fix(nemesis): Fix wrong nemesis usage#13287
Conversation
738cc79 to
0f84590
Compare
yarongilor
left a comment
There was a problem hiding this comment.
the manager backup nemesis works fine in performance test in:
scylla-enterprise-perf-manager-native-backup-nemesis.jenkinsfile
It is not used by Sisyphus, no need to modify it.
Why it is not used by Sisyphus? It has class, it is in the yamls, so it should be used |
Discussed offline: |
|
@pehala @yarongilor why this nemesis are not part of Sisyphus ? where is the reason documented ? |
Due to bucket and object storage requirements that are region specific. It is not documented and I was confused by it as well, I can add documentation in this PR I also created task to enable them and we will tackle it in followup |
yarongilor
left a comment
There was a problem hiding this comment.
LGTM, if nemeses tested and passed.
|
@pehala new branch |
75f1b48 to
a17c294
Compare
|
@pehala |
a17c294 to
e729287
Compare
Removed arguments so they are now run as expected. Simplify method for backup
It was never executed in the expected way Add disrupt_rolling_restart_cluster_random to effectively reenable it
…veBackup Explains why they are not part of Sisyphus
e729287 to
33313b6
Compare
|
v2:
|
Was not executed as part of Sisyphus and served no real purpose
33313b6 to
331b367
Compare
Nemesis execution is not intuitive and the actual execution mechanism does not execute code inside
disruptmethod in nemesis classes, it is only used for pattern matching and then the same method is executed with default arguments.This PR fixes it and as a consequence we will now test more nemesis, so needs to carefully tested.
ManagerBackup nemesis are in COMPLEX_NEMESIS and as such are not executed by Sisyphus, so for them this PR is not critical but it improves them regardless and brings them closer to how nemesis should look like
Fixes #13039
Testing
PR pre-checks (self review)
backportlabels