Wallace
Wallace
I made a mistake. I only see that memtable allow one thread write.
This proposal shall be use a thread-pool for memtable writing.
Do we need to consider how to split a manifest version of sst into multiple?
> > Do we need to consider how to split a manifest version of sst into multiple? > > What's it used for? To split one engine into two engine....
It seems complex. Why can not we launch multiple process and each process runs some test with a single thread?
I can not agree. Raft is essentially a special multipaxos algorithm. I think we shall rename `XPaxos` to `XRaft`.
For `B-Tree data structure`, have we known which component memory is used for Key-Value and which component memory is just allocated for B-Tree Node?
> Discussed offline. We think updating the metric in sstable iterator can be problematic. > > https://github.com/risingwavelabs/risingwave/blob/18eed8fa8543f51c244aa8c0ee4c01049020ce7b/src/storage/src/hummock/sstable/forward_sstable_iterator.rs#L115 This bug has been fixed
The `skip_key_count` comes from ` __internal_nexmark_q7_1_hashjoinright_1007`. It was caused by `SELECT MAX(price) AS maxprice`
How about use `poll_next` in `compact` ? For example: ```rust async fn compact(...) { while iter.is_valid() { iter.poll_next(); if iter.is_pending() { iter.await_next().await; } } } ```