Page MenuHomePhabricator

Fix dnsmasq in nova to not cache
Closed, DeclinedPublic

Description

Author: laner

Description:

Both dnsmasq and our recursors are caching lookups to private DNS. This
is really annoying when an instance is deleted, cached, then recreated,
because we need to SIGHUP dnsmasq and we need to wipe the cache on the
recursors to make the instance immediately available.
Either dnsmasq or our recursors should cache, not both.

Details

Reference
rt2111

Event Timeline

rtimport raised the priority of this task from to Medium.Dec 18 2014, 1:05 AM
rtimport added a project: ops-core.
rtimport set Reference to rt2111.

dvanliere wrote:

Hey Ryan,
Is this still an issue?
D

Both dnsmasq and our recursors are caching lookups to private DNS. This
is really annoying when an instance is deleted, cached, then recreated,
because we need to SIGHUP dnsmasq and we need to wipe the cache on the
recursors to make the instance immediately available.

Either dnsmasq or our recursors should cache, not both.

Status changed from 'new' to 'open' by RT_System

GIving this to Marc to follow up, since he's labs point person now.

On Thu Apr 17 07:09:53 2014, ariel wrote:

GIving this to Marc to follow up, since he's labs point person now.

There are other issues caused/made worse by dnsmasq and the entire thing needs
to be rethought. I'll keep this on my radar, but that's not going to be
anywhere sooner than Q3.

Yeah, I'm planning on creating a proper recursor this week that should fix most/all of those issues.

coren raised the priority of this task from Medium to High.Jan 5 2015, 2:37 PM
coren edited projects, added Cloud-Services; removed ops-core.
coren changed the visibility from "WMF-NDA (Project)" to "Public (No Login Required)".
coren changed the edit policy from "WMF-NDA (Project)" to "All Users".
coren set Security to None.
yuvipanda lowered the priority of this task from High to Medium.Jan 8 2015, 6:22 PM

Actually, *this* should be declined. It *wasn't* caching, and that almost killed it.

Recreating dead instances with same name should be rare enough for the 5min TTL to not be a big deal, I think.

If it ever turns out to be an issue, reducing the TTL slightly would be acceptable (but no lower than 2 min or so - it takes that long to build an instance anyways)