https://1byteone.github.io/en/tags/%E6%80%A7%E8%83%BD/性能HugoBlox Kit (https://hugoblox.com)en-usFri, 21 Aug 2026 00:00:00 +0000https://1byteone.github.io/media/icon_hu_1c0e9cb08cfb822a.pnghttps://1byteone.github.io/en/tags/%E6%80%A7%E8%83%BD/https://1byteone.github.io/en/blog/jvm-memory-gc/Fri, 21 Aug 2026 00:00:00 +0000https://1byteone.github.io/en/blog/jvm-memory-gc/<p><em>Whiteboard note: the goal is not to memorize boxes, but to connect each arrow to a request, a contract, and an operational signal.</em></p> <h2 id="start-with-a-real-scenario">Start with a real scenario</h2> <p>Use this situation as the mental model: <strong>A recommendation API protects p99 latency by measuring allocation bursts rather than guessing GC flags.</strong>. A useful backend diagram answers four questions: where does input enter, who owns state, which boundary can fail, and how do we know the system is healthy? The flow is <strong>Allocation → promotion → reachability → collection</strong>.</p> <h2 id="deconstructing-the-architecture">Deconstructing the architecture</h2> <ul> <li><strong>Contract</strong>: define input, output, and error shape before adding implementation detail.</li> <li><strong>Mechanism</strong>: GC Root → Reachability Analysis; Young Gen: Eden + Survivor From / To. This is the part that determines latency, throughput, and testability.</li> <li><strong>State and resources</strong>: Minor GC → copy live objects; Promotion → long-lived objects enter Old Gen. Explain creation, ownership, cleanup, and recovery.</li> <li><strong>Failure path</strong>: Mark → Sweep / Compact; combine it with <strong>G1 / ZGC → choose for latency goals</strong> when choosing a fallback, retry, or rollback.</li> </ul> <h2 id="a-small-implementation-boundary">A small implementation boundary</h2> <p>The following sketch is intentionally small. It shows where a production implementation should place validation and ownership rather than pretending that a happy path is enough.</p> <div class="highlight"><pre tabindex="0" class="chroma"><code class="language-text" data-lang="text"><span class="line"><span class="cl">request → bounded resource → validated boundary → observable result </span></span></code></pre></div><p>Keep external systems behind an adapter, repository, client, or gateway. The domain layer should depend on a stable contract so tests can use fakes and incidents can be isolated to one boundary.</p> <h2 id="engineering-decisions-for-the-scenario">Engineering decisions for the scenario</h2> <ol> <li><strong>Protect dependencies first.</strong> Pools, queues, semaphores, caches, and worker counts need explicit limits. An unbounded queue only hides overload.</li> <li><strong>Preserve correctness.</strong> Use an idempotency key, transaction boundary, lock, version check, or schema validation where retries or concurrency can duplicate work.</li> <li><strong>Optimize with evidence.</strong> Use traces, slow-query data, GC pauses, queue depth, or p99 latency before choosing a cache, index, batch size, or concurrency setting.</li> <li><strong>Make recovery testable.</strong> Timeout, retry, rate limit, circuit breaking, rollback, and alerting should be exercised in a drill.</li> </ol> <h2 id="common-mistakes">Common mistakes</h2> <ul> <li>Drawing only the happy path while omitting timeout, empty data, rejection, rollback, or replica lag.</li> <li>Treating framework defaults as business contracts and discovering their limits after an upgrade.</li> <li>Scaling machines before measuring pool exhaustion, lock contention, slow SQL, allocation, or event-loop blocking.</li> <li>Treating logs as the only observability tool; metrics, traces, sampling, and redaction are also required.</li> </ul> <h2 id="production-checklist">Production checklist</h2> <ul> <li><input disabled="" type="checkbox"> Input, output, and error responses have explicit schemas</li> <li><input disabled="" type="checkbox"> Dependencies have timeout, retry budgets, rate limits, and fallback behavior</li> <li><input disabled="" type="checkbox"> Pools and queues expose capacity, waiting, and rejected-work metrics</li> <li><input disabled="" type="checkbox"> Important state has idempotency, transaction, or recovery semantics</li> <li><input disabled="" type="checkbox"> Logs, metrics, and traces share a request or correlation ID</li> <li><input disabled="" type="checkbox"> Regression tests, load tests, and failure drills use production-like data</li> </ul> <h2 id="summary">Summary</h2> <p>You understand this whiteboard when you can start from one request, explain every arrow&rsquo;s contract, state how each risk is handled, and name the signal that would tell you to change the design.</p>https://1byteone.github.io/en/blog/python-asyncio-concurrency/Fri, 21 Aug 2026 00:00:00 +0000https://1byteone.github.io/en/blog/python-asyncio-concurrency/<p><em>Whiteboard note: the goal is not to memorize boxes, but to connect each arrow to a request, a contract, and an operational signal.</em></p> <h2 id="start-with-a-real-scenario">Start with a real scenario</h2> <p>Use this situation as the mental model: <strong>A profile page fans out to several services while a semaphore protects the slowest dependency.</strong>. A useful backend diagram answers four questions: where does input enter, who owns state, which boundary can fail, and how do we know the system is healthy? The flow is <strong>Coroutine → Event Loop → Await I/O → Gather Results</strong>.</p> <h2 id="deconstructing-the-architecture">Deconstructing the architecture</h2> <ul> <li><strong>Contract</strong>: define input, output, and error shape before adding implementation detail.</li> <li><strong>Mechanism</strong>: async def → coroutine object; create_task → scheduled work. This is the part that determines latency, throughput, and testability.</li> <li><strong>State and resources</strong>: await → yield while I/O is pending; gather / TaskGroup → coordinate tasks. Explain creation, ownership, cleanup, and recovery.</li> <li><strong>Failure path</strong>: Semaphore → bound downstream concurrency; combine it with <strong>Cancellation + timeout → cleanup path</strong> when choosing a fallback, retry, or rollback.</li> </ul> <h2 id="a-small-implementation-boundary">A small implementation boundary</h2> <p>The following sketch is intentionally small. It shows where a production implementation should place validation and ownership rather than pretending that a happy path is enough.</p> <div class="highlight"><pre tabindex="0" class="chroma"><code class="language-text" data-lang="text"><span class="line"><span class="cl">request → bounded resource → validated boundary → observable result </span></span></code></pre></div><p>Keep external systems behind an adapter, repository, client, or gateway. The domain layer should depend on a stable contract so tests can use fakes and incidents can be isolated to one boundary.</p> <h2 id="engineering-decisions-for-the-scenario">Engineering decisions for the scenario</h2> <ol> <li><strong>Protect dependencies first.</strong> Pools, queues, semaphores, caches, and worker counts need explicit limits. An unbounded queue only hides overload.</li> <li><strong>Preserve correctness.</strong> Use an idempotency key, transaction boundary, lock, version check, or schema validation where retries or concurrency can duplicate work.</li> <li><strong>Optimize with evidence.</strong> Use traces, slow-query data, GC pauses, queue depth, or p99 latency before choosing a cache, index, batch size, or concurrency setting.</li> <li><strong>Make recovery testable.</strong> Timeout, retry, rate limit, circuit breaking, rollback, and alerting should be exercised in a drill.</li> </ol> <h2 id="common-mistakes">Common mistakes</h2> <ul> <li>Drawing only the happy path while omitting timeout, empty data, rejection, rollback, or replica lag.</li> <li>Treating framework defaults as business contracts and discovering their limits after an upgrade.</li> <li>Scaling machines before measuring pool exhaustion, lock contention, slow SQL, allocation, or event-loop blocking.</li> <li>Treating logs as the only observability tool; metrics, traces, sampling, and redaction are also required.</li> </ul> <h2 id="production-checklist">Production checklist</h2> <ul> <li><input disabled="" type="checkbox"> Input, output, and error responses have explicit schemas</li> <li><input disabled="" type="checkbox"> Dependencies have timeout, retry budgets, rate limits, and fallback behavior</li> <li><input disabled="" type="checkbox"> Pools and queues expose capacity, waiting, and rejected-work metrics</li> <li><input disabled="" type="checkbox"> Important state has idempotency, transaction, or recovery semantics</li> <li><input disabled="" type="checkbox"> Logs, metrics, and traces share a request or correlation ID</li> <li><input disabled="" type="checkbox"> Regression tests, load tests, and failure drills use production-like data</li> </ul> <h2 id="summary">Summary</h2> <p>You understand this whiteboard when you can start from one request, explain every arrow&rsquo;s contract, state how each risk is handled, and name the signal that would tell you to change the design.</p>