Botond Dénes
Botond Dénes
New in v2: * Use `compaction_group_for_token()` and `compaction_group_for_token_range()` instead of iterating though all compaction groups.
New in v3: * Use return value instead of output parameter in `hold_gate_for_compaction_groups()`.
New in v4: * Add reserve() and fix typo.
> Why isn't holding the effective_replication_map_ptr sufficient? I don't know, but the use-after-free says it is not sufficient. > What's the impact of the change on performance? I don't think...
New in v5: * Don't process first range twice.
New in v6: * Use pre-increment instead of post increment
> > What's the impact of the change on performance? > > I don't think there is any, but I will measure it. For the hot path we care about,...
> Is this still needed? Possibly not. If it was a "free" change, I would say let's do it anyway for safety, but there is a performance impact, so one...
> I'm concerned (without definite proof) that we're not hermetically covered with closing the compaction_groups before destroying a storage_group upon tablet_cleanup, since some queries could have started using them before...
> > > I'm concerned (without definite proof) that we're not hermetically covered with closing the compaction_groups before destroying a storage_group upon tablet_cleanup, since some queries could have started using...