[masakari] Wallaby PTG Summary

Radosław Piliszek radoslaw.piliszek at gmail.com
Tue Nov 3 16:11:55 UTC 2020


Hello Everyone!

Masakari had 3 2-hour sessions and we managed to discuss a broad range
of proposals and assign tasks to interested parties.
(And this was my first time chairing a PTG as PTL).
Thank you, everyone who participated!

Here is a summary of the PTG discussions.
The whole discussion has been written down in the etherpad. [1]

We acknowledged that the project has lost its momentum during the last
cycle (Victoria) and we hope to restore it now.
To this end, we are reviving interesting old blueprints and specs.
Some of these are already at the implementation stage, ready for reviews.

Some of the discussed proposals fall into the 'low code impact, high
(positive) usability impact' category, the most prominent being ways
to have more flexibility in disabling/enabling the HA scope as well as
generally more friendly approach towards user errors (not evacuating
'pinned' instances).

On the other hand, some proposals involve deeper architectural changes
to Masakari, such as the shift from reliance on Pacemaker/Corosync
stack to Consul and/or etcd (possibly also for coordination), as well
as making Masakari inspect the host status via both inbound and
outbound means (including power management for checks and fencing). To
this end, we plan to investigate a possible integration with Ironic as
our gate to the baremetal world.

Another proposal worth noting is the Masakari 'restore previous state'
workflow, meant to help operators move instances back to their
original location once the local problem is solved.

Yet another proposal is increasing the resilience under the current
scope (Pacemaker/Corosync) by retrying host checks (several sampled
observations rather than one) and implementing adaptive rules to
consider large-scale failures.

We also discussed general maintenance tasks. We decided to give some
love to the documentation and make sure it's not self-contradictory
and also in agreement with the actual code, as well as include general
recommendation on how Masakari is supposed to be deployed to have it
work optimally.

Finally, we discussed a few miscellaneous proposals such as making
Masakari a better container world citizen by removing the reliance on
systemd.

[1] https://etherpad.opendev.org/p/masakari-wallaby-vptg

Kind regards,

-yoctozepto



More information about the openstack-discuss mailing list