Andrew Kryczka

Consider picking from our up-for-grabs issues: https://github.com/facebook/rocksdb/issues?q=is%3Aissue+is%3Aopen+label%3Aup-for-grabs

> Is this 6ms reasonable for a 8Kb random read ? Probably not. If you enable statistics and provide them here, we can try to help debug - https://github.com/facebook/rocksdb/wiki/Statistics To...

> because even if i set block size to 32kb , the blocks that get read from ssd are in chunks of 4KB ? Do you mean that RocksDB is...

Oh ok. I'm a little bit interested in the topic of splitting I/Os because I ran into a pathological case before where the kernel always split large RocksDB reads into...

This is an unexpected setting: `target_file_size_multiplier=10`. It is probably resulting in enormous files in the bottom level, which will indeed nearly double the space when compaction runs. We use 1...

> enable this once we figure out how to adjust kill odds I think this was necessary because some of the db_stress invocations produced by whitebox crash test mainly/only crash...

> I don't remember what the issue was but we can try it and see what happens. Couldn't find any issue after specifying `-reopen=0`. I probably didn't want to do...

> can we add more kill points to the code? Would it make sense to add more kill points tointeresting code paths like middle of a flush, during compaction or...

> `earliest_seqno` is actually smallest seqno - 1, this violates the document that "It can then be assumed that any write with a larger(or equal) sequence number will be present...

There is no ritual for fixing the API documentation. Please feel free to submit a PR like the diff in your screenshot; we can review it then.