Use JNI directly for posix_fadvise

[merlimat]

Motivation

For calling posix_fadvise, we are using JNA which is a wrapper on top of JNI. While JNA is convenient because it's dynamically creating a stub for accessing native methods, it's also very inefficient compared to direct JNI access.

Since we already have JNI modules, we should add posix_fadvise there and remove the dependency on JNA.

[hangc0276]

Great job!

[hangc0276]

Please fix the failed check:

Error:  /home/runner/work/bookkeeper/bookkeeper/bookkeeper-server/src/main/java/org/apache/bookkeeper/util/PageCacheUtil.java:37:28: Name 'FADVISE_POSSIBLE' must match pattern '^[a-z][a-zA-Z0-9]*_?$'. [StaticVariableName]
Audit done.
[INFO] There is 1 error reported by Checkstyle 9.3 with buildtools/src/main/resources/bookkeeper/checkstyle.xml ruleset.
Error:  src/main/java/org/apache/bookkeeper/util/PageCacheUtil.java:[37,28] (naming) StaticVariableName: Name 'FADVISE_POSSIBLE' must match pattern '^[a-z][a-zA-Z0-9]*_?$'.

[codecov-commenter]

Codecov Report

Merging #3824 (6ed509d) into master (284cf96) will increase coverage by 19.04%.
The diff coverage is 47.05%.

@@              Coverage Diff              @@
##             master    #3824       +/-   ##
=============================================
+ Coverage     41.40%   60.45%   +19.04%     
- Complexity     4034     5851     +1817     
=============================================
  Files           473      473               
  Lines         40963    40963               
  Branches       5240     5241        +1     
=============================================
+ Hits          16962    24764     +7802     
+ Misses        22160    13981     -8179     
- Partials       1841     2218      +377     

Flag	Coverage Δ
bookie	39.90% <47.05%> (?)
remaining	29.54% <47.05%> (?)
replication	41.38% <47.05%> (-0.03%)	⬇️
tls	20.98% <47.05%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
...java/org/apache/bookkeeper/util/PageCacheUtil.java	53.12% <40.00%> (ø)
...a/org/apache/bookkeeper/bookie/JournalChannel.java	62.93% <100.00%> (+2.58%)	⬆️
...g/apache/bookkeeper/proto/WriteEntryProcessor.java	69.01% <0.00%> (-9.86%)	⬇️
...pache/bookkeeper/server/service/BookieService.java	61.76% <0.00%> (-2.95%)	⬇️
...va/org/apache/bookkeeper/proto/BookieProtocol.java	83.57% <0.00%> (-0.72%)	⬇️
...ols/cli/commands/client/LedgerMetaDataCommand.java	0.00% <0.00%> (ø)
...rg/apache/bookkeeper/meta/LedgerMetadataSerDe.java	76.06% <0.00%> (+0.77%)	⬆️
...rg/apache/bookkeeper/client/BookieWatcherImpl.java	56.25% <0.00%> (+0.78%)	⬆️
.../java/org/apache/bookkeeper/client/BookKeeper.java	46.51% <0.00%> (+0.89%)	⬆️
...a/org/apache/bookkeeper/client/LedgerCreateOp.java	53.80% <0.00%> (+0.95%)	⬆️
... and 219 more

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[horizonzy]

We should check if nativeIO.posix_fdavise works or not. new NativeIOImpl() din't ensure posix_fdavise work

[wenbingshen]

@horizonzy I also found this when I run unit tests on mac os.

We don't need to check whether nativeIO.posix_fdavise is valid, because in use, once an error occurs, we will close posix_fdavise, so we can just check whether the library is normal, this is the error I encountered.

I have submit a PR: #3832

    public static void bestEffortRemoveFromPageCache(int fd, long offset, long len) {
        if (!fadvisePossible || fd < 0) {
            return;
        }
        try {
            NATIVE_IO.posix_fadvise(fd, offset, len, POSIX_FADV_DONTNEED);
        } catch (Exception e) {
            log.warn("Failed to perform posix_fadvise: {}", e.getMessage());
            fadvisePossible = false;
        }
    }