blog
Blog — page 4 of 39
Posts 73–96 of 928, newest first.
Customer-Authored Policy Code: Rego, CEL, and When a Rules Engine Needs a VM
Every B2B platform eventually lets customers write the rule. The honest answer is a three-rung ladder, and most teams should stop two rungs below the one I sell — a microVM create is 179 ms p50, which is three orders of magnitude slower than a CEL eval.
Ajay KumarTop 9 Ephemeral Development Environment Platforms (2026)
Three different products are sold as "ephemeral development environments," and most comparisons rank all three in one list. Here is the split that matters, nine platforms graded inside it, and why idle cost and TTL semantics are the two line items to read first.
Ajay KumarTop 5 Scratch Postgres Platforms for CI (2026)
Before you shop for a disposable-Postgres product: a tuned postgres:16 container plus CREATE DATABASE ... TEMPLATE beats every platform on the board for unit and most integration tests. Five options for the cases where it genuinely doesn't, including the one where my own platform is the slowest thing here.
Ajay KumarMGLRU and MicroVM Density: Reclaim When Every Page Is Someone's Guest
The host kernel picks reclaim victims using access information already filtered through a second kernel. MGLRU gives it generations instead of two buckets — which changes the shape of your bad days, not how many guests fit on the box.
Ajay KumarThe I/O Scheduler Inside a MicroVM Is Almost Always Wrong
Your guest's block layer is sorting requests by sector number to optimise seeks on a platter that is a file on somebody else's filesystem. Why none is the answer, why it still merges, and why readahead is the knob that actually moves the number.
firecracker-containerd: Running OCI Images in MicroVMs, and What It Costs You
It lets you keep your images, your registry and your containerd tooling, and swap the thing at the bottom for a microVM. That is a genuinely small diff for a large change in blast radius — and it is still the wrong tool if what you wanted was a fast sandbox API.
Ajay KumarFirecracker vs FreeBSD Jails: Two Honest Answers to the Same Question
Jails predate Linux containers by a decade and are better designed than them. They also share the host kernel, which is the entire comparison. Everything else is downstream of that one sentence.
Ajay KumarAudit Trails When the Machine Is Gone in 200 Milliseconds
The thing that makes a sandbox safe — it is destroyed — is exactly what makes "what did the agent do at 03:14?" unanswerable. You cannot keep the machine. You can keep its memory and disk.
Ajay KumarTop 10 Disposable Development Environment Platforms in 2026
Creating environments is the easy half. This is a roundup graded on the hard half: what actually gets destroyed when you destroy one, what survives that you did not plan for, and who else feels it.
Ajay KumarTop 6 Throwaway Postgres Platforms for Testing in 2026
A database for a human to poke at and a database for forty parallel CI jobs are not the same product. The axes that decide the automated case are parallel isolation, seed-cost amortisation, and what leaks when a run is cancelled at 40%.
Ajay KumarRunning MATLAB and Octave Workloads in Isolated microVMs
Most runtime isolation problems are technical. This one starts with a contract: you cannot fan out 200 MATLAB processes without the entitlements to match. Here is the honest version — licence manager, MATLAB Runtime, Octave's real compatibility cliff, and why the guest boundary wants to be a VM.
Ajay KumarSelf-Hosting a Compiler Explorer: Running Strangers' Compilers Safely
"I'm not running their code, I'm only compiling it" is the most expensive sentence in this genre. A compiler is a programmable machine with a filesystem, a plugin loader and an unbounded appetite for RAM — and on a compiler explorer, the flags are user input too.
Ajay KumarOptimisation Solver as a Service: Isolating Jobs That Run for Hours
Nearly every compute platform assumes it can predict how long a job will take from the shape of its input. A mixed-integer program is a counterexample: the same model class solves in 200 ms or runs until you stop it, and nothing in the file tells you which. Everything downstream of that — timeouts, memory, placement, reproducibility — has to be designed for the bimodal case.
Ajay KumarNotebooks in Production: Parameterised Runs in Disposable microVMs
The quarterly board metric comes out of cell 34, which must be run in order, by Dmitri, on his laptop. The notebook is not the problem. The laptop is. Papermill turns the notebook into a batch artefact; a disposable microVM per run turns it into one you can trust.
Ajay KumarBuild, Boot, Test: Kernel CI on MicroVMs (and Where It Stops)
"Does this patch boot" should be a per-commit check, not a nightly. A microVM makes the boot step sub-second and a git bisect a coffee break — right up to the device model, the nested-virt wall, and the fact that you cannot swap our guest kernel.
Ajay KumarShadow Traffic and Replay: Testing a Rewrite Against Real Requests
Nobody wrote down what the old service does, and some of its bugs are load-bearing. Production traffic is the only artefact that has never been wrong about it — which makes diffing two implementations against the same requests the only honest test you have. Then the mirrored POST charges the card twice.
Ajay KumarMutation Testing at Scale: Thousands of Broken Builds on Purpose
A test suite is a claim: that if this code were wrong, something here would fail. Mutation testing is the only honest way to check that claim — break the code on purpose, thousands of times, and see what the suite notices. Which makes it a scheduling problem.
Ajay KumarYour Dependency Bot Runs Strangers' Install Scripts
An automated dependency bot is a privileged robot on a cron whose entire job is to fetch third-party code and run it. The fix is not to stop it executing — it is to make sure the thing executing and the thing holding the token are never the same machine.
Ajay KumarTop 7 Ephemeral Development Environment Platforms in 2026
An environment is ephemeral when it is created from a definition, nobody is sad when it dies, and the 400th costs the same as the 4th. Most "cloud dev environments" fail at least one of those tests.
Ajay KumarTop 8 Throwaway Postgres Platforms for 2026
A disposable database is not a feature, it is one of three mechanisms with genuinely different properties — and the one that starts empty can only ever run your unit tests.
Ajay KumarLighthouse and axe-core Audits, One MicroVM Each
Forty Lighthouse runs sharing a kernel all report a worse LCP than the truth, and the variance swallows the regression you were hunting. The isolation here is not a security control — it is a scientific one.
Ajay KumarOne microVM per Attendee: Hands-On Labs for a Whole Room at Once
Two hundred people arrive inside the same ninety seconds, all do the same thing at the same time, and all vanish at 5pm. That load shape breaks pre-provisioning in both directions — and it is exactly what snapshot-restore-per-create is for.
Ajay KumarExtracting Untrusted Archives: Zip Bombs, Zip Slip, Symlink Escape
Archive extraction is one of the few operations where the attacker picks the amplification factor. You cannot inspect your way out of that. You can only put it somewhere you are willing to lose.
Ajay KumarA Regex Is Untrusted Code That Does Not Look Like Code
Nobody reviews a regex in a form field as code. But a nested quantifier on a backtracking engine is a denial-of-service primitive with a one-line source file — and the attack payload is usually a badly-typed email address, not a pattern.
Ajay Kumar