[openstack-dev] [glance] PTG summary, Thursday Feb 23
Brian Rosmaita
rosmaita.fossdev at gmail.com
Fri Feb 24 13:41:17 UTC 2017
The Glance PTG schedule etherpad contains links to the etherpads for
each of the sessions discussed below:
https://etherpad.openstack.org/p/glance-pike-ptg-schedule
1. glance/glare/artifacts/images
The tldr; is Glance needs to complete image import before taking on
anything else, and the Glare project should continue to mature
independently. Even though there is some conceptual overlap between the
two projects, Glance needs to focus on providing the OpenStack Image
Service.
2. multi-store use cases and multiple image locations
The tldr; is that efficient use of multiple stores via multiple image
locations is a complicated problem that requires a complicated solution,
possibly a microservice to provide location sorting instead of trying to
shoehorn this into a custom location strategy in Glance. We're trying
to balance an effective strategy with a convenient way for operators to
modify strategies given environmental changes (e.g., a store goes offline).
3. organizing the Glance community
The consensus is that we need to introduce a few more events during the
cycle: a multi-day bug smash, virtual midcycle, and virtual operators'
midcycle (I mean "operators' virtual midcycle"). We're also going to
revive bi-weekly bug/patch reviews at a scheduled time TBD instead of
relying on everyone independently reviewing bugs/patches. We'll also
continue the operators' surveys, next one will be about using the Glance
cache.
4. eliminate redundant downloads
hemanthm outlined an idea to make downloads more efficient along with
some ideas of how to eliminate a single point of failure that
ultra-efficient downloads from the backend could introduce.
5. image import, continued
We discussed a few unclarities in the current spec; rosmaita will put up
a patch to clear these up.
6. atomicity in race conditions
dharinic and nikhil outlined some race conditions that can occur around
the change in image status in Glance. We discussed a few ways to fix
this; dharinic will explore these in order of complexity (least to most,
of course). The good news is that these are mostly benign, and the one
condition that could be problematic only occurs in a non-default,
non-recommended configuration.
We'll be discussing priorities, roadmapping for Pike, and holding an
Ocata postmortem on Friday morning.
cheers,
brian
More information about the OpenStack-dev
mailing list