Page MenuHomePhabricator

Add a second Zuul merger instance on scandium.eqiad.wmnet
Closed, DeclinedPublic


When we have installed scandium.eqiad.wmnet T95046 we have provisioned it with two SSD. The original intent was to have two zuul-merger, each doing operating on a different SSD.

I have missed that during the installation. With the phase out of gallium, I would like to get rid of its zuul-merger and add a second one on scandium.eqiad.wmnet.

We would need to repartition the disk to have something like:

Mount pointDescription
/ssd with raid over first and second SSD
/srv/ssd1/No raid, first SSD
/srv/ssd2/No raid, second SSD

Currently we have:

$ cat /proc/mdstat 
Personalities : [raid1] 
md2 : active raid1 sda3[0] sdb3[1]
      145416192 blocks super 1.2 [2/2] [UU]
      bitmap: 0/2 pages [0KB], 65536KB chunk

md0 : active raid1 sdb1[1] sda1[0]
      9756672 blocks super 1.2 [2/2] [UU]
md1 : active (auto-read-only) raid1 sdb2[1] sda2[0]
      976320 blocks super 1.2 [2/2] [UU]
/dev/md0 on / type ext3 (rw,relatime,errors=remount-ro,data=ordered)
/dev/md2 on /srv/ssd type xfs (rw,noatime,nodiratime,attr2,nobarrier,inode64,logbufs=8,noquota)

So looks like we can get rid of md2 made of sda3 and sdb3 and repurpose those disk partition as:


Data on /srv/ssd/ can be ditched out entirely. Zuul merger will recreate them. We will want to stop the zuul-merger process.

Then we need some puppet work so we can get multiple zuul-merger instances on a single host with the associated git-daemon listening on different ports (which is also referenced in the zuul-merger.conf).

Event Timeline

Partitioning will depends on SRE , would need a different task.

Puppet work I am sure Release-Engineering-Team can handle it

Declining for now. Might revisit later on but really I have filled too many random tasks.

We have been with a single zuul-merger instance for months now, and nothing indicate it is causing a trouble right now.