> On Sat, Mar 14, 2015 at 1:09 PM, Roberto De Ioris <[email protected]>
> wrote:
>>
>>> I have setting "harakiri = 60" set.  I'm using ugreen so I have
>>> multiple async cores per worker.  I'm guessing there's separate timers
>>> per "core", but it seems like it kills the whole worker process.
>>> Maybe it's not practical to just kill a core and context switch out,
>>> or maybe it's trying that and failing, I'm not really sure how it's
>>> working.  However, is there any way to tell which core's timer tripped
>>> the harakiri?  I suspect it's which ever line has the oldest "since
>>> [timestamp]"?
>>>
>>> Here's my logs:
>>>
>>> Fri Mar 13 16:23:28 2015 - *** HARAKIRI ON WORKER 2 (pid: 20648, try:
>>> 1)
>>> ***
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI !!! worker 2 status !!!
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 178] 208.90.100.6 - POST
>>> /walldisplay_json?timestamp=2015-03-13T16%3A19%3A14.329000%2B0000&count=3
>>> since 1426263775
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 179] 208.90.101.56 - POST
>>> /walldisplay_json?timestamp=2015-03-13T16%3A23%3A11.076000%2B0000&count=3
>>> since 1426263793
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 183] 130.63.114.112 - POST
>>> /sitrep_json?timestamp=2015-03-13T16%3A22%3A43.778000%2B0000&count=3
>>> since 1426263772
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 185] 208.90.101.206 - POST
>>> /walldisplay_json?timestamp=2015-03-13T16%3A21%3A51.968000%2B0000&count=3
>>> since 1426263769
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 186] 208.90.100.6 - POST
>>> /walldisplay_json?timestamp=2015-03-13T14%3A13%3A43.594000%2B0000&count=1
>>> since 1426263768
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 187] 208.90.101.12 - POST
>>> /walldisplay_json?timestamp=2015-03-13T16%3A21%3A16.535000%2B0000&count=3
>>> since 1426263781
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 188] 208.90.100.6 - POST
>>> /walldisplay_json?timestamp=2015-03-13T16%3A19%3A14.329000%2B0000&count=3
>>> since 1426263772
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 189] 130.63.114.112 - POST
>>> /login since 1426263808
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 191] 208.90.101.12 - POST
>>> /walldisplay_json?timestamp=2015-03-13T16%3A19%3A04.271000%2B0000&count=3
>>> since 1426263795
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 192] 208.90.101.206 - POST
>>> /walldisplay_json?timestamp=2015-03-13T16%3A23%3A20.857000%2B0000&count=1
>>> since 1426263801
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 193] 208.90.101.192 - POST
>>> /walldisplay_json?timestamp=2015-03-13T14%3A13%3A43.594000%2B0000&count=1
>>> since 1426263771
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 194] 208.90.101.192 - POST
>>> /walldisplay_json?timestamp=2015-03-13T16%3A23%3A02.528000%2B0000&count=3
>>> since 1426263784
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 195] 208.90.101.192 - POST
>>> /walldisplay_json?timestamp=2015-03-13T16%3A09%3A17.970000%2B0000&count=3
>>> since 1426263773
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 196] 208.90.101.12 - POST
>>> /walldisplay_json?timestamp=2015-03-13T16%3A21%3A16.535000%2B0000&count=3
>>> since 1426263780
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI [core 198] 208.90.100.6 - POST
>>> /walldisplay_json?timestamp=2015-03-13T16%3A22%3A43.778000%2B0000&count=3
>>> since 1426263771
>>> Fri Mar 13 16:23:28 2015 - HARAKIRI !!! end of worker 2 status !!!
>>> DAMN ! worker 2 (pid: 20648) died, killed by signal 9 :( trying respawn
>>> ...
>>> Respawned uWSGI worker 2 (new pid: 21832)
>>
>> Hi, per-core harakiri is a 2.1 thing (you can take it from github
>> master)
>>
>
> Since the oldest running core is 40 seconds and I set it to 60
> seconds...  what could have caused this?  Or is my interpretation of
> the "since 1426..." wrong?
> _______________________________________________
>


In pre-2.1 releases every new activated core overwrites the harakiri
memory area of the previous one. So every usage of harakiri in
multithreaded or async mode is unreliable


-- 
Roberto De Ioris
http://unbit.com
_______________________________________________
uWSGI mailing list
[email protected]
http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi

Reply via email to