[openstack-dev] [all] DevStack switching from MySQL-python to PyMySQL

Sean Dague sean at dague.net
Mon Jun 15 11:30:30 UTC 2015


On 06/11/2015 06:29 AM, Sean Dague wrote:
> On 06/09/2015 06:42 PM, Jeremy Stanley wrote:
>> As discussed in the Liberty Design Summit "Moving apps to Python 3"
>> cross-project workshop, the way forward in the near future is to
>> switch to the pure-python PyMySQL library as a default.
>>
>>     https://etherpad.openstack.org/p/liberty-cross-project-python3
>>
>> To that end, support has already been implemented and tested in
>> DevStack, and when https://review.openstack.org/184493 merges in a
>> day or two this will become its default. Any last-minute objections
>> or concerns?
>>
>> Note that similar work is nearing completion is oslo.db with
>> https://review.openstack.org/184392 and there are also a lot of
>> similar changes in flight for other repos under the same review
>> topic (quite a few of which have already merged).
> 
> Ok, we've had 2 days fair warning, I'm pushing the merge button here.
> Welcome to the world of pure python mysql.

As a heads up for where we stand. The switch was flipped, but a lot of
neutron jobs (rally & tempest) went into a pretty high failure rate
after it was (all the other high volume jobs seemed fine).

We reverted the change here to unwedge things -
https://review.openstack.org/#/c/191010/

After a long conversation with Henry and Armando we came up with a new
plan, because we want the driver switch, and we want to figure out why
it causes a high Neutron failure rate, but we don't want to block
everything.

https://review.openstack.org/#/c/191121/ - make the default Neutron jobs
set some safe defaults (which are different than non Neutron job
defaults), but add a flag to make it possible to expose these issues.

Then add new non-voting check jobs to Neutron queue to expose these
issues - https://review.openstack.org/#/c/191141/. Hopefully allowing
interested parties to get to the bottom of these issues around the db
layer. It's in the check queue instead of the experimental queue to get
enough volume to figure out the pattern for the failures, because they
aren't 100%, and they seem to move around a bit.

Once https://review.openstack.org/#/c/191121/ is landed we'll revert
revert - https://review.openstack.org/#/c/191113/ and get everything
else back onto pymysql.

	-Sean

-- 
Sean Dague
http://dague.net



More information about the OpenStack-dev mailing list