Skip to main content
automationtester.in
Engineering publication

The Automation Brief

Practical quality-engineering guides for people building reliable tests, delivery systems, developer tools, and AI-assisted workflows.

Tools & Reviews articles

19 published
SRShashank Rawlani
Tools & Reviews

Build a k6 Load Model from Real Traffic, Not a VU Count

In k6's closed-model executors the next iteration waits on the previous one, so a degrading system quietly receives less load and the run still reports green. This walks through converting peak-hour analytics into an arrival rate, driving it with constant-arrival-rate and ramping-arrival-rate, sizing preAllocatedVUs from median iteration duration times rate, and reading dropped_iterations as a capacity finding rather than a misconfiguration. It closes on thresholds — per-flow tag sub-metrics, p(95) and p(99) instead of avg, abortOnFail with delayAbortEval, and the duplicate-key trap that silently discards a gate.

14 min read
A dark technical illustration read as one left-to-right timeline. Along the top, a bright green curve ramps out of the left edge and holds a plateau to the right, with vertical ticks hanging from it that grow denser as it rises. Below, a wide panel of twelve horizontal lanes fills with green segments whose idle gaps shrink until, past a dashed orange vertical line, every lane is a solid unbroken bar. To the right of that line the ticks turn orange, fall as dashed lines straight past the full lanes, and accumulate as a rising stack of small orange squares along the bottom, continuing a flat green baseline that runs back to the left edge.
SRShashank Rawlani
Tools & Reviews

Test Cron Schedules by Their Fire Times, Not Their Strings

Reviewing a five-field string cannot tell you when a job runs: the two day fields are combined with OR in cron but AND in systemd, the spring gap deletes a run and the autumn fold duplicates one, and Kubernetes then decides separately whether a matched run may start. This shows you how to compute fire times from a pinned base time using the parser your scheduler actually vendors — robfig/cron for CronJobs, systemd-analyze calendar --base-time for timers — and how startingDeadlineSeconds, concurrencyPolicy and the 100-missed-schedule ceiling interact. It ends with the evidence to keep: the cronjob-scheduled-timestamp annotation diffed against actual start times, so a run that never happened shows up as a gap.

14 min read
Dark technical illustration of two schedule timelines running edge to edge. The upper lane shows wall-clock fire times as a row of green ticks hanging from a rule; at one highlighted transition column a tick is missing, drawn as a dashed outline with a crossed-out circle above it, and at a second transition column a single slot carries two overlapping orange ticks ringed together. Curved droppers connect every wall-clock tick to the lower lane, where the same schedule in UTC is a perfectly even row of green ticks with no gap and no duplicate, the fold's two instants landing as separate ringed nodes. A strip of small cells along the bottom repeats the pattern, one cell empty and one doubled.
Newsletter

The Automation Brief

Test suites, agents, workflows and pipelines — and how you prove any of it actually works. Practical notes from the tools being built here.

No spam, no sending cadence promised. Unsubscribe any time.