You’re 100% right.  So I ended-up discovering three things:

- The “new” hosts we’re migrating to that were having the problems had 1/4 the 
memory they ideally should have had
- The segmentation faults were caused by the Linux oom-killer coming in and 
killing the child processes to try and recover memory
- The application requires that much memory because of a 
previously-undiscovered memory leak in the application and what it stores in a 
set of in-memory dicts

Appreciate the help!

—



James Goin | Software Engineer, Liberator Squad

> On Jan 10, 2019, at 1:03 PM, Roberto De Ioris <[email protected]> wrote:
> 
> 
>> 
>> However, after moving from kernel version 4.4.0-134-generic over to
>> 3.13.0-164-generic, the segfaults stopped.  I’m wondering if there’s a
>> memory management bug somewhere in the mix.  Anyone have any ideas on how
>> I might be able to track this back and provide a definitive bug report?
>> —
>> 
>> 
> 
> Hi, can you post your uWSGI configuration ? those "scary" errors are
> generally triggered (unless there is a bug in uWSGI) from mis-usage of
> fork()'s copy on write
> 
> 
> -- 
> Roberto De Ioris
> https://urldefense.proofpoint.com/v2/url?u=http-3A__unbit.com&d=DwIGaQ&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=-TrLS3TpQre-21b9Hcu4pBUeXmQyQEOnP5OVIhuR78s&m=bFOJKaGNqg463DUGrHjfcflJRCoxxF3C1Xqz-NlbiD0&s=csgYXDyeeqiQwjpcaoj8tlcWr37gLZ6UGsHiS1I5too&e=
> _______________________________________________
> uWSGI mailing list
> [email protected]
> https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.unbit.it_cgi-2Dbin_mailman_listinfo_uwsgi&d=DwIGaQ&c=RoP1YumCXCgaWHvlZYR8PZh8Bv7qIrMUB65eapI_JnE&r=-TrLS3TpQre-21b9Hcu4pBUeXmQyQEOnP5OVIhuR78s&m=bFOJKaGNqg463DUGrHjfcflJRCoxxF3C1Xqz-NlbiD0&s=8MghID1ORnmmcOwce9sapqMz_Trp7V7XtFjKR5Afn_Y&e=

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

Reply via email to