moka: Memory corruption observed when using Moka v0.9.6
Split from #279.
Note: I started looking into Moka because recently we tried using it in https://github.com/MaterializeInc/materialize/pull/19614, but had to revert it because we were observing memory corruption.
- Moka v0.9.6
- No stack traces and core dumps are available.
- The target platform (CPU arch and OS)
are yet unknown.=> Linux x86_64 moka::sync::Cacheare used in async contexts.- Multiple (many?)
Cacheinstances. (An instance per blob URL. Do not know how many) - Weigher and eviction listener (immediate mode) are configured.
I am trying to reproduce the issue using mokabench but no luck: https://github.com/moka-rs/moka/issues/279#issuecomment-1596820321
CC: @ParkMyCar
About this issue
- Original URL
- State: closed
- Created a year ago
- Comments: 15 (14 by maintainers)
FYI, I got the RAM modules for my PC, and confirmed it can run the test with at least 50,000 actions. I allocated 24GiB of memory and only 8 logical CPU cores to WSL2.
The reason I did not try 200,000 actions is that it will take long time (Estimated ~30 hours). 50,000 actions took 7 hours and 14 minutes. The memory utilization was fine so I had feeling that it could run 200,000 without issue.
Also, before I got the RAM module, I was also able to run 80,000 actions. I found that reducing the number of logical CPU cores for WSL2/Docker will reduce the memory utilization of the test. So I allocated only 8 cores (instead of 20) and 10GiB of memory (instead of 8GiB) to WSL2 and it was able to run 80,000 actions.
@0xdeafbeef Thanks for the offer. Wow, you have a great machine!
I think I am going to be okay. I already ordered the RAM modules and am waiting for them to arrive. I think 32GB should be enough for now.