prefix_random distribution from FAST 2020 paper doesn't actually use key prefix
Open
- Dominant language
- C++
- Stars
- 32.1k
- Forks
- 6.9k
- Avg merge
- 32m
- Merged PRs (30d)
- 1
Description
The prefix_random distribution as described in Appendix A of https://www.usenix.org/system/files/fast20-cao_zhichao.pdf doesn't actually generate a distribution based on the keyrange parameters. If key_dist_a and key_dist_b are 0, then `use_random_modeling` gets set to true, and then `DistGetKeyID()` doesn't get called.
It seems the behavior was broken by https://github.com/facebook/rocksdb/pull/6795 which purported to fix the `all_random` workload but apparently broke this one.
Contributor guide
Assessment
This issue has not been assessed yet.