-
Improvement
-
Resolution: Unresolved
-
Major
-
Jenkins 2.73.3, docker-workflow-plugin 1.14, kubernetes 1.7.5
Since version 1.6 kubernetes is using a custom cgroup format:
Enhance Kubelet QoS:
Pods are placed under a new cgroup hierarchy by default. This feature requires draining and restarting the node as part of upgrades. To opt-out set --cgroups-per-qos=false.
The new cgroup looks something like "x:cpu:/kubepod/qos-level/container_id", here is an example from my kubernetes cluster:
cat /proc/9/cgroup 11:hugetlb:/kubepods/besteffort/pod60070ae4-c63a-11e7-92b3-0adc1ac11520/931fc44a96d53e4acbcbb037e22b7a5f55da235ca9b336105294d4f109cd8c59 10:perf_event:/kubepods/besteffort/pod60070ae4-c63a-11e7-92b3-0adc1ac11520/931fc44a96d53e4acbcbb037e22b7a5f55da235ca9b336105294d4f109cd8c59 9:net_cls,net_prio:/kubepods/besteffort/pod60070ae4-c63a-11e7-92b3-0adc1ac11520/931fc44a96d53e4acbcbb037e22b7a5f55da235ca9b336105294d4f109cd8c59 8:cpuset:/kubepods/besteffort/pod60070ae4-c63a-11e7-92b3-0adc1ac11520/931fc44a96d53e4acbcbb037e22b7a5f55da235ca9b336105294d4f109cd8c59 7:devices:/kubepods/besteffort/pod60070ae4-c63a-11e7-92b3-0adc1ac11520/931fc44a96d53e4acbcbb037e22b7a5f55da235ca9b336105294d4f109cd8c59 6:pids:/kubepods/besteffort/pod60070ae4-c63a-11e7-92b3-0adc1ac11520/931fc44a96d53e4acbcbb037e22b7a5f55da235ca9b336105294d4f109cd8c59 5:memory:/kubepods/besteffort/pod60070ae4-c63a-11e7-92b3-0adc1ac11520/931fc44a96d53e4acbcbb037e22b7a5f55da235ca9b336105294d4f109cd8c59 4:cpu,cpuacct:/kubepods/besteffort/pod60070ae4-c63a-11e7-92b3-0adc1ac11520/931fc44a96d53e4acbcbb037e22b7a5f55da235ca9b336105294d4f109cd8c59 3:freezer:/kubepods/besteffort/pod60070ae4-c63a-11e7-92b3-0adc1ac11520/931fc44a96d53e4acbcbb037e22b7a5f55da235ca9b336105294d4f109cd8c59 2:blkio:/kubepods/besteffort/pod60070ae4-c63a-11e7-92b3-0adc1ac11520/931fc44a96d53e4acbcbb037e22b7a5f55da235ca9b336105294d4f109cd8c59 1:name=systemd:/kubepods/besteffort/pod60070ae4-c63a-11e7-92b3-0adc1ac11520/931fc44a96d53e4acbcbb037e22b7a5f55da235ca9b336105294d4f109cd8c59
The matcher does not recognize this as a cgroup of running in a container and the docker plugin behaves as if Jenkins is not running in a container, i.e. volume mounts fail.
It would be awesome if the matcher could be improved to recognize this so we do not have to run custom options on the kubelet just for Jenkins.
- is related to
-
JENKINS-48502 image.inside stopped working on ECS agents
- Closed