Last modified: 2014-02-10 17:02:56 UTC
The Jenkins slaves on labs projects (deployment-prep and integration projects) use the user jenkins-deploy to setup the instance as a Jenkins slave. The LDAP entry points the home of that user to /home/jenkins-deploy which is a GlusterFS partition for the integration project. The slave jobs are run under /mnt/jenkins-workspace since we don't want the jobs workspaces to be on a shareable disks to avoid obvious race conditions. To avoid relying on GlusterFS, we should get jenkins-deploy home to point to /mnt/jenkins-deploy . That needs some puppet change to create the home directory then the LDAP entry can be altered.
Change 112458 had a related patch set uploaded by Hashar: contint: move jenkins-deploy user home to /dev/vdb https://gerrit.wikimedia.org/r/112458
Change 112458 merged by ArielGlenn: contint: move jenkins-deploy user home to /dev/vdb https://gerrit.wikimedia.org/r/112458
All instances now have a /mnt/home/jenkins-deploy homedir which belong to jenkins-deploy. The next step is to have the jenkins-deploy user home dir to be updated in LDAP to point to /mnt/home/jenkins-deploy Currently: # ldaplist -l passwd jenkins-deploy|grep homeDirectory homeDirectory: /home/jenkins-deploy # Then reconnect all Jenkins slaves and make sure jobs can still be triggered.
Apergos reviewed the puppet changes and Andrew updated the homeDirectory. I have rebooted most slave instance (beside deployment-parsoid2 and deployment-bastion). integration-slave01 Jenkins slave has reconnected and the npm command now use /mnt/home/jenkins-deploy for its cache thus solving the issue I had. Still have to double check the other jobs/slaves.
I have disconnected and reconnected all slaves. Bye bye GlusterFS.