Peter Dillinger
Peter Dillinger
> I thought the size of compressed blocks is typically known -- that is "compression format 2" which writes the size at the beginning. That doesn't help you predict the...
By the way, we are finally very interested in this line of features. Thanks for working on it!
Hmm, I have upset the Java gods by making the default value depend on a constant (derived from a preprocessor define)
Visual Studio 2019 appears to "just work" but gcc 9.x from Ubuntu 20 makes it difficult/ugly for us in CI at the moment. #12415
Unfortunately, this doesn't fix an internal failure in this test that we are seeing.
Super weird to see the warnings as error in Snappy because ``` cd snappy-1.1.10/build && CFLAGS='-arch x86_64 ' CXXFLAGS='-arch x86_64 ' LDFLAGS=' ' cmake -DCMAKE_POSITION_INDEPENDENT_CODE=ON -DSNAPPY_BUILD_BENCHMARKS=OFF -DSNAPPY_BUILD_TESTS=OFF --compile-no-warning-as-error .. &&...
> I also _borrowed_ some fixes from your [PR](https://github.com/facebook/rocksdb/pull/12426) FWIW, I want those fixes to be committed atomically with fixing the Windows CI. Then we can look at tightening the...
To be more precise, there is one mutable memtable per column family. There can be immutable memtables as well. We have an existing community contributed feature SstPartitioner that sets boundaries...
In your contrived example you assume essentially twice as much memory devoted to memtables in the second case as in the first case. If you simply devote the same amount...
By the way, I have a refactoring in progress that should completely solve this issue