Upcoming changes to clusterplans

[Zelldon]

We will move back to S, M and L clusterplans, see here the upcoming changes written down by Bernd https://docs.google.com/spreadsheets/d/1wbsB8-VuQITxYUx1ms31MD4kbwkgDhgzE4omL3ZbDRg/edit?usp=sharing

The plans are now prefixed with G-S, G-M and G-L. Where ID is the generation id, in our case the old GA plans will become G2-S and the new plans get the G3- prefix. Related discussion https://camunda.slack.com/archives/C02UT3T8DQR/p1645007586782299?thread_ts=1644575068.922369&cid=C02UT3T8DQR

I would suggest that we should test against all clusterplans again, which makes it possible again to run different experiments (based on topology etc.) ideally we map the clusterplan name to the old prod-s, -m and -l then we don't need to change the the structure in the chaos repo. But I'm open for discussion of course.

Btw we should do this sooner than later, and maybe also sync with @Sijoma otherwise our testbench runs will fail if the clusterplans are changed without our knowledge

[pihme]

I would suggest that we should test against all clusterplans again, which makes it possible again to run different experiments (based on topology etc.) ideally we map the clusterplan name to the old prod-s, -m and -l then we don't need to change the the structure in the chaos repo. But I'm open for discussion of course.

Not sure what you mean here. On the level of testbench, the names are used to look up the UUID of the cluster plan to be generated. On the level of the chaos tests, sure a mapping can be done. This would, in my mind, need to happen in the chaos test itself. I also wonder whether the mapping could be more flexible. Maybe not all tests depend on topology. So it might make sense to distinguish chaos tests which do depend on topology from those that don't and make the mapping a little more complex

[pihme]

Update: the new cluster plans are already in use on int. And testbench didn't break. The names of the cluster plan types have not changed, so our name based lookup still works and finds the old cluster plan types. (which are mapped to the new cluster plans)

This still leaves the issue of how to deal with cluster plan changes when the name stays the same.

[pihme]

I will keep this issue open for now, because the question which cluster plan types we want to test in the future is still open.