issue: #47827
Key optimizations:
- O(1) task lookup in scheduler via unissuedTasksIndex map
- Channel→indexed slice in searchSegments, eliminating channel
alloc/sync
- Single segment fast path skipping errgroup/goroutine overhead
- Single channel fast path in LB policy
- Shallow copy SearchRequest instead of proto.Clone deep copy
- Pipeline.String()→p.name in debug log to avoid eager evaluation
- TimeRecorder: fix double time.Now(), pre-compute logLabel, IsRecording
guard
- Batch 2D array allocation for resultOffsets in search reduce
- Pre-allocate Scores/Topks slices with nq*topk capacity
- Replace fmt.Sprint(nodeID) with paramtable.GetStringNodeID() on hot
paths
- Eliminate fmt.Sprintf in CtxElapse calls, use static strings
Signed-off-by: Li Liu <li.liu@zilliz.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Li Liu
congqixia
Xiaofan
gcmutator
jaime