[
https://issues.apache.org/jira/browse/MESOS-4947?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Joerg Schad updated MESOS-4947:
-------------------------------
Description:
TL;DR: In a newly created DCOS cluster with a running framework and actually
used dynamic reservations and persistent volumes the /slaves API does not list
the persistent volumes either (as described here:
https://github.com/apache/mesos/blob/master/docs/persistent-volume.md#listing-persistent-volumes).
Situation: There are Mesos agents in the cluster that have dynamic reservations
as well as persistent volumes for role "arangodb" with principal "arangodb" but
the corresponding framework does no longer exist (was "destroyed" by clicking
in the Marathon UI). Let's call these "Zombie persistent volumes". We try to
cleanup this mess manually (or automatically).
Effect: According to
https://github.com/apache/mesos/blob/master/docs/persistent-volume.md#listing-persistent-volumes
one should be able to list these zombies using
http://<mesos-master-url>/mesos/slaves JSON/REST endpoint. We see a summary of
the dynamic reservations, but the persistent disks do not appear. As a
consequence we can neither use the /destroy-volumes API nor the /unreserve API
do perform manual or automatic cleanup.
Additional information:
- If we start a new framework with role "arangodb" and principal "arangodb"
it will receive resource offers containing the dynamic reservations *and* the
persistent volumes.
- In a newly created DCOS cluster with a running framework and actually used
dynamic reservations and persistent volumes the /slaves API does not list the
persistent volumes either. So this might not be limited to zombie persistent
volumes.
was:
Situation: There are Mesos agents in the cluster that have dynamic reservations
as well as persistent volumes for role "arangodb" with principal "arangodb" but
the corresponding framework does no longer exist (was "destroyed" by clicking
in the Marathon UI). Let's call these "Zombie persistent volumes". We try to
cleanup this mess manually (or automatically).
Effect: According to
https://github.com/apache/mesos/blob/master/docs/persistent-volume.md#listing-persistent-volumes
one should be able to list these zombies using
http://<mesos-master-url>/mesos/slaves JSON/REST endpoint. We see a summary of
the dynamic reservations, but the persistent disks do not appear. As a
consequence we can neither use the /destroy-volumes API nor the /unreserve API
do perform manual or automatic cleanup.
Additional information:
- If we start a new framework with role "arangodb" and principal "arangodb"
it will receive resource offers containing the dynamic reservations *and* the
persistent volumes.
- In a newly created DCOS cluster with a running framework and actually used
dynamic reservations and persistent volumes the /slaves API does not list the
persistent volumes either. So this might not be limited to zombie persistent
volumes.
> Persistent volumes are not listed
> ---------------------------------
>
> Key: MESOS-4947
> URL: https://issues.apache.org/jira/browse/MESOS-4947
> Project: Mesos
> Issue Type: Bug
> Affects Versions: 0.27.1
> Environment: DCOS 1.6.1 on AWS
> Reporter: Max Neunhöffer
>
> TL;DR: In a newly created DCOS cluster with a running framework and actually
> used dynamic reservations and persistent volumes the /slaves API does not
> list the persistent volumes either (as described here:
> https://github.com/apache/mesos/blob/master/docs/persistent-volume.md#listing-persistent-volumes).
> Situation: There are Mesos agents in the cluster that have dynamic
> reservations as well as persistent volumes for role "arangodb" with principal
> "arangodb" but the corresponding framework does no longer exist (was
> "destroyed" by clicking in the Marathon UI). Let's call these "Zombie
> persistent volumes". We try to cleanup this mess manually (or automatically).
> Effect: According to
> https://github.com/apache/mesos/blob/master/docs/persistent-volume.md#listing-persistent-volumes
> one should be able to list these zombies using
> http://<mesos-master-url>/mesos/slaves JSON/REST endpoint. We see a summary
> of the dynamic reservations, but the persistent disks do not appear. As a
> consequence we can neither use the /destroy-volumes API nor the /unreserve
> API do perform manual or automatic cleanup.
> Additional information:
> - If we start a new framework with role "arangodb" and principal "arangodb"
> it will receive resource offers containing the dynamic reservations *and* the
> persistent volumes.
> - In a newly created DCOS cluster with a running framework and actually
> used dynamic reservations and persistent volumes the /slaves API does not
> list the persistent volumes either. So this might not be limited to zombie
> persistent volumes.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
