Page MenuHomePhabricator

Drop some Trusty permanent slaves from integration labs project
Closed, ResolvedPublic

Description

We have 15 Trusty instances each with 8 GBytes of RAM. Some jobs have been migrated to Nodepool trusty images already so we should be able to reclaim some of the instances.

They got spawned en masse back when we switched MediaWiki from PHP5.3 to PHP5.5 due to a surge of demand.

Seems we can drop at least half of them.

WARNING: due to labs capacity there is no return. Those instances will NOT be able to be recreated in a short term.

The Jenkins graph for executors business:

https://integration.wikimedia.org/ci/label/UbuntuTrusty/load-statistics?type=hour

graph-3.png (500×950 px, 33 KB)

Event Timeline

Restricted Application added subscribers: Zppix, Aklapper. · View Herald Transcript

The Jenkins graph above is average so it does not accomodate for spikes :(

I created some more graphs on https://grafana.wikimedia.org/dashboard/db/releng-zuul to shows the max wait and # of builds for each of the three labels we use on permanent slaves (DebianJessie, UbuntuPrecise and UbuntuTrusty).

We have spikes for jobs on UbuntuTrusty , but I am pretty sure they are due to a conflict with the Jenkins Throttle plugin and the Gearman plugin, were a build is incorrectly assigned on a node.

I am thus dropping a couple 8GBytes RAM nodes: integration-slave-trusty-1024 and integration-slave-trusty-1025

Mentioned in SAL [2016-07-07T13:10:38Z] <hashar> deleting integration-slave-trusty-1024 and integration-slave-trusty-1025 to free up some RAM. We have enough permanent Trusty slaves. T139535

They were both on labvirt1010 which recovered 16GBytes of memory :-]

labvirt1010-2x8GB_deletion.png (422×677 px, 17 KB)

2 big ones got dropped. That is good enough for now. More will be deleted as jobs are shifted to Nodepool instances.