backend-performance
Measure and improve latency, throughput or resource use in server-side request handling, background processing or data access. Use when profiling or a server-side budget identifies a backend bottleneck.
- License 许可
- MIT
- 最近更新
- 2026-11-14
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- backend-performance
- description
- Measure and improve latency, throughput or resource use in server-side request handling, background processing or data access. Use when profiling or a server-side budget identifies a backend bottleneck.
03 SKILL.md 正文
<!-- Generated by Paved plugin paved@2.29.0 from core.performance.backend-performance@0.1.2; edit Paved Core instead. -->
Backend performance
When to use
In the planning and implementation phases of a performance run, once profiling has located
the bottleneck in server-side code. It always starts from a measurement.
Required context
The baseline measurement and profile from profiling. Useful: the architecture context
(data stores, caches, remote services on the path) and the integrations context.
Preconditions
- A baseline measurement and a profile exist.
- The metric is a percentile, not only an average: tail latency is what users feel under load.
Procedure
- Classify the time the profile shows: computing, waiting on storage, waiting on remote calls, waiting on locks or pools. Each class has different fixes.
- Look for the usual causes, in this order, because they are common and cheap to
confirm:
- one query or remote call per item inside a loop, instead of one for the batch;
- data loaded and then discarded: unused columns, unused relations, missing limits;
- lookups the data store cannot serve from an index;
- remote calls made one after another that do not depend on each other;
- pools (connections, threads) too small for the load, so requests queue;
- repeated work on unchanged input that a cache could serve.
- Fix one cause, then measure again with
profiling. - For caching, decide before writing it: what the key is, how entries are invalidated, how stale a result may be, what happens on a miss storm. A cache without an invalidation answer is a correctness bug waiting.
- For concurrency, check shared state, ordering assumptions and limits on the collaborators you now call in parallel.
- Check the effect on collaborators: a faster service can overload the next one.
Tools
None. Measurement tools come from profiling.
Rules
None beyond those that apply by scope; the measurement rule is enforced through
profiling.
Verification
Required: performance. Recommended: integration, to show data access and remote
calls still behave the same.
Evidence
Baseline and result measurements for the changed path.
Completion criteria
- The measured percentile meets the target, measured as the baseline was.
- Cache invalidation and concurrency limits are stated for any cache or parallelism added.
Failure modes
| Failure | Signal | Response |
|---|---|---|
| Stale data | Cache serves results that should have changed | Fix invalidation before keeping the cache |
| Load moved, not removed | Downstream latency or errors rise | Measure the collaborator; add limits |
| Average improved, tail did not | Median drops, high percentile unchanged | Profile the slow requests specifically |
| Contention introduced | Throughput drops under parallel load | Revisit shared state and pool sizes |
References
None.
Paved integration
Use the project-local Paved contract as the source of truth. Discover state with node "${CLAUDE_PLUGIN_ROOT}/scripts/paved.mjs" status --json, load only relevant .paved/project/ context, and invoke verification with node "${CLAUDE_PLUGIN_ROOT}/scripts/paved.mjs" verify --json. Always run it from the repository being worked on, never from the plugin directory. Do not edit generated integration files or bypass Paved lifecycle and ownership checks.
04 同仓库更多 Skills
sample-skill
Does a sample task. Use when a sample is needed.
unexplained-contract
Does a sample task. Use when a sample is needed.
shell-block
Does a sample task. Use when a sample is needed.
project-identifier
Does a sample task. Use when a sample is needed.
oversized
Does a sample task. Use when a sample is needed.
no-activation
Does a sample task.