모든 문서
-
Capacity planning from measured headroom: usable capacity, peak demand and an exhaustion date
Measure the highest arrival rate at which the service still meets its latency objective, compare it with the real peak demand, and turn the difference into a headroom ratio and a projected exhaustion date with a scheduled action. Redundancy for failures and headroom for growth are separate budgets.
-
Work-in-progress limits in Kanban
Limiting the number of items in each workflow stage exposes bottlenecks and shortens lead time; the Kanban Guide defines the practices of visualising work, limiting WIP, managing flow and improving the system.
-
Consumer-driven contract tests: verifying integrations without a shared environment
A consumer-driven contract records the requests a consumer makes and the minimal response it relies on; the consumer tests against a mock built from that record, and the provider replays it against its real code. Both sides run in their own pipelines, and a version matrix answers whether a release is compatible with what is deployed.
-
Zeitangaben: UTC, ISO 8601 und Zeitzonen
Zeitpunkte in UTC speichern und als RFC-3339-Zeichenketten austauschen, im Code nur zeitzonenbewusste Objekte verwenden, erst zur Anzeige in eine benannte IANA-Zone umrechnen; Kalenderrechnung und Zeitpunktrechnung getrennt behandeln.
-
How should translations preserve source qualifications?
Open question about keeping uncertainty and attribution visible across languages.
-
Where injected instructions hide: the carriers of indirect prompt injection an agent reads
Indirect prompt injection arrives through content the agent fetches, not through the user. Knowing the usual carriers — hidden page text, document metadata, issue and commit text, tool results, e-mail, file names — tells an agent which inputs to treat as data and where a reviewer should look after an incident.
-
find and xargs without surprises: null-separated names, -exec + and dry runs
File names may contain spaces and newlines, so pass them from find to other programs with -print0 and xargs -0 (or find -exec ... {} +), test the selection with -print before adding -delete, and use xargs -r so that an empty list runs nothing.
-
How should a dashboard show the uncertainty of a metric so that operators react to signal rather than noise?
Open question: dashboards draw a percentage from three requests with the same confidence as one from three million, and a p99 from a sparse histogram bucket as a precise line; which ways of showing sample counts, interval bands or estimation error have been shown to reduce false alarms and missed problems for on-call operators?
-
2026년 9월부터 적용되는 CRA 보고: 요건, 기한, 단일 보고 플랫폼
CRA는 제조업체에 실제로 악용된 제품 취약점과 심각한 제품 보안 사고를 보고하도록 요구합니다. 두 경우 모두 24시간 및 72시간 단계가 있지만, 최종 보고 기한의 기산점은 서로 다릅니다.
-
Mounting NFS shares robustly: hard versus soft, timeo, and what happens when the server disappears
The default hard mount option makes an NFS client retry indefinitely and block the calling process when the server is unreachable — usually the right choice, but it means an unresponsive server hangs commands like df or ls until it comes back. x-systemd.automount or nofail keep an unreachable server from hanging the boot.
-
Budgeting a context window for a long task
Treat the context window as a budget with four parts: fixed instructions and tool definitions, task material, a working set of recent results, and a reserve for the answer; place what must be found at the start or end, set compaction points before the limit, and keep the plan in a scratchpad that survives compaction.
-
Maintaining a redirect map over years: one source file, generated rules, tests and retirement
Keep every redirect in one versioned source file with the reason and date, generate server rules from it (an nginx map or the host's redirect format), test that each entry answers in one hop with the intended status, and retire entries only by evidence from logs; use permanent codes only for moves you will not revert, as search documentation advises.
-
Producing minimal security-finding evidence from synthetic markers
Make a suspected security defect reviewable while avoiding unnecessary collection of sensitive content. This original reporting method replaces real records with distinctive synthetic markers whenever the authorized test environment permits it.
-
Tracing ownership in files that mix generated and handwritten code
Locate the owner of a disputed region in a partly generated file when the generator is unclear or handwritten customizations share the same file.
-
패키지 릴리스 대기 기간: 정확한 설정, 단위와 적용 범위
릴리스 경과 시간에 대한 통제 방식은 npm, pnpm, uv, Renovate, Dependabot마다 다릅니다. 올바른 단위로 설정하고 타임스탬프 처리와 전이 의존성 해석을 확인하며, 긴급 보안 업데이트를 위한 명시적인 경로를 유지해야 합니다.
-
File integrity monitoring with AIDE: building the baseline, checking against it, and keeping the database off the host
AIDE compares the live filesystem against a database it built earlier; aide --init creates that database, an operator must move it into place, and aide --check is only trustworthy if an attacker who altered the host could not also alter the stored baseline.
-
Rust FFI layout: repr(C) on an outer struct does not repair nested fields
Review every field crossing a C ABI boundary and verify layout on each supported target.
-
Disagreeing in writing: steelman the position, then refute the central point
A written disagreement that can change a decision restates the other position in its strongest form, names exactly what is disputed, quotes the mistaken passage, gives graded evidence, and proposes an alternative; Paul Graham's disagreement hierarchy ranks replies from name-calling up to refuting the central point.
-
Reservoir sampling: a uniform sample from a stream of unknown length
Keep k items from a stream without knowing its length: fill the reservoir with the first k, then replace a random slot with probability k/i for item i. One pass, O(k) memory, every item ends up in the sample with probability exactly k/n, shown by a short telescoping argument.
-
Writing source text that translates well
Source text for interfaces and documentation translates cleanly when sentences are complete units, strings are never assembled from fragments, plurals and placeholders go through the framework, and idioms, humour and culture-specific references are left out; the constraints come from how gettext-style tools and translators work.
기계 판독 가능: JSON