uWSGI - unresponsive after reloading workers - uwsgi

Question

I'm trying to let uWSGI refresh its workers, because they are leaking a bit of memory. However, after reloading the workers it's no longer responsive. Am I forgetting something?
uwsgi --http-socket 0.0.0.0:8000 --wsgi-file entry.py --processes 3 --master --req-logger file:/log/reqlog --logger file:/log/errlog --harakiri 15 --max-requests 3
max-requests 3 is to test the reloading:
mapped 291072 bytes (284 KB) for 3 cores
*** Operational MODE: preforking ***
2018-02-01 13:31:04,416 root [INFO] Starting
WSGI app 0 (mountpoint='') ready in 2 seconds on interpreter 0x17d9dc0 pid: 1 (default app)
*** uWSGI is running in multiple interpreter mode ***
spawned uWSGI master process (pid: 1)
spawned uWSGI worker 1 (pid: 11, cores: 1)
spawned uWSGI worker 2 (pid: 12, cores: 1)
spawned uWSGI worker 3 (pid: 13, cores: 1)
flask#4078cdd3df37:/app$ curl localhost:8000
{"message": "ok"}
flask#4078cdd3df37:/app$ curl localhost:8000
{"message": "ok"}
flask#4078cdd3df37:/app$ curl localhost:8000
{"message": "ok"}
flask#4078cdd3df37:/app$ curl localhost:8000
{"message": "ok"}
flask#4078cdd3df37:/app$ curl localhost:8000
{"message": "ok"}
...The work of process 11 is done. Seeya!
flask#4078cdd3df37:/app$ curl localhost:8000
worker 1 killed successfully (pid: 11)
Respawned uWSGI worker 1 (new pid: 33)
{"message": "ok"}
flask#4078cdd3df37:/app$ curl localhost:8000
{"message": "ok"}
flask#4078cdd3df37:/app$ curl localhost:8000
{"message": "ok"}
...The work of process 13 is done. Seeya!
flask#4078cdd3df37:/app$ curl localhost:8000
{"message": "ok"}
...The work of process 12 is done. Seeya!
flask#4078cdd3df37:/app$ curl localhost:8000 --max-time 10
worker 3 killed successfully (pid: 13)
Respawned uWSGI worker 3 (new pid: 40)
worker 2 killed successfully (pid: 12)
Respawned uWSGI worker 2 (new pid: 43)
curl: (28) Operation timed out after 10001 milliseconds with 0 bytes received
flask#4078cdd3df37:/app$ curl localhost:8000 --max-time 10
curl: (28) Operation timed out after 10001 milliseconds with 0 bytes received
Eg uWSGI is no longer responding (connection is alive forever, unless i use curl --max-time). How does uWSGI communicate internally? How does the master process know how to reach the workers? I think something is going wrong there.

Answer 1

I ran into this this same issue. It appears that when master flag is unset, this issue goes away. For those using emperor, the 'master flag' is the one in the vassal configuration.