30
520 14:35:31

(That last second of mallocs is only smaller because I stopped the client, so the request was aborted)

I hope that debug log turns up something informative --- and thank you again for your time on this.

Cheers,

Gulli