Petr Hála
Petr Hála
> We need to avoid name changes or handle it here so comparison would match. > We can't lose history. This wont be backported, for master this is not an...
> parallel nemesis test and perf test with nemesis worth testing too That was a good call, I found a few more minor issues with the code, like #13329 v6:...
> Each case is different. Here, if we want to keep the names matched, we need info how to match nemesis name with disrupt function. @fruch @roydahan if comparing history...
> option 2, is exactly my ask, to accommodate in our APIs room for changes like this. > > don't mind turning it an issue on it's own, but it...
> @pehala you need to fix conflicts before we can merge v7: * Rebased
> How is it acceptable to have a user facing queries to fail for the sake of the raft majority ? > > Like when a user will face that,...
> We have numerous tools, none of them is capable of being told to expect failure. > > All of them have some retry configuration, but it doesn't seem to...
> Tried running just the enospc nemesis. It eventually fails every time. Yes, the nemesis needs to be disabled for tablets from 2025.4 onwards
> Even with the new enospc prevention, there are other reasons that can cause a node to reach enospc scenarios. > For example: coredumps on the same fs, bloated snapshots,...
> so is it a scylla issue or a driver issue? > ([#12989 (comment)](https://github.com/scylladb/scylla-cluster-tests/issues/12989#issuecomment-3687717684)) Bit of both, Scylla should provide more information so driver can react