[#2145] feat(server): The server-side supports choosing Netty's ByteBufAllocator.

[zhengchenyu]

What changes were proposed in this pull request?

The server-side supports choosing Netty's ByteBufAllocator.

Why are the changes needed?

Fix: #2145

Does this PR introduce any user-facing change?

Introduce rss.server.netty.pooled.allocator.enabled to choose allocator.

How was this patch tested?

test in cluster.

[rickyma]

The Uniffle server does not support PooledByteBufAllocator for now. You can refer to #1534.

[github-actions]

Test Results

 2 744 files   -  46   2 744 suites   - 46   5h 8m 4s ⏱️ - 16m 25s
 1 019 tests ±  0   1 016 ✅ + 1   1 💤 ±0  2 ❌  - 1 
12 562 runs   - 102  12 541 ✅  - 97  15 💤 ±0  6 ❌  - 5 

For more details on these failures, see this check.

Results for commit 84ab3d4. ± Comparison against base commit 8f4494b.

[zhengchenyu]

The Uniffle server does not support PooledByteBufAllocator for now. You can refer to #1534.

Thanks for you quick reply! It is just a draft.
I think you mean that if we used pooled allocator, we allocate chunkSize, but DIRECT_MEMORY_COUNTER in netty is not matched with usedMemory in uniffle, then may result to OOM. Am I correct in my understanding?
If so, just close this pr.

[rickyma]

Yes, you are right.

[zhengchenyu]

Yes, you are right.

Thanks for your reply! I will revert it in our repo, then I'm trying to use mimalloc that you talked about earlier.

[xianjingfeng]

I need this feature. I found that in our production environment, memory allocation and deallocation are very slow on the machines with large memory(1.5T). We will reserve sufficient memory, so I think the risk of OOM is relatively low for us. So I think this feature can be used as an option. cc @rickyma @zhengchenyu @jerqi @zuston

[zuston]

I need this feature. I found that in our production environment, memory allocation and deallocation are very slow on the machines with large memory(1.5T). We will reserve sufficient memory, so I think the risk of OOM is relatively low for us. So I think this feature can be used as an option. cc @rickyma @zhengchenyu @jerqi @zuston

Make sense to me. Let @zhengchenyu take a deep look.