Without restart it will keep failing (not 100% of the time but frequently)
My logging level is error
Dec 8 08:04:08 srv103 imgproxy[4490]: embedded profile incompatible with image Dec 8 08:04:14 srv103 imgproxy[4490]: embedded profile incompatible with image Dec 8 08:04:31 srv103 imgproxy[12668]: embedded profile incompatible with image Dec 8 08:05:46 srv103 imgproxy[4490]: embedded profile incompatible with image Dec 8 08:05:56 srv103 imgproxy[4490]: embedded profile incompatible with image Dec 8 08:06:45 srv103 imgproxy[8904]: embedded profile incompatible with image Dec 8 08:06:51 srv103 imgproxy[4490]: embedded profile incompatible with image Dec 8 08:07:25 srv103 imgproxy[8904]: embedded profile incompatible with image Dec 8 08:08:09 srv103 imgproxy[4490]: embedded profile incompatible with image Dec 8 08:08:18 srv103 imgproxy[12668]: embedded profile incompatible with image Dec 8 08:08:24 srv103 imgproxy[12668]: embedded profile incompatible with image
That's the only thing I see in imgproxy logs
Earlier some of the problems were caused by : too large source images images with cmyk colorspace
Sergey Alexandrovich
@DarthSim
I can only guess one of your instances is OOMed. What's the exit code of the failing instance?
aleksystasiak
@aleksystasiak
no it is not oom
there is plenty of ram available
instance does not die, just spews couple od dozens of 504, works for a while and fail again
in the same time other instances work correctly
I was wandering if it's tied to default settings like concurrency and max clients as they were not touched
I'm allowing one instance to receive 32 simultainous connections
all instances are behind haproxy loadbalancer
(which is also caching but that is not connected I think)
The funny thing is, when I spot and restart misbehaving instance it will be healthy, but after some time one of 4 will start failing
aleksystasiak
@aleksystasiak
and cycle repeats
Sergey Alexandrovich
@DarthSim
504 means gateway timeout. Your LB doesn't receive the response from imgproxy in the expected time. What happens if you send a request to the fallen instance directly?
aleksystasiak
@aleksystasiak
When I spot another occurence I will try But I don't think it is caused by timeout (20 seconds read|write|download on imgproxy side 40 seconds is lb tolerance)
Sergey Alexandrovich
@DarthSim
I recommend setting imgproxy concurrency to num CPUs*2. You run 4 instances with 32 threads each, this means you need 64 cores to run smoothly.
Hi! Just to be clear — avif support hasn't been released yet? I remember reading about imgproxy supporting avif somewhere, probably on twitter, but I just tried with 2.15.0 and it doesn't know what .avif extension is.
Sergey Alexandrovich
@DarthSim
Hey! No, it hasn’t. You can use the latest image to play with avif
Ewan Higgs
@ehiggs
If no bg color is set when converging png with alpha channel to jpeg, is there a default colour, or is it arbitrary? It seems to be white but I have reports from the wild where it’s sometimes another color. (i know the solution is to set IMGPROXY_PRESET=default=bg:fffff just curious if I’m imagining the behaviour)
Sergey Alexandrovich
@DarthSim
@ehiggs By default, imgproxy flattens images with white when saving them to jpeg. If you have a sample image where it uses another color, I'd take a look