Pre-conference preparation breakage

09 Apr 2016

Just a quick post to help others who get stuck on the problem I solved today that could save a lot of time!


Of course it’s Saturday (what else is there to do?!), and there is a little over a week before I present on running a Galera Cluster MySQL on Kubernetes and while trying to record the demo, everything was apparently broken. Here’s the tale of woe in a nutshell.

Initial problem

So, when trying to run my kubernetes cluster, when running the following:

patg@kubernetes-master001:~$ ./kubectl exec pxc-node2-hjjrw -i -t -- mysql -u root -p -h pxc-cluster
Content-Type specified (plain/text) must be 'application/json'

Normally, this would bring me into an active MySQL client session. Instead, it would just hang. Figures. It worked a week ago, I swear!

Diagnosis and side-effect of cure

A bit of searching, seems it’s something that Docker changed in the client API recently (1.22) and requires a new kubernetes version. The version that originally broke was 1.0.6. An upgrade to 1.1.1. Still, no go. So finally, and upgraded to the latest 1.2.

The build script is useful for building binaries for Debian-based systems which I then build into debian packages:

~/kubernetes/cluster/ubuntu$ export KUBE_VERSION=1.2.2
~/kubernetes/cluster/ubuntu$ ./ 

The binaries are copied over to the minion VMs and restarted (kube-proxy and kubelet). They are not registering with the master, so the journal shows:

Apr 09 11:03:52 kubernetes-minion003 kubelet[13645]: I0409 11:03:52.213704   13645 kubelet.go:2365] skipping pod synchronization - [Failed to start ContainerManager system validation failed - Following Cgroup subsystem not mounted: [memory]]

Cure to side-effects of original cure

Not as bad as statin drug side-effects, but something that had to be fixed. The cause is that Kubernetes does a check for whether the memory cgroup is mounted. Looking at the source (Use the source!) line 128 of container manager in Kubernetes.

if expectedCgroups.Len() > 0 {
    return f, fmt.Errorf("%s - Following Cgroup subsystem not mounted: %v", localErr, expectedCgroups.List())

More searching, and found this page with a similar issue on lxc.

The fix for cgroups is:

  1. Add to /etc/fstab

     cgroup /sys/fs/cgroup cgroup defaults,blkio,net_cls,freezer,devices,cpuacct,cpu,cpuset,memory,clone_children 0 0
  2. Add to /etc/default/grub:

     GRUB_CMDLINE_LINUX_DEFAULT="quiet cgroup_enable=memory,namespace"
  3. Run update-grub
  4. Reboot

And this fixes the problem!

mysql-galera patg$ kubectl exec -it pxc-node3-iu89r -- mysql -u root -p -h pxc-cluster
Enter password: 
Welcome to the MySQL monitor.  Commands end with ; or \g.
Your MySQL connection id is 8
Server version: 5.6.24-72.2-56-log Percona XtraDB Cluster (GPL), Release rel72.2, Revision 43abf03, WSREP version 25.11, wsrep_25.11

Copyright (c) 2009-2015 Percona LLC and/or its affiliates
Copyright (c) 2000, 2015, Oracle and/or its affiliates. All rights reserved.

Oracle is a registered trademark of Oracle Corporation and/or its
affiliates. Other names may be trademarks of their respective

Type 'help;' or '\h' for help. Type '\c' to clear the current input statement.

mysql> show status like 'wsrep_cluster_size';
| Variable_name      | Value |
| wsrep_cluster_size | 3     |
1 row in set (0.00 sec)

Phew! I was seriously fit to be tied!

comments powered by Disqus