[#1598] fix(server) Fix inaccurate used_direct_memory_size metric

[rickyma]

What changes were proposed in this pull request?

Fix the inaccurate used_direct_memory_size metric.
Also used_direct_memory_size_by_netty and used_direct_memory_size_by_grpc_netty metrics are added to provide more detailed indicators for locating and analyzing in production.

Why are the changes needed?

Fix #1598.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

Tested in our env.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 75.00000% with 3 lines in your changes are missing coverage. Please review.

Project coverage is 54.99%. Comparing base (32d533d) to head (6280ac3).

Files	Patch %	Lines
...pache/uniffle/server/NettyDirectMemoryTracker.java	62.50%	3 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##             master    #1599      +/-   ##
============================================
+ Coverage     54.01%   54.99%   +0.97%     
- Complexity     2863     2865       +2     
============================================
  Files           438      418      -20     
  Lines         24850    22498    -2352     
  Branches       2114     2114              
============================================
- Hits          13423    12372    -1051     
+ Misses        10586     9358    -1228     
+ Partials        841      768      -73     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[jerqi]

LGTM, cc @lifeSo

[github-actions]

Test Results

 2 340 files  ±0   2 340 suites  ±0   4h 32m 0s ⏱️ -1s
   908 tests ±0     907 ✅ ±0   1 💤 ±0  0 ❌ ±0 
10 541 runs  ±0  10 527 ✅ ±0  14 💤 ±0  0 ❌ ±0 

Results for commit 6280ac3. ± Comparison against base commit 32d533d.

[lifeSo]

Great!

[lifeSo]

@rickyma
I am interested in is there direct memory problem in your prd ?

[rickyma]

@rickyma I am interested in is there direct memory problem in your prd ?

No. I've fixed a lot of serious issues related to off-heap memory when enabling Netty.
There is only one issue #1596 left.
And I find out that gRPC does not work very well when the concurrency is above 11200. There is a huge problem when enabling gRPC. The load of CPU will be a lot higher than using Netty. I'll mention this later in another issue. But I might not fix this, because we will use Netty instead of gRPC. gRPC should be deprecated later on.