On Fri, 11 Aug 2023 15:32:58 GMT, Aleksey Shipilev <sh...@openjdk.org> wrote:
>> When backporting [JDK-8312127](https://bugs.openjdk.org/browse/JDK-8312127), >> I realized there are no targeted tests for `FileDescriptor.sync` that can be >> used to qualify the changes in that area. >> >> Additionally, we use `FD.sync` for durability in Java databases, and we want >> to make sure at least some smoke tests are available in OpenJDK. Asserting >> durability would be hard, but let's at least test the Java code does not >> throw unexpected exceptions and native code does not crash the VM. >> >> The benchmark will show, among other things, that the recent change to >> `FileDescriptor.sync` does not affect the performance much, compared to the >> cost of the `fsync` itself. It deliberately targets tmpfs to provide the >> lowest actual FS overhead. >> >> >> Benchmark Mode Cnt Score Error Units >> >> # Before JDK-8312127 >> FileDescriptorSync.sync avgt 15 351,688 ? 2,477 ns/op >> >> # After JDK-8312127 >> FileDescriptorSync.sync avgt 15 353,331 ? 2,116 ns/op >> >> >> The new regression test completes in <0.5s on my Mac. > > Aleksey Shipilev has updated the pull request incrementally with one > additional commit since the last revision: > > Review comments AutoDelete is probably a bit overkill but okay. ------------- Marked as reviewed by alanb (Reviewer). PR Review: https://git.openjdk.org/jdk/pull/15231#pullrequestreview-1575704528
