807415 - puppet should setup directory structure on foopies

Status: RESOLVED INVALID

(Infrastructure & Operations Graveyard :: CIDuty, task)

Description

[Kim Moir [:kmoir] ET]

Right now we just run a script after the foopies have been puppetized to create the dirs for the pandas, would be nice if this was puppetized.

Comment 1

[Kim Moir [:kmoir] ET]

Attachment: patch

This is just a better shell script than the one that currently exists.  For example

./create_dirs.sh 687 697 buildbot-master41 panda

Still working out the details to run this through puppet.

Comment 2

[Justin Wood (:Callek)]

Comment on attachment 683163 [details] [diff] [review]
patch

>diff --git a/buildfarm/mobile/create_dirs.sh.in b/buildfarm/mobile/create_dirs.sh.in
>--- a/buildfarm/mobile/create_dirs.sh.in
>+++ b/buildfarm/mobile/create_dirs.sh.in
>@@ -1,33 +1,59 @@
> #!/bin/bash
> 
>-tegras="sedTEGRALISTsed"
>+#script to run as root on the foopies to setup device dirs
> 
>-echo "setting up tegra directories"
>+if [ $# != 4 ]; then
>+    echo "usage: create_dirs.sh start end master [panda|tegra]"
>+    exit
>+fi

Suggestion, use python+json.tool+devices.json to extract a useable list based on the foopy?

Comment 3

[Kim Moir [:kmoir] ET]

Also, puppet should configure logrotate on the foopies to only store a fixed number of log files, right now their are too many

Comment 4

[Justin Wood (:Callek)]

Comment on attachment 683163 [details] [diff] [review]
patch

Review of attachment 683163 [details] [diff] [review]:
-----------------------------------------------------------------

f+, which for this, is enough to land it. However see my comment earlier about using devices.json to get the list of devices to create. Since always assuming a contiguous number will prove flawed (especially with tegras)

Comment 5

[Kim Moir [:kmoir] ET]

Yes, I'm going to look at parsing devices.json and using puppet to set this up, but I have to fix some other bugs first.

Comment 6

[Chris Cooper [:coop] (he/him)]

Callek: did your (or Pete's) recent work with the foopies already fix this?

Comment 7

[Justin Wood (:Callek)]

(In reply to Chris Cooper [:coop] from comment #6)
> Callek: did your (or Pete's) recent work with the foopies already fix this?

No its still a todo, just not urgent at this point

Comment 8

[Kim Moir [:kmoir] ET]

This should be easier now that mobile devices are managed by slavealloc.  We just need to create the directory name, not additional files in this dir.

Comment 9

[Chris Cooper [:coop] (he/him)]

I've talked to Callek about this recently, but I'm still not sure how exactly we want this to work. 

I think we want to add foopies to slavealloc, and then periodically poll slavealloc to see which mobile devices a particular foopy should have attached, possibly moving old devices aside, and setting up new device dirs as required. 

Since foopies *should* have long uptimes, we should treat puppet-on-foopies the same way we do puppet-on-masters and run it on some regular cadence. Under that model, there's no reason to implement another periodic check for which mobile devices should be attached to a given foopy, we should just do it from puppet.

Lots of intervening steps to get to that model though.

Comment 10

[Kim Moir [:kmoir] ET]

Think this can be closed since pandas are on there way out in bug 1186615