Hi Laura,
W dniu 17.09.2025 o 11:09, Laura Hofer pisze:
> Hi Krzysztof, Hi Roman,
>
> we are currently planning to switch our Unity setup to an HA setup
> with one master and two workers.
> This raised the question of how best to proceed with Unity updates: Do
> all Unity instances have to be stopped and updated at the same time,
> or can the instances be updated gradually (e.g. within a few hours)?
> Could the second option cause problems with the database if, for
> example, the master is already running a newer version of Unity, but
> the workers are still running an older version?
Newer Unity can not work on older DB - actually at the startup it will
try to update DB schema.
That said, if you want to have close-to-zero downtime deployment
blue-green sounds as the proper approach.
0. Blue is running old version (and can have HA, e.g. 3 instance)
1. install the new version on Green
2. (critical; see notes below) stop Blue, clone DB from Blue to Green
3. start the stack on Green
4. switch routing to Green
5. deprovision Blue
In this scheme #2 is critical, this is when your downtime starts (ends
at #4). Typically such short downtime is OK: that should take few
minutes, and in case of troubles with new installation rollback is
immediate - you just start Blue again.
But we can do better here: instead of fully stopping Unity, we can have
a feature to put it into RO mode. This would require some development
and precise definition (note: in Unity even a simple new sign-in is a
write operation from DB standpoint), but could help a lot: e.g. existing
sessions/tokens could be serviced w/o interruptions. But that would
require a new feature.
HTH,
Krzysztof
|
