Especially since the beginning of November we can see that the system load spikes have gotten bigger.
Description
Details
Related Objects
Event Timeline
Seems to have gotten better since I fixed the big memory leak, but we'll have to wait and see.
The spikes in load average on thumbor machines keep happening, it correlates with high iowait too. The iowait doesn't seem related to swap activity, rather to application writes overloading the disks.
e.g.
$ sudo find /srv/thumbor/tmp/ -type f -ls | grep -i magi 11796587 169460 -rw------- 1 thumbor thumbor 173522752 Dec 12 18:52 /srv/thumbor/tmp/thumbor@8801/magick-2fGojBNXq9JkK 11796607 83192 -rw------- 1 thumbor thumbor 168633504 Dec 12 19:00 /srv/thumbor/tmp/thumbor@8801/magick-2KuUwLZ7ceX3S 11796621 169416 -rw------- 1 thumbor thumbor 173475840 Dec 12 18:54 /srv/thumbor/tmp/thumbor@8816/magick-2JlKQIpqDTGEW 11796740 33604 -rw------- 1 thumbor thumbor 171488096 Dec 12 18:59 /srv/thumbor/tmp/thumbor@8816/magick-2QT-pK5zCsNP4 11796692 167544 -rw------- 1 thumbor thumbor 171488096 Dec 12 19:00 /srv/thumbor/tmp/thumbor@8808/magick-2yWb4SZ2tyeuo 11796756 163748 -rw------- 1 thumbor thumbor 172824288 Dec 12 19:00 /srv/thumbor/tmp/thumbor@8837/magick-208oPYfR0ouIa 11796578 169480 -rw------- 1 thumbor thumbor 173462016 Dec 12 18:58 /srv/thumbor/tmp/thumbor@8837/magick-2RRNpe2roDP2f 11796667 164688 -rw------- 1 thumbor thumbor 168633504 Dec 12 18:57 /srv/thumbor/tmp/thumbor@8824/magick-2iwYvowKJWTk4 11796630 167628 -rw------- 1 thumbor thumbor 171639936 Dec 12 18:54 /srv/thumbor/tmp/thumbor@8833/magick-2Tuz8W3BuocOt 11796864 156624 -rw------- 1 thumbor thumbor 168633504 Dec 12 18:59 /srv/thumbor/tmp/thumbor@8834/magick-2kpWvkp4i18ip 11796803 166616 -rw------- 1 thumbor thumbor 170607424 Dec 12 18:52 /srv/thumbor/tmp/thumbor@8836/magick-2fDmqwCc8SCsl 11796577 29548 -rw------- 1 thumbor thumbor 171488096 Dec 12 19:00 /srv/thumbor/tmp/thumbor@8812/magick-2C_D1xW7n8Jou
root@thumbor1001:~# find /srv/thumbor/tmp/ -type f -ls | grep -i magi 19398878 28004 -rw------- 1 thumbor thumbor 171488096 Dec 12 18:55 /srv/thumbor/tmp/thumbor@8824/magick-29wQc9DEvgSIL 19399037 298436 -rw------- 1 thumbor thumbor 338000000 Dec 12 19:03 /srv/thumbor/tmp/thumbor@8832/magick-2IDwM7M1WfPea 19399001 167544 -rw------- 1 thumbor thumbor 171488096 Dec 12 19:03 /srv/thumbor/tmp/thumbor@8812/magick-2lh7Sok65bIaf 19398818 31188 -rw------- 1 thumbor thumbor 170607424 Dec 12 18:57 /srv/thumbor/tmp/thumbor@8833/magick-2p03qSUENZauW 19398826 169728 -rw------- 1 thumbor thumbor 173765696 Dec 12 18:55 /srv/thumbor/tmp/thumbor@8833/magick-27DfV-Xr96G2j 19399324 55664 -rw------- 1 thumbor thumbor 171488096 Dec 12 19:03 /srv/thumbor/tmp/thumbor@8806/magick-2kJhQlpFwIt2U 19399216 167520 -rw------- 1 thumbor thumbor 171488096 Dec 12 19:01 /srv/thumbor/tmp/thumbor@8808/magick-2z7Dr_YRDN1Sa 19398987 84432 -rw------- 1 thumbor thumbor 171488096 Dec 12 18:57 /srv/thumbor/tmp/thumbor@8826/magick-2t3QhovEAkzA2 19399257 136336 -rw------- 1 thumbor thumbor 170607424 Dec 12 18:59 /srv/thumbor/tmp/thumbor@8816/magick-23LPqFwpOMRD5 19398815 30332 -rw------- 1 thumbor thumbor 170607424 Dec 12 19:03 /srv/thumbor/tmp/thumbor@8829/magick-24XGeTlH6KdXE 19398752 133640 -rw------- 1 thumbor thumbor 171488096 Dec 12 19:03 /srv/thumbor/tmp/thumbor@8835/magick-2dtnKbkyHaF3J 19399326 167496 -rw------- 1 thumbor thumbor 171488096 Dec 12 19:01 /srv/thumbor/tmp/thumbor@8801/magick-28b9tez3e8qRj 19398783 50672 -rw------- 1 thumbor thumbor 171488096 Dec 12 19:03 /srv/thumbor/tmp/thumbor@8836/magick-2Rak7B42JrLPj 19399533 167488 -rw------- 1 thumbor thumbor 171488096 Dec 12 18:54 /srv/thumbor/tmp/thumbor@8836/magick-2921OB0phUD68 19399266 110592 -rw------- 1 thumbor thumbor 171488096 Dec 12 19:01 /srv/thumbor/tmp/thumbor@8805/magick-2JnOydF4QvNHf 19399099 168780 -rw------- 1 thumbor thumbor 172824288 Dec 12 18:53 /srv/thumbor/tmp/thumbor@8834/magick-2QLDfkVTqZnwA 19399033 169416 -rw------- 1 thumbor thumbor 173475840 Dec 12 18:54 /srv/thumbor/tmp/thumbor@8822/magick-2C5cp7R7batVF 19398973 29540 -rw------- 1 thumbor thumbor 171488096 Dec 12 19:03 /srv/thumbor/tmp/thumbor@8822/magick-2UYm55iFKJmqw 19399535 122528 -rw------- 1 thumbor thumbor 171488096 Dec 12 19:03 /srv/thumbor/tmp/thumbor@8837/magick-2Nzfg6Tjq7Zic
I haven't confirmed it but I'm assuming this traffic might be related to a single big file being requested. Things like T151066: Implement PoolCounter support in Thumbor or T151067: Implement rate limiter in Thumbor would help mitigating this I'd say
Change 328713 had a related patch set uploaded (by Filippo Giunchedi):
swift: drain thumbor traffic
Change 333222 had a related patch set uploaded (by Filippo Giunchedi):
swift: enable back shadow traffic to thumbor
Change 333222 merged by Filippo Giunchedi:
swift: enable back shadow traffic to thumbor
@fgiunchedi please open read rights for me on the nginx access logs on thumbor100*
This way I can look for some kind of pattern in the requests around the time the iowait spikes begin.
I'm closing this, as the load spikes have considerably lowered and now look reasonable compared to the amount of cores. Memory consumption halved as well since we deployed the subprocess-based IM engine.