Mark
Mark
This has been implemented in https://github.com/duckdb/duckdb/pull/14717
Thanks! LGTM. Could you just look at the failing CI run?
Ah apologies, that CI failure is unrelated and should already have been resolved by a different PR. I meant the other two failing CI runs that seem to be related...
Thanks for the PR - but I'm not sure if this is safe. Also when not vacuuming row groups are modified quite invasively that could cause issues with concurrent operations....
Thanks - but this needs a lot more work. I think this is best tackled by someone from the core team.
For more context, the actual write to disk happens [right before executing the first query as part of the `sync` which is used for flushing the buffers](https://github.com/ClickHouse/ClickBench/blob/main/clickhouse/run.sh#L15). This is not...
Just to clarify - the issue here is not with the *read* not using I/O - as that is consistent across all systems. The issue here is with the *write*...