Last modified: 2014-04-11 22:27:44 UTC
To make the beta varnish instances as close to production as possible, we should use an XFS disk to hold the cached data. Mark proposed to craft some image under /dev/vdb and do a loopback mount of them where the production manifests expect them (aka /srv/sda3 , /srv/sdb3 ). /dev/vdb should be mounted somewhere, a file of a given size is created there using dd and then we can mkfs it and finally have puppet mount it with something like: mount -t xfs /some/file.img /srv/sda3 -o loopback,... That will let us get rid of some labs exception in the puppet manifests.
Change 119534 had a related patch set uploaded by Hashar: Support eqiad labs secondary disk https://gerrit.wikimedia.org/r/119534
The EQIAD varnishes comes with XFS, the Gerrit change #119534 is applied on the local puppetmaster there. Keeping the bug open until beta cluster got migrated and we verified all varnishes are indeed using XFS.
Change 119534 merged by Ottomata: Support eqiad labs secondary disk https://gerrit.wikimedia.org/r/119534