[perfloop-agent]

Comprehensive Summary of your change

For positive n below the maximum integer, local registry tag lists now fetch names strictly after last in byte order, with one lookahead row, instead of hydrating every tag. The existing response helper still produces JSON and continuation links. Explicit byte ordering preserves Go comparisons under non-byte database collations, including arbitrary cursor bytes.

With PostgreSQL 17 and n=1000, median Go-allocated bytes for the existing two-tag v1/v2 input fell from 13071 to 9175 B/request over 87 paired runs. Larger inputs extend that naming pattern as constructed scaling workloads.

Times are medians in ns/request; brackets bound the paired median time saved:

Tags	Handler before → after [bounds]	Handler change	PostgreSQL before → after [bounds]	PostgreSQL change	Pairs
2	401623 → 385151 [6676, 47464]	−4.6% (−11.8% to −1.7%, n=87)	33106 → 33751 [−1561, 1019]	~ (−3.1% to +4.7%, n=87)	87
1500	4079196.5 → 1530553 [2337894, 2637931]	−62.1% (−64.7% to −57.3%, n=16)	1057470.5 → 657691.5 [326619, 446871]	−37.9% (−42.3% to −30.9%, n=16)	16
10000	22071357 → 3121612 [16870971, 20084972]	−86.2% (−91.0% to −76.4%, n=16)	5230749.5 → 2193850 [2447397, 3445275]	−58.0% (−65.9% to −46.8%, n=16)	16

The two-tag paired SQL estimate is 743 ns/request slower, but its bounds span zero; zero overhead is not established. Lower allocations and large-list gains motivate the bounded query.

PostgreSQL time includes SELECT planning and execution, not driver/network waits or CPU time.

Native benchmark: src/server/registry/tag_benchmark_test.go. Further reproduction from src, with the DAO tests' PostgreSQL settings: go test ./server/registry -run '^$' -bench '^BenchmarkTagList$' -benchtime=8s -count=1 -benchmem.

The database may still scan or sort repository tags. Unpaginated requests, maximum-integer counts and proxy local/remote merging retain their existing paths. Production traffic savings are not measured.

Issue being fixed

Paged local tag lists retrieve and hydrate the whole repository before selecting a page.

Validation and contribution checklist

Go tests and race checks passed for tag DAO, manager and controller behavior, registry handling and registry pagination utilities. The database differential test compares response status, JSON and headers against the whole-collection response across mixed names, cursor positions and page boundaries. The native benchmark also completed a one-iteration smoke run at each size.

Please indicate you've done the following:

• Well Written Title and Summary of the PR
• Label the PR as needed. "release-note/ignore-for-release, release-note/new-feature, release-note/update, release-note/enhancement, release-note/community, release-note/breaking-change, release-note/docs, release-note/infra, release-note/deprecation"
• Accepted the DCO. Commits without the DCO will delay acceptance.
• Made sure tests are passing and test coverage is added if needed.
• Considered the docs impact and opened a new docs issue or PR with docs changes if needed in website repository.

Case and evidence

---

Made by Perfloop. Human sponsor: Tomás Senart.

[tsenart]

Hi, I'm Tomás, I build Perfloop. I approved this contribution and will handle review feedback here.

@chlins, this is in the registry tag list handler (src/server/registry/tag.go): when a client sets a positive n, Harbor now reads only the requested page of a local repository's tags instead of loading every tag first. Requests without n keep the existing path. We found it while hunting performance improvements with Perfloop. We've had changes merged in Distribution, Prometheus, ClickHouse, Redis, and more.

If you'd like anything in this patch changed, just let me know and I'll update it.