Skip to content
This repository was archived by the owner on Dec 5, 2017. It is now read-only.

etcd_proxy consuming unallocated port in k8s/dcos #790

Closed
jdef opened this issue Feb 18, 2016 · 3 comments
Closed

etcd_proxy consuming unallocated port in k8s/dcos #790

jdef opened this issue Feb 18, 2016 · 3 comments

Comments

@jdef
Copy link

jdef commented Feb 18, 2016

user reported port conflict on a slave node because etcd_proxy was consuming a port that hadn't been properly allocated:

E0217 13:27:18.112458 6688 manager.go:1864] Failed to create pod infra container: API error (500): Cannot start container 4c74ded9de8418510f0c7966faef1a5adc834029b3e35829744bdcf6ee783a47: failed to create endpoint k8s_POD.e67a4f09_pod-38_default_50f9670b-d5a3-11e5-8fb3-005056ad6f59_e60cef0d on network bridge: Bind for 0.0.0.0:1052 failed: port is already allocated 
@jdef
Copy link
Author

jdef commented Feb 20, 2016

running k8s-mesos with etcd-mesos I see the following:

core@ip-10-0-2-150 ~ $ sudo netstat -natlp |grep -e etcd|grep -e LISTEN
tcp        0      0 127.0.0.1:4001          0.0.0.0:*               LISTEN      1933/etcd
tcp        0      0 10.0.2.150:1025         0.0.0.0:*               LISTEN      1264/./etcd
tcp        0      0 10.0.2.150:1026         0.0.0.0:*               LISTEN      1264/./etcd
tcp        0      0 10.0.2.150:38594        0.0.0.0:*               LISTEN      1250/./etcd-mesos-e
tcp6       0      0 :::1027                 :::*                    LISTEN      1250/./etcd-mesos-e
  • etcd-mesos isn't properly allocating a driver port from the resource pool
  • the etcd proxy running in the k8sm uber container is using local port 4001, not properly allocating that from the resource pool; since the container uses HOST networking mode this could create conflicts

@jdef
Copy link
Author

jdef commented Feb 20, 2016

going to change the k8s/dcos scheduler docker container to run in BRIDGE mode, requires some changes in the scheduler driver setup so that the correct IP is published to mesos, namely advertising an IP that matches that of LIBPROCESS_IP (if present)

upstream PRs:

there's also some packaging changes required for this:

@jdef
Copy link
Author

jdef commented Feb 20, 2016

cherry-picked into v0.7.3, removing cherry-pick label

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Projects
None yet
Development

No branches or pull requests

1 participant