[Labs-l] dns changes this week

Mon Jun 1 22:39:48 UTC 2015

The slow march to update labs dns will hit a few milestones this week.  
If all goes well, labs users won't notice anything at all. If we run 
into trouble there may be brief outages; if you notice an outage please 
notify me immediately.

2015-06-02:  Starting tomorrow, new instances will use the new dns 
system by default.

2015-06-02 through -05:  During the week, Yuvi will be moving a few 
critical hosts (proxies, mostly) over to the new system.

2015-06-09:  On Monday I'll switch remaining instances to the new system.

If you have an instance using self-hosted puppet, today is a good day to 
rebase your puppet code so that you're set up to cope with these 
changes.  Honestly, every day is a good day to rebase your puppet code :)

Details, for the technically interested:

During each of the above steps, sets of instances will be moved over to 
using a new dns server.  That server is named 
'labs-recursor0.wikimedia.org' and uses pdns-recursor. Pdns-recursor is 
backed in turn by a pdns implementation running on the same server.  The 
records in pdns are populated via openstack designate-sink, a tool which 
hooks instance creation and creates an A record.

Use of the old (dnsmasq) system is maintained via the puppet setting 
'use_dnsmasq'.  So when I say I'm 'moving' an instance, in reality I'm 
just removing the use_dnsmasq flag.

The new system supports old-school names (name.eqiad.wmflabs) as well as 
project-qualified names (name.projectname.eqiad.wmflabs) and also 
supports rudimentary reverse lookups.