GuideObservability
Tracing an AI request through Cloudflare and its origin
Follow a slow AI request from Cloudflare to an instrumented origin, configure sampling, and verify that the trace preserves the evidence needed to investigate.
Articles
Guides, news and explanations about AI and reliable systems.
GuideObservability
Measuring a latency SLO with Prometheus histograms
Build a latency SLI from Prometheus histogram buckets, include failed requests, and compare classic and native queries with a worked search-service example.
GuideIncident response
An SRE incident handoff in ChatGPT Space
A synthetic order-service incident follows a shared handoff through page edits, new findings and shift change, including the review and access work involved.
GuideObservability
OpenTelemetry tail sampling through a checkout failure
Tail sampling selects recorded traces by error, latency and other evidence. A checkout timeout shows how routing, waiting time and memory affect the trace a responder can retrieve.
GuideAI in production
CodeRabbit: how to use AI code review
CodeRabbit brings AI feedback into pull requests, editors and the CLI. A retry loop that exceeds its caller’s deadline shows the kind of finding that can repay review time.
GuideAI in production
Recovering an agent task after OpenShell containment
OpenShell can stop an agent after Kubernetes has accepted its scaling request. Recovery then depends on the Deployment state, available replicas and customer-facing result.
GuideAI in production
Meta Muse keeps research and connected tasks running
A release-note digest and a changed maintenance event show two uses of Muse’s persistent agent: gathering information over time and preparing a connected-account update.
GuideObservability
OpenTelemetry Collector queues and the wait for fresh data
A durable queue can preserve telemetry through an outage, then take longer to catch up. Arrival rate, export capacity and event age explain what recovery looks like.
GuideAI in production
Recovering an automation write after a timeout
A missing response can leave a worker already created. Durable operation IDs and the destination’s idempotency contract let a replacement executor recover the original request.
GuideObservability
Why a Prometheus error-rate query can return no data
Zero errors, no traffic and absent series produce different PromQL results. Synthetic tests show how ratios, fallback zeros and absence checks behave for a checkout service.
GuidePlatform engineering
Disruption budgets and the capacity left behind
Plan an inference-service node drain with a scoped disruption budget, replacement-capacity checks and a practical maintenance worksheet.
GuideAI in production
Prompt caching: measure the cost of useful work
What prompt-cache hits save once retries and failed work are included in task cost, with a local calculator and ledger.