[perfloop]

hashKey runs before YCSB worker operations and while preloading keys. It keeps the eight-byte FNV-1a sweep and replaces fmt.Sprintf with a 20-byte stack buffer and strconv.AppendUint. The returned key stays user-prefixed, lowercase, and unpadded. The formatter-equivalence test covers zero, positive, negative, and signed int boundaries.

The direct hash benchmark reports per-call time and bytes. A fixed 100,000-call batch gates at least 100,000 fewer allocation events, one for each hash call. A separate RunYCSBWithContext workload runs 100,000 read-only operations with eight workers, 1,000 keys, and 100-byte values against an in-memory loopback gRPC server. Its full-run allocation result is a sampled Go alloc_space estimate attributed to hashKey, including preload; it is not total process allocation or production traffic. Worker-time and throughput are guards, not speedup claims; Result.Duration includes worker hashes but excludes preload and final reporting. The direct-hash and large-run profile harnesses are retained measurement support and are not included in this PR.

On workload YCSB hashKey for 100,000 nonnegative key numbers per sample, median metric hashkey_allocs_per_100000_calls changed from 200009 to 100000; paired median delta -100008 (-50% of baseline; at least 19/20 confidence interval -100010 to -100007 from 10 pairs).

Workload	Measure	Metric	Before median	After median	Paired change
YCSB hashKey for 100,000 nonnegative key numbers per sample	time	ns/op	115.2 ns	48.13 ns	-58.5%
YCSB hashKey for 100,000 nonnegative key numbers per sample	allocated bytes	B/op	32 B	24 B	-25%
YCSB hashKey for 100,000 nonnegative key numbers per sample		hashkey_allocs_per_100000_calls	200009	100000	-50%
YCSB Workload C read-only, 100,000 operations, 8 workers, 1,000 keys, 100-byte values, loopback gRPC		hashkey_alloc_bytes_per_ycsb_run	3356450	2451285	-28.4%

Each workload ran 10 paired runs.

The existing BenchmarkRunYCSBReadOnly/64KiB is an additional allocation guard; each Go benchmark iteration is one 64-operation read-only RunYCSBWithContext invocation with four workers, one key, and 64KiB values. Three one-second diagnostic samples reported allocs/op of 24,902 / 24,805 / 24,818 before the change and 24,747 / 24,757 / 24,809 after it. These diagnostic runs are separate from the official paired claim results. No new benchmark or build target is added to the PR.

YCSB Workload C, 64 operations, 4 workers, 1 key, 64KiB values, loopback gRPC (10 paired runs): declared guards passed.

make test-client, make build-cli, make fmt-check, make license-check, and client go vet ./... pass. The full repository make lint remains for target CI.

---

Generated by Perfloop.

---

Note

Low Risk
Behavior-preserving hot-path optimization with an equivalence test; key format and hashing logic are unchanged.

Overview
Optimizes YCSB hashKey so preload and every worker op spend less time and heap on key strings. The FNV-1a hash is unchanged; only string assembly moves from fmt.Sprintf("user%x", h) to a fixed user prefix on a 20-byte stack buffer plus strconv.AppendUint (lowercase, unpadded hex).

Adds TestHashKeyMatchesFormatter, which compares the new path to the old formatter for zero, typical values, and signed int min/max so key names stay compatible with existing caches and tests.

^{Reviewed by Cursor Bugbot for commit 133bd25. Bugbot is set up for automated code reviews on this repo. Configure here.}

[greptile-apps]

Confidence Score: 5/5

[Low impact] The PR appears safe to merge; key names and caller behavior stay unchanged.

Summary

hashKey replaces fmt.Sprintf with strconv.AppendUint and a local 20-byte buffer.

• YCSB builds hash keys without formatter allocations.

_{Reviews (1) · Last reviewed commit: "perf: avoid formatter allocations in YCS..." · Reviewed by Greptile}