Optimize memory allocation for index lookup

[zyguan]

Enhancement

After profiling some index lookup workloads, we found that the go runtime overhead are remarkable. By optimizing memory allocation on some critical path, performance can be improved by more than 10%. So we plan to merge those optimizations, and here is the task breakdown.

• Reuse goroutines (in copr, disksql, etc.) to avoid the overhead of new/grow stack.
  • related issues:
    • use goroutine pool to copr.(*copIteratorWorker).run #52887
    • There is a 7% performance regression in Taobench benchmark #51852
• Reduce memory allocations of ExecDetails and RuntimeStats.
  • related path: distsql fetch resp, unmarshal resp in recv loop, TableHandlesToKVRanges, etc.
• Reduce memory allocations of BuildCopIterator.
  • related path: BuildRPCContext, SplitKeyRangesByLocations, BatchLocateKeyRanges, batchStoreTaskBuilder, etc.