Imarlier added a comment.

@BBlack @ema Couple of questions for you about Nginx:

  • Do we have nginx configured to handle a specific number of requests on a given worker process/thread, and then shut that down?
  • If it possible for nginx to be restarted (interrupting existing persistent connections) due to config updates or the like, and if so, is there a record of times when that has happened?

What I'm poking at is the issue described here: https://williamsbdev.com/posts/no-http-response-exceptions/.

The Wikidata updater process has a pool of connections to the text cache nodes, eg

imarlier@wdqs1003:~$ sudo netstat -anet | grep 208.80.154.224
tcp6       0      0 10.64.0.14:50536        208.80.154.224:443      ESTABLISHED 499        283749063
tcp6       0      0 10.64.0.14:50278        208.80.154.224:443      ESTABLISHED 499        283748769
tcp6       0      0 10.64.0.14:51154        208.80.154.224:443      ESTABLISHED 499        283750591
tcp6       0      0 10.64.0.14:50566        208.80.154.224:443      ESTABLISHED 499        283733664
tcp6       0      0 10.64.0.14:50564        208.80.154.224:443      ESTABLISHED 499        283749076
tcp6       0      0 10.64.0.14:51184        208.80.154.224:443      ESTABLISHED 499        283738772
tcp6       0      0 10.64.0.14:51648        208.80.154.224:443      ESTABLISHED 499        283750982
tcp6       0      0 10.64.0.14:50794        208.80.154.224:443      ESTABLISHED 499        283749269
tcp6       0      0 10.64.0.14:51662        208.80.154.224:443      ESTABLISHED 499        283750987
tcp6       0      0 10.64.0.14:50796        208.80.154.224:443      ESTABLISHED 499        283738743

The implication of the article linked above is that if the process that represents the remote side of these connections were to shut down while the connection itself is idle, it's possible that the socket would enter a CLOSE_WAIT state on the local side, and that in turn would result in the NoHttpResponseException that we're seeing in this case.

I'm doing some super super naive testing of this hypothesis on one of the wdqs hosts, but I'm curious whether it strikes you as being possible based on what we see here.


TASK DETAIL
https://phabricator.wikimedia.org/T207718

EMAIL PREFERENCES
https://phabricator.wikimedia.org/settings/panel/emailpreferences/

To: Imarlier
Cc: BBlack, ema, Gehel, Aklapper, Smalyshev, Legado_Shulgin, Nandana, thifranc, AndyTan, Davinaclare77, Qtn1293, Imarlier, Lahi, Gq86, Lucas_Werkmeister_WMDE, GoranSMilovanovic, Th3d3v1ls, Hfbn0, QZanden, EBjune, merbst, LawExplorer, Vali.matei, Zppix, _jensen, D3r1ck01, Jonas, Xmlizer, Wong128hk, jkroll, Wikidata-bugs, Jdouglas, aude, Tobias1984, Manybubbles, faidon, Mbch331, Jay8g, fgiunchedi
_______________________________________________
Wikidata-bugs mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/wikidata-bugs

Reply via email to