receive: Add map pooling and optimize sendLocalWrite to reduce alloca…

[mkusumdb]

…tions

This commit implements three optimizations targeting ~612GB of allocations (11% of total) identified through production CPU profiling.

Optimization 1: Distribution Map Pooling (357GB, 6.41%)

• Added sync.Pools for distribution maps (endpointReplicaMapPool, tenantSeriesMapPool)
• Added helper functions for map clearing with size guards
• Modified distributeTimeseriesToReplicas to use pools
• Added cleanup in fanoutForward

Optimization 2: sendLocalWrite Improvements (255GB, 4.57%)

Part A: Pool tenantSeriesMapping (10.30GB)

• Added tenantTimeseriesMappingPool sync.Pool
• Modified sendLocalWrite to get/return map from pool

Part B: Optimize loop and pre-allocate (244.65GB)

• Changed loop iteration from value to index (avoids copying TimeSeries structs)
• Pre-allocate slices with capacity hints (single-tenant and multi-tenant cases)
• Lazy allocation for multi-tenant scenarios

Memory Safety:

• Size guards prevent pool ballooning (max 100 entries)
• Maps cleared before reuse (no data leakage)
• Error path cleanup prevents pool leaks

Testing:

• Added TestDistributionMapPooling
• Added TestClearMapFunctions
• All existing tests pass

Profiling Source:
Cluster: dev-aws-us-east-1-obs-integrationtest
Namespace: pantheon
Pod: pantheon-db-rep0-0
Date: 2026-02-06

• I added CHANGELOG entry for this change.
• Change is not relevant to the end user.

Changes

Verification