| Gehel added a comment. |
In T200563#4623531, @Smalyshev wrote:Great work!
Thanks (I'll forward to @Volans)
I am not sure though why logging would be that much of an issue, shouldn't the log code take care of batching it, etc.? As for not logging nginx - do we have these logs somewhere else? If yes, then I guess we can stop that. We could probably tune Kafka logging too - right now it's kinda verbose and we most likely don't want that in the logstash.
nginx sends logs over syslog, so one UDP packet per log message. Honestly, I'm not sure it will make much difference. The dropped packets are on RX, and nginx/syslog should be only TX. They still compete for the same CPU, so that could help.
In the end, the problem is CPU contention, so anything that can reduce it is good to take. Nginx logs is the low hanging fruit. Spreading NIC related workload to more CPU should be doable. And making sure that Blazegraph CPU consumption is bounded is what we should really address (but probably quite a bit more complex).
As far as I understand, access logs should be available via varnish kafka, in the analytics cluster / hive.
Cc: gerritbot, Volans, Stashbot, Gehel, Aklapper, Smalyshev, Nandana, NebulousIris, Gaboe420, Versusxo, Majesticalreaper22, Giuliamocci, Adrian1985, Cpaulf30, Lahi, Gq86, Baloch007, Darkminds3113, Bsandipan, Lordiis, Lucas_Werkmeister_WMDE, GoranSMilovanovic, Adik2382, Th3d3v1ls, Ramalepe, Liugev6, QZanden, EBjune, merbst, LawExplorer, Lewizho99, Maathavan, Jonas, Xmlizer, jkroll, Wikidata-bugs, Jdouglas, aude, Tobias1984, Manybubbles, Mbch331
_______________________________________________ Wikidata-bugs mailing list Wikidata-bugs@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/wikidata-bugs
