blog

Blog — page 9 of 39

Posts 193–216 of 928, newest first.

10 min read

Per-Tenant SQLite: One File, One microVM, and No Noisy Neighbours

One SQLite file per customer gets the data model right: blast radius is a path, erasure is rm. Then you put a thousand of them in one process and discover that a file boundary is not a failure boundary.

sqlitemulti-tenancymicrovmAjay Kumar
10 min read

Bazel Remote Execution Workers on Firecracker microVMs

A remote cache poisoned by a leaky worker is a supply-chain compromise with excellent uptime — every developer pulls it, nobody rebuilds it, and the digests all check out.

bazelremote-executionbuild-systemsAjay Kumar
10 min read

Ephemeral Jenkins Agents on Firecracker MicroVMs

An eight-year-old build agent is a museum with a Jenkinsfile. Here is how to give every build a fresh microVM instead — EC2-plugin freshness at container-plugin latency.

jenkinsci-cdmicrovmAjay Kumar
11 min read

The Best Self-Hosted CI Runners in 2026

You've decided to run your own build infrastructure. Now you have to pick the runner software and, whether you meant to or not, an isolation model — because a CI runner is a machine that executes whatever is in a YAML file someone can open a pull request against.

ci-cdself-hostedcomparisonAjay Kumar
9 min read

GPU Passthrough and microVMs: The Honest Answer

The question comes up weekly, and the answer is no. Firecracker has no PCI bus and no VFIO device assignment, on purpose. Here is what to build instead.

gpufirecrackermicrovmAjay Kumar
10 min read

Observability for Machines That Live 200 Milliseconds

Every observability tool you know assumes the host is still there when you go looking. A fleet of two-second Firecracker VMs is not a scrape target — it is a stream of machines that leave no note.

observabilityloggingfirecrackerAjay Kumar
10 min read

SOC 2 for a Platform That Runs Other People's Code

An auditor is not testing whether your system is secure. They are testing whether you described a control and whether the evidence shows it operated every single time. That gap is the whole engagement.

soc2complianceauditAjay Kumar
10 min read

Running Code Over PHI Without Expanding Your HIPAA Blast Radius

A snapshot of guest memory is a full-fidelity copy of whatever PHI the process was holding, sitting in a bucket. Here is where protected health information really goes when you run code over it, and the architecture that keeps its residency measured in seconds.

hipaaphihealthcareAjay Kumar
10 min read

PSI: The Right Instrument for Memory Oversubscription

Load average and free -m cannot tell you whether a host is comfortably oversubscribed or one page fault from the OOM killer. PSI measures the only thing that matters — time lost to waiting — and that is what turns memory overcommit from a hope into an engineering decision.

linuxmemoryinternalsAjay Kumar
10 min read

Compiling User-Submitted LaTeX Without Handing Over a Shell

You added LaTeX because the typography is beautiful. You also added a macro language with a documented primitive for running shell commands, and pointed it at strangers.

latexmicrovmsecurityAjay Kumar
9 min read

A Spreadsheet Is a Programming Language Your Users Don't Call One

Your users write programs in your product every day. They spell them =SUMPRODUCT(...) instead of def main(), and that is the only difference. Then someone asks for PDF export and you shell out to an office suite.

spreadsheetsmicrovmmulti-tenantAjay Kumar
9 min read

Running Customer Git Hooks in Isolated microVMs

A git hook is a shell script your user wrote that your server agreed to run. The exploit is not the scary part — the scary part is a policy hook that greps a 4 GB monorepo on every push and takes the whole push queue with it.

gitmicrovmisolationAjay Kumar
10 min read

The Best Elasticsearch Hosting Platforms in 2026

Your search cluster is a database that has opinions about your heap. Here is where to run it, and when you should not run one at all.

comparisonelasticsearchopensearchAjay Kumar
10 min read

The best Meilisearch hosting platforms in 2026

Meilisearch is one stateful Rust binary with a memory-mapped index on a disk. That sentence explains every hosting option below, including the one where you do not host it at all.

comparisonmeilisearchsearchAjay Kumar
9 min read

The Best MinIO Hosting and S3 Alternatives in 2026

Object storage is the one service where the failure mode is not a restart, it is losing your customers' data. Here is where to actually put the bucket.

comparisonminioobject-storageAjay Kumar
10 min read

The best durable execution platforms in 2026

Durable execution means your workflow survives the process running it. It also means your bug survives, and the workflow you shipped in March is still out there making progress.

comparisondurable-executionworkflowsAjay Kumar
8 min read

How to Run a BullMQ Worker Alongside Your Node App

A few thousand jobs a day does not need a second always-on service. It needs a second process. Here is how to run BullMQ next to your Node app — including the Redis part nobody says out loud.

nodetypescriptbullmqAjay Kumar
9 min read

How to Self-Host Sentry for Error Tracking

Most people searching for this are looking at a SaaS bill and assuming the self-hosted version is one container. It is a distributed system with a bug tracker on top. Here is the honest cost, and the lighter answer.

self-hostingobservabilitysentryAjay Kumar
8 min read

Top 8 MySQL Hosting Platforms in 2026

MySQL hosting looks like a commodity until you compare connection limits, version pinning, PITR windows and replica lag. Here's what each option is optimised for — and where we don't compete.

comparisonmysqldatabasesAjay Kumar
8 min read

Top 7 Celery Hosting Platforms in 2026

A Celery worker is a process that starts and then sits there for weeks. That one sentence disqualifies half the places people try to deploy it. Here's where it does work, and what each option makes you pay for.

comparisoncelerypythonAjay Kumar
10 min read

Top 7 Keycloak Hosting Platforms in 2026

Keycloak is free the way a puppy is free. Here's an honest comparison of where to run it, and the four config mistakes that cause every redirect-loop support thread.

comparisonkeycloakidentityAjay Kumar
10 min read

Top 8 Prometheus Hosting Platforms in 2026

Vanilla Prometheus is a single-node database with a disk. Everything else on this list exists because of that sentence. Here's how the options differ.

comparisonprometheusobservabilityAjay Kumar
8 min read

The 8 Best Platform.sh Alternatives in 2026

Platform.sh gets one big thing right: a branch is an environment, with the parent's data cloned into it. Here's what you actually give up, and gain, on each of the eight places teams move to.

comparisonplatform-shupsunAjay Kumar
10 min read

The 8 Best AWS SQS Alternatives in 2026

Nobody sets out to replace SQS. You just want a job to run in forty minutes, or a priority lane, or yesterday's events back. Here are eight honest options and what each one is actually good at.

comparisonqueuessqsAjay Kumar