Last modified: 2013-04-18 18:43:51 UTC
From Jeff Green in RT #2524 Looking at srv248 and srv249: /dev/sda1 7.4G 5.8G 1.2G 83% / /dev/sda6 1.9G 126M 1.7G 8% /tmp /dev/sda7 59G 7.4M 59G 1% /a Now that we have heterogeous deployment, it's easy to blow up that first partition with multiple 1.1GB mw footprints plus rotated syslogs. Meanwhile we allocated a huge chunk of space to the unused /a partition. Is there a good reason for this scheme that I've missed? Otherwise maybe we should we do something more like: 10GB / 5 GB /tmp 5 GB /var/log * /a
We need to really do this for the scalars too. Switching to Swift will involve much more temp file creation.
I am wondering if there is a good reason for those partitions anyway. Filesystems support the reservation of space for the root user, and log rotation can (IIRC) be based on space usage. So why not use a single root partition?
[adding "ops" keyword as proposed by Greg]
This has already been done by notpeter. RT-2524 has more info on this and is resolved. summary is "done" (no /a on newly installed servers in eqiad) root@mw1024:~# mount | grep ^/ /dev/sda1 on / type ext4 (rw,errors=remount-ro) /dev/sda3 on /tmp type ext4 (rw)