Dockerized container consuming too much CPU resource

[kweng2]

Building docker image from source then running in container as suggested results in consuming ~100% of CPU:

docker build ./ -t tasks_emulator
docker run -p 8123:8123 tasks_emulator -host 0.0.0.0 -port 8123 -queue projects/dev/locations/here/queues/anotherq

I have not tried running this via go directly.

[aertje]

Hi @kweng2, I tried to reproduce this but haven't been able to, either in a docker container or running it straight up.
Do you observe this behaviour simply after starting the emulator (i.e. even when it's not executing any tasks or anything)?

[aertje]

If anyone else experienced this behaviour, can you please provide some more information under what conditions this was observed? Thanks!

[acoulton]

FWIW I haven't observed any issues. We have it running in our CI cluster under k8s, and monitoring is showing consistently tiny CPU usage for that pod. It's not handling a lot of tasks yet, it's a clean pod for every build and we purge the queue for each scenario, so it's maybe not 100% representative.

[Thammada]

I'm running it from docker-compose on macOS 11.0.1
For a single queue, the average CPU usage is around 10% for an idle queue (while not 100% as the OP, but I still think it's a bit too high for an idle queue).
For my application, I need 12 separate queues and the container uses around 120% CPU resource, which is a bit too high for local development.

[aertje]

Ok, I'm also seeing an increase in CPU consumption when i add a queue, but only about 1% per queue. At first I suspected the max concurrent dispatches (as it creates a go routine for each) but i created one with 1 million without any noticeable increase in CPU, so that does not seem to be it.
Will keep looking.

[aertje]

Possibly related: docker/for-mac#5044

[aertje]

I've merged @phalimee's fix; please let me know if you are seeing an improvement in CPU consumption with this @kweng2 and @Thammada.

[Thammada]

@aertje I'm observing an average of 0.00% CPU utilization on idle queues and around 1-3% for active queues 👍👏🎉

[aertje]

Thanks @Thammada, will close this issue then.