SIGTERM to master, gevent http-socket workers will not re-start

[edevil]

Using the latest master branch, if I configure my gevent workers to use http-socket instead of having a fastrouter, they will not restart cleanly after a SIGTERM is sent to the master process:

*** running gevent loop engine [addr:0x46bcb0] ***
Traceback (most recent call last):
  File "core.pyx", line 420, in gevent.core.loop.io (gevent/gevent.core.c:7346)
  File "core.pyx", line 739, in gevent.core.io.__init__ (gevent/gevent.core.c:12782)
ValueError: fd must be non-negative: -1
Traceback (most recent call last):
  File "core.pyx", line 420, in gevent.core.loop.io (gevent/gevent.core.c:7346)
  File "core.pyx", line 739, in gevent.core.io.__init__ (gevent/gevent.core.c:12782)
ValueError: fd must be non-negative: -1
DAMN ! worker 1 (pid: 24079) died :( trying respawn ...
Respawned uWSGI worker 1 (new pid: 24098)
DAMN ! worker 2 (pid: 24080) died :( trying respawn ...
Respawned uWSGI worker 2 (new pid: 24099)

[unbit]

can you report the full uWSGI configuration ? thanks

[edevil]

uwsgi:
    uid: www-data
    gid: www-data
    module: discoapi.wsgi
    pythonpath: /servers/disco/sites/discoapi
    pythonpath: /servers/disco/discolib
    chdir: /servers/disco/sites/discoapi
    env: DJANGO_SETTINGS_MODULE=discoapi.settings
    processes: 2
    harakiri-verbose: 1
    auto-procname: 1
    no-orphans: 1
    vacuum: 1
    procname-prefix-spaced: discoapi
    master: 1
    listen: 2048
    virtualenv: /servers/python-environments/discosite
    limit-post: 2147483648
    http-timeout: 60
    gevent: 10000
    enable-threads: 1
    shared-socket: 0.0.0.0:80
    http-socket: =0
    buffer-size: 16384
    socket-timeout: 30
    add-header: Connection: close
    http-var: SERVER_PROTOCOL=HTTP/1.0
    cron: 0 -1 -1 -1 -1 /servers/conf/discoapi/clean_unused_blocks.sh
    cron: 30 -1 -1 -1 -1 /servers/conf/discoapi/check_old_revisions.sh
    cron: 15 3 -1 -1 -1 /servers/conf/discoapi/purge_zombies.sh
    threaded-logger: 1
    reload-on-rss: 500
    evil-reload-on-rss: 1000
    touch-chain-reload: /servers/conf/discoapi-chain-reload
    worker-reload-mercy: 120
    lazy-apps: 1
    memory-report: 1
    shared-import: discoapi.uwsgi_init
    ignore-sigpipe: 1
    ignore-write-errors: 1
    logformat: {RSS: %(rssM)MB} [pid: %(pid)|wid: %(wid)] %(addr) [%(ltime)] %(method) %(uri) => generated %(rsize) bytes in %(msecs) msecs (%(proto) %(status)) %(headers) headers in %(hsize) bytes (%(switches) switches on core %(core)) "%(uagent)"

[unbit]

ok, i was able to reproduce it, using :80 instead of 0.0.0.0:80 should work (while i fix it)