Why is apache + nginx in image processing making 500 mistakes?
-
There's a php crypt that handles a large number of images (resays, windmarks, etc.). The problem is that the 40-minute eclipt of work makes 500 mistakes, and then, with each later launch, falls back early, even though the time is about an hour. It's a logic. upstreamly closed connection while reading response header from upstream
Apache
- max_execution_time 3600
- max_input_time 3600
- upload_max_filesize 512M
- post_max_size 512M
- memory_limit 512M
Nginx
- proxy_connect_timeout 3600s
- proxy_send_timeout 3600s
- proxy_read_timeout 3600s
- proxy_buffer_size 512k
- proxy_buffers 512 4096k
- proxy_busy_buffers_size 4096k
- proxy_temp_file_write_size 4096k
- client_max_body_size 30m
- client_body_buffer_size 4096k
- large_client_header_buffers 8 32k
In the violin itself, after processing each catalogue, the disc cash is cleared. How can this problem be solved without resorting to the kroner and/or launching the crypt directly on the server?
-
It turns out the apache had eaten the whole operation, but he didn't write anything in the log, and he wrote nginx. gc_collect_cycles() fixed the situation.
