This review attempts to provide better cost metrics for the upgrade
steps. I do a few things:
- restart mysql before the upgrades to flush innodb buffers and
caches
- run each migration as a separate job
- log the innodb counters from mysql after each run
This will provide some first order metrics which are better than
wall time -- for example the number of written rows is probably
a more interesting metric than the number of seconds on a loaded
instance with a noisy neighbour.
We also need to do git pulls on stable branches, as zuul doesn't
do that for us.
Change-Id: I39e964a9bbe43717b5a5b5dfd236ee2fa2c62a00
ffe7a2cf34