You are browsing a read-only backup copy of Wikitech. The primary site can be found at wikitech.wikimedia.org

Depooling servers

From Wikitech-static
Revision as of 15:31, 29 January 2016 by imported>Giuseppe Lavagetto (→‎cp*** machines: Add a note on the "depool" script)
Jump to navigation Jump to search

Contains various ways to depool different servers. In a glorious future, we'd just need to do conftool depool <service> <node> but we aren't there yet!

cp*** machines

These are the varnish frontends, and should be depooled via conftool. To fully depool a server, do as root on palladium:

confctl --tags dc=eqiad,cluster=<cluster-name>,service=varnish-be --action set/pooled=no <server-fqdn>
confctl --tags dc=eqiad,cluster=<cluster-name>,service=varnish-be-rand --action set/pooled=no <server-fqdn>
confctl --tags dc=eqiad,cluster=<cluster-name>,service=varnish-fe --action set/pooled=no <server-fqdn>
confctl --tags dc=eqiad,cluster=<cluster-name>,service=nginx --action set/pooled=no <server-fqdn>

You can find out <cluster-name> from looking at conftool-data/nodes/ in the operations/puppet repository.

An easier option is to ssh into the server and just run

$ depool

mw* machines (and others)

The mw* application servers and other servers managed via pybal/etcd are depooled by using conftool. This is documented at https://wikitech.wikimedia.org/wiki/LVS#Etcd_as_a_backend_to_Pybal_.28All_of_production.29