Pavel Emelyanov

> ... so existing nodes should be able to know this is a cache cold node. So do we really need to know the per-table cache hit rate? Maybe node...

> Assume that the greater uptime the more populated the cache is? Of course, there's no other decision that's apparently better than this one > How much more? Dunno. When...

@bhalevy , yes, but unfortunately didn't make visible progress :(

@roydahan , is there a dtest specifically for local disk ENOSPC?

@bhalevy , it looks like the fix a980510654e5a504da1b58bf8398c05676e01572 is awesome, but needs high-precision tuning. It captures `system_error`, however, as we've seen in https://github.com/scylladb/scylladb/issues/10304#issuecomment-1148234936 , ENOSPC is propagated via `storage_io_error` and...

Regarding https://github.com/scylladb/scylladb/issues/11245#issuecomment-1250774786: The e4e92d44ae3015d81c3fb3f7386c65eacad7a790 needs fixing. It stops compaction manager with `deferred_stop()` which is the generic stopping wrapper that doesn't care about ~~ENOSPC~~ any exception that may appear. However, the...

I'm trying to wrap my head around `ENOSPC` vs `storage_service::do_isolate_on_error()` and in parallel cooking patches for two aforementioned places

> @xemul whay else has to be done to close this issue? I'd call for testing. And storage_service's IO error subscription still haunts my mind... > weren't the fixes merged...

LGTM , but I'd like to ask @nyh to comment on the rpc_address -> broadcast_rpc_address replacement

Awesome :) it's a pity that aarch64 is the blocker. Is it possible to teach zdtm.py not to use binary logging on this arch? If yes, we could merge the...