Field notes and guides
Guides to machine observations, service recovery and workload performance. Engineering tools and earlier website, search and business-workflow articles are collected by topic.
Machine observations and service recovery
Read operating records, investigate missing measurements and plan recovery checks.
Specify your data center’s monitoring requirements
Prepare a monitoring review around operating decisions, available observations, physical measurement gaps, access requirements and acceptance checks.
↗Document storage media retirement
Keep storage-media identity, sanitization evidence, acceptance and destination traceable when retiring data-center equipment.
↗Set acceptance checks before a configuration change
Write the expected result, evidence, reviewer and rollback conditions before changing a data-center system configuration.
↗Before adding equipment, review the rack’s phase currents
Prepare a rack power review with separate phase readings, verified outlet and branch mapping, representative demand, and an approved connection plan.
↗An event can carry two timestamps
Compare source and collection timestamps without losing their meaning. Check clock synchronization, time-zone handling and the fields used in an incident timeline.
↗Close a cable move with matching labels and connection records
Review cable identifiers, recorded endpoints and the completed change together so the next maintenance task starts with a usable connection record.
↗Plan UPS runtime around the work that must finish during an outage
Define the required outage response, record the UPS model and load, and use manufacturer runtime data to review compatible battery configurations.
↗Trace bypass and recirculation paths
Identify cool-air bypass and warm-air recirculation, inspect relevant rack openings, and document the evidence needed for an airflow review.
↗Measure air at the equipment inlet
Plan a rack inlet-temperature review with documented probe locations, equipment requirements and operating conditions that make the readings useful.
↗Data center operations after deployment
A practical review of equipment records, measurements, change history and operating responsibilities after a data center deployment.
↗Data-center power monitoring starts with a better handoff
DOE's oscillation-monitoring guidance raises a practical handoff question for data centers: what can the monitoring see, and who owns the gaps?
↗A service can be running before it's ready
For data-center and IT operators: separate process startup, readiness and recovery before the next restart sends work to a service that can't handle it.
↗Review ownership and recovery before deleting an unused Kubernetes volume
Kubernetes now exposes an unused-PVC signal. Turn it into an owned storage review, with retention and recovery checks, before treating it as a cleanup list.
↗Incremental Kubernetes backups need a tested restore path
Changing Kubernetes backup tooling? Check the changed block tracking API migration, retained recovery points, and application restore evidence before rollout.
↗Karmada has graduated. Your failover plan still needs a rehearsal.
Karmada's CNCF graduation is a project milestone. Use this operator checklist to examine recovery capacity, application state, and failover evidence.
↗Review a service incident path with OpenTelemetry
Use OpenTelemetry's graduation as a reason to review one service's incident path: useful signals, Collector configuration, ownership, and response.
↗Workload performance
Compare correct output, latency, memory and resource limits under stated conditions.
Your inference benchmark changed. Did the system get faster?
Moving from GenAI-Perf to AIPerf? Align reasoning-token metrics before comparing inference latency, usable capacity or the case for more hardware.
↗Test the workload before counting on Memory QoS
Kubernetes v1.37 changes Memory QoS defaults. A practical evaluation for data-center teams: check active policy, memory pressure, latency and useful output.
↗JRuby vs CRuby performance: measure the work your application does
Compare JRuby and CRuby using separate startup and steady-state tests, verified output, memory limits and the same application workload.
↗Ruby concurrency: when to use threads, fibers, Ractors or processes
Choose a Ruby concurrency model by the work it must handle: waiting on I/O, using CPU cores, protecting shared state or isolating failures.
↗Rails memory keeps growing. Is it a leak, a larger workload or the worker layout?
Investigate Ruby on Rails memory growth with a repeatable workload, process-level measurements, GC counters and a clear test for retained objects.
↗NVHBM changes the memory design. Start with the workload.
NVIDIA's NVHBM announcement raises a workload question: where is time actually spent? A practical evaluation plan for compute and data-center teams.
↗Systems research
Why computers still waste time and energy.
Read the systems studies separately from Valen Systems' field notes.
→The Scheduler Can’t See the Work
Why a general-purpose scheduler sees runnable work but not the request, dependency, cache, or accelerator context behind it.
→An Idle CPU Isn’t Always the Right CPU
Why CPU placement has to account for cache, NUMA distance, capacity, and cooperating work instead of idle time alone.
→Migration Trades Locality for Balance
Why moving work can restore balance while discarding cache and memory locality that the task had already earned.
→Measure the Latency and Throughput Costs of Scheduler Slices
Why responsiveness and throughput pull scheduler decisions in different directions.
→Express Workload Priorities in CPU Scheduling Policy
Why fair-share accounting can’t by itself know which work matters most to a person or a business.
→History Arrives After the Work Changes
Why historical utilization is useful for prediction but can lag behind sudden workload changes.
→Engineering tools and history
Use the versioned tool releases and development guides, and read the history behind our systems work.
Set up and verify a Rails project on Omarchy
Set up an existing Rails project on Omarchy without confusing a global Ruby install, project dependencies and a production-ready application.
↗24 open-source tools for Cloudflare work
Download 24 free MIT-licensed tools for Cloudflare Workers, D1, R2, email, releases, routes, tunnels, MCP and operational checks.
→Check Worker request policies before wiring them in
Download Worker No-Egress and Rate Limit Contract: two free local tools for testing outbound request policies and synthetic rate-limit responses.
↗Email Fixture Lab: inspect an email file locally
A free CLI and local MCP tool for inspecting .eml structure, nested messages and reported delivery-status fields without uploading the message.
↗Three small Cloudflare tools, now open source
Download Worker Request Kit, Wrangler Config Doctor and Cloudflare Release Proof: three free v0.1 betas with public source.
↗Two more free tools for Cloudflare logging
Download OAuth Log Redactor and CF OTel Policy: two free, local-first beta tools for filtering JSON logs and configuring Worker logging policies.
↗Red Eagle Fenv and the 2006 Microsoft Accessibility Competition
The documented story of Red Eagle Fenv, Troy Mallory, the 2006 Microsoft accessibility competition, and the technical history still being recovered.
↗Websites, search and business workflows
Earlier field notes on client websites, search measurement and customer-inquiry handling.
Before you connect your inbox to your CRM
A practical inquiry-to-CRM handoff guide: define the record, owner, next action and failure cases before paying for an integration.
↗Choose Which URLs Belong in a Public Sitemap
How sitemap families help search engines discover and classify pages, and why inclusion cannot make a weak page competitive.
↗Measure Branded and Non-Branded Search Separately
Why ranking for a company name is different from ranking for a service, and what a local business must build to compete for non-branded demand.
↗Use Separate Measures for Edge Traffic and Search Performance
Why edge requests, crawler activity, and internal tooling traffic cannot stand in for impressions, rankings, qualified visits, or leads.
↗Constant Crawling Didn’t Make the Site Rank
Why constant website crawling does not improve rankings, what a five-minute crawl loop actually proved, and how we rebuilt it around controlled releases.
↗Separate Client Website Ownership from Reusable Platform Infrastructure
A founder's field note on client website ownership and the boundary between contracted deliverables and reusable platform infrastructure.
↗What We Learned Building Local Service Pages Without Thin Content
A practical account of moving from city-and-service templates toward pages with real service detail, local proof, internal authority, and useful next steps.
↗