While populating data from sanitarium to the new clouddb hosts (T267090) some sections showed InnoDB errors after the upgrade and as soon as replication started.
Examples:
s1 T267090#6629364
s2 T267090#6644949
s4 T267090#6640399
The workaround for this (that has worked so far) is copying the data from the sanitarium master instead and sanitize clouddb hosts.
This makes me thing that maybe upgrading sanitarium hosts might trigger InnoDB errors for the sections above.
Let's take two hosts from {T267043} and use them as temporary sanitarium with 10.4, copying the data from the existing ones to them and see what happens with the upgrade.
Chosen hosts:
* db1154
[x] s1
[] s3
[x] s5
[x] s8
* db1155
[] s2
[] s4
[] s6
[] s7
There are two scenarios:
a) It works fine, so once ready we can just move clouddb hosts under them when ready and re-use the other ones for core
b) The upgrade doesn't work and we need to "convert" them by using the workaround above (copying the data from sanitarium masters and sanitize them).
[] Ensure InnoDB is compressed on the new hosts