Hi Stefan,
One Question, to see if we are talking about the same:
You mean that only the database is running on the vm's? or you also
wanna start a sequoia controller inside the vm?
I am not sure yet about the controllers but if they can fetch their
configuration through http, that might also be an option which would
greatly simplify the cleanup procedure. The logs would still have to be
captured though to diagnose test failures.
What where the synchronization issues you had to face?
On Testing with sequoia controllers in vm' i often get really unusal
errors (pending writes....) and i think this might be because of the
time drift between the two nodes.
AFAIK the local system time is used for the checkpoints in the recovery
log and if one node falls behind the other time, that might cause the
problem.
Am i right, or is the time synchronization not so important?
Time synchronization should not be that important. Using virtual
machines can exhibit a different scheduling behavior that is not
noticeable at low loads. This can ease the reproduction of
synchronization issues though it might hide races that can only be seen
with higher concurrency levels. All in all, time skews should not be a
problem in Sequoia where global timestamps are just informational.
I will start branching Sequoia 4 today and start to add JIRA entries as
I move forward.
Cheers,
Emmanuel
_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia
