Dind image seems broken

[0xmichalis]

https://ci.openshift.redhat.com/jenkins/job/test_pull_request_origin_extended_networking_minimal/3042/consoleFull#-14181563465898c58db7602c31c0eab717

[INFO] Starting 'networking' extended tests
[INFO] [CLEANUP] Cleaning up temporary directories
[INFO] Building docker-in-docker images
Building container images
/data/bin/imagebuilder
--> Image fedora:25 was not found, pulling ...
--> Pulled 0/2 layers, 4% complete
--> Pulled 1/2 layers, 50% complete
no such image
[ERROR] PID 6395: hack/common.sh:750: `imagebuilder -t 'openshift/dind' '/data/src/github.com/openshift/origin/images/dind/'` exited with status 1.

@openshift/networking

[smarterclayton]

It's not fully blocking queue, some are still merging

[bparees]

saw it here too
https://ci.openshift.redhat.com/jenkins/job/test_pull_request_origin_extended_networking_minimal/3055/

[0xmichalis]

Removing P0 since it's not blocking but adding P1 since it happens fairly often since yesterday.

[danwinship]

It looks like it's failing while pulling the fedora:25 image, which shouldn't happen. Maybe some transient docker hub problem. Let's see if it goes away by itself...

[danwinship]

no, looks like this is new since #14522. @stevekuznetsov

[stevekuznetsov]

@smarterclayton any clue why imagebuilder would fail to pull fedora:25?

[stevekuznetsov]

Haven't seen this since we systemctl restart docker in openshift-eng/aos-cd-jobs#330 ... so it's definitely a memory hog from the daemon issue. /cc @runcom we restart the daemon after instances of docker cp and docker build now since it's holding on to too much memory, and we see no adverse effects. Especially with docker cp once the client call is done the memory stays in use for a while. Unclear what the daemon is doing with it.