mkustermann
commented
9 years ago cc: @whesse
We (Rico, Bill, ...) have discussed these things many times and the bottom line is: The best thing we could do is to have no bot modifications at all.
I think the right step is to work with chrome infra to get either our modifications into VM images & puppet configuration (disadvantage: very inflexible, takes a long time) or get chrome infra to get us a way / a hook for us to run certain things as root (installation, mounting, chroot, ...).
Bot modifications get lost if
- we get new bots
- if a bot gets suddenly re-imaged (as with the GCE VMs we got)
- puppet overrides our config
- we get new VMs
- we use a VM for a different builder, which needs more things
- ...
As for documentation, the slave*-m3 GCE VMs need the following additional steps at the time of this writing:
- boto file (for archiving of tarballs & coredumps)
- disable apport
- coredump sysctl entries
- install ARM cross compiler
- install libfdt for running our qemu in third_party
Some of our build-bot machines are GCE instances, and they get re-imaged once in awhile. This cause them to stop working as buildbot machines, as additional installs and configuration is required.
We should at least document the installation and configuration that is required, so it is not trial and error every time.
If at all possible we should automate it.
@kustermann can you please update this bug with the list of actions you performed on the re-imaged bots today?