Page MenuHomePhabricator

Create reuse recipes for tendril/zarcillo/dbprov/backup hosts
Closed, ResolvedPublic


T251768: Make partman/custom/no-srv-format.cfg work covered all normal DB hosts, as they all use the same partitioning scheme. The remaining db hosts and all backup hosts also need reuse recipes to be crafted.

Event Timeline

Kormat triaged this task as Medium priority.Jun 18 2020, 12:38 PM
Kormat moved this task from Triage to Backlog on the DBA board.
Kormat renamed this task from Create reuse recipes for tendril/zarcillo/dbprov hosts to Create reuse recipes for tendril/zarcillo/dbprov/backup hosts.Jun 25 2020, 2:56 PM
Kormat updated the task description. (Show Details)

Note backup hosts are very unique ones- with different hardware (some have 1 or 2 external arrays of disks). I believe they were setup manually each one, as each one has different hardware and there is no standardization. That means there is really no recipe for 1st install (wipe). Aiming for "something that will not destroy data" would be enough, rather than a perfect recipe.

Change 608012 had a related patch set uploaded (by Kormat; owner: Kormat):
[operations/puppet@production] install_server: Reuse partitions for dbprov* hosts

tendril/zarcillo are probably unsupportable by reuse-parts. They have multiple mdraid arrays, and the documentation says that the naming of arrays is non-determinstic, so a recipe which mentions /dev/md0 could, in theory, match any of the arrays. No bueno.

Aand the same applies to the backup* hosts.

Change 608012 merged by Kormat:
[operations/puppet@production] install_server: Reuse partitions for dbprov* hosts /608012

As @jcrespo pointed out - we cannot currently test the dbprov reuse recipe, but the next time we're installing a new dbprov host we should test it then.


  • tendril/zarcillo: not currently feasible, and not worth the effort repartitioning them to make it feasible.
  • dbprov: done! (needs testing the next time a new dbprov host is imaged)
  • backup: not feasiable. reuse-parts would need to know which physical disks belong to which mdraid devices, and the naming of those mdraid devices, which would be a huge amount of work for a tiny number of hosts that are almost never reimaged.