blog

Blog — page 17 of 39

Posts 385–408 of 928, newest first.

10 min read

fsfreeze and the Three Tiers of Snapshot Consistency

Copying a live disk gives you a crash-consistent image — a polite way of saying you pulled the power cord and hope the journal is forgiving. Here's the ladder from there to application-consistent, and the fsfreeze mechanics that get you up a rung.

snapshotsfilesystemconsistencyAjay Kumar
9 min read

IPv6 for Firecracker MicroVM Fleets

Every microVM networking guide is IPv4, TAP, NAT, a /30 per guest. That works until you run out of address space — or until someone notices your carefully written egress allowlist only speaks one address family. Here's the IPv6 version, including the parts snapshots make awkward.

networkingfirecrackermicrovmAjay Kumar
11 min read

The best Rust hosting platforms in 2026

Hosting a Rust service is backwards from every other language: the artifact is the easiest thing in the world to run, and getting to it can take eight minutes and more RAM than your builder has. Evaluate the build machine first.

rusthostingapp-hostingAjay Kumar
10 min read

The best .NET and ASP.NET Core hosting platforms in 2026

.NET left Windows behind years ago, but a lot of hosting advice never got the memo. The decisions that actually matter now are your publish model, JIT warmup, the GC's appetite in a memory-capped box, and whether your platform tolerates a WebSocket that lives for an hour.

dotnetaspnet-corehostingAjay Kumar
10 min read

The best MongoDB Atlas alternatives in 2026

Half the people shopping for a MongoDB Atlas alternative want another document database. The other half want Postgres and have not admitted it yet. Sorting out which half you are in is the whole exercise.

mongodbatlaspostgresAjay Kumar
10 min read

Cloudflare Sandbox SDK vs E2B for Agent Code

Cloudflare's Sandbox SDK binds code execution to Workers and Durable Objects. E2B is a standalone sandbox API you can call from anywhere. Same shortlist, genuinely different bets — here's how to tell which one your architecture wants.

comparisonsandboxesai-agentsAjay Kumar
9 min read

The best Aiven alternatives in 2026

Most teams land on Aiven for managed Postgres and end up paying for a multi-service data platform they use a tenth of. Here's how the alternatives actually differ.

postgresmanaged-databasesalternativesAjay Kumar
9 min read

The best Hono hosting platforms in 2026

Hono's selling point is that the same routes run on five runtimes. That makes it unusually easy to host and unusually easy to pick the wrong target.

honohostingapp-hostingAjay Kumar
9 min read

The best Angular hosting platforms in 2026

Half the Angular deploy problems people hit come from treating an SSR build like a static one. Check what your build actually emits, then pick the host.

angularhostingapp-hostingAjay Kumar
9 min read

The best Koyeb alternatives in 2026

Koyeb sits in a crowded middle: nicer than raw Kubernetes, more infrastructure-shaped than Vercel. Here's how the neighbours differ on the things that actually bite.

koyebapp-hostingalternativesAjay Kumar
10 min read

How to add pgvector to a Postgres database

Adding pgvector is one statement. Getting queries to use the index — and getting the distance operator to match how the index was built — is where people lose an afternoon.

postgrespgvectorhow-toAjay Kumar
10 min read

How to tail logs and debug a running app

"Check the logs" is unhelpful advice when there are three kinds. Here's how to tell which one holds the answer, before you start guessing.

debuggingobservabilityhow-toAjay Kumar
10 min read

How to add OpenTelemetry tracing to a deployed app

Tracing answers the question logs can't: where did the 3 seconds go? Here's the shortest path from a running app to a trace you can read.

opentelemetryobservabilityhow-toAjay Kumar
9 min read

How to give a LlamaIndex agent a code execution tool

LlamaIndex agents are great at retrieval and bad at arithmetic. A code tool fixes that, and the interesting part is what you hand back to the model.

llamaindexai-agentshow-toAjay Kumar
9 min read

How to give a Pydantic AI agent a code execution tool

Pydantic AI's dependency injection is the right place to put a sandbox. Here's the tool, the lifecycle, and how ModelRetry turns errors into progress.

pydantic-aiai-agentshow-toAjay Kumar
10 min read

How to use Prisma with a managed Postgres database

Prisma with managed Postgres works perfectly until you put a pooler in front of it. Here's the configuration that avoids every version of that afternoon.

prismapostgreshow-toAjay Kumar
10 min read

From Prompt Injection to RCE: The Agent Tool-Call Attack Chain

Prompt injection isn't a chatbot party trick. Give the agent a shell tool and injected text becomes a command running with your service account's credentials. Here's the chain, link by link, and the defenses ranked by what they actually buy you.

securityprompt-injectionai-agentsAjay Kumar
10 min read

node:vm Is Not a Sandbox (And Neither Was vm2)

vm.runInNewContext looks like a sandbox, is named like a sandbox, and is documented as not being one. Here's what it actually gives you, why the escape class is structural rather than a bug, and what boundary to reach for instead.

nodejsjavascriptsecurityAjay Kumar
10 min read

Shrinking PCI DSS Scope With Per-Transaction microVMs

Scope is the cost function of PCI DSS, and a shared kernel turns your segmentation boundary into a twenty-page argument that one config change invalidates. Here is the per-transaction microVM shape — and an honest list of what it does not buy you.

pci-dsscompliancemicrovmAjay Kumar
9 min read

Honeypots on microVMs: a machine you expect to lose

A high-interaction honeypot is, by definition, a machine you expect an attacker to root. Putting one on a kernel you share with anything else is a strange decision to make on purpose.

securityhoneypotdeceptionAjay Kumar
9 min read

Customer-Authored dbt Runs, One microVM Per Tenant

A dbt project looks like SQL and YAML. It is Jinja macros, run-operation, hooks, and packages fetched from wherever — all executing on your worker, against whatever connection profiles.yml hands them.

dbtdata-platformmulti-tenantAjay Kumar
11 min read

KSM Memory Deduplication for MicroVMs — And Why It's a Trap

Eighty microVMs running the same kernel and the same libc, and the host is storing eighty copies. KSM will merge them — for a permanent CPU tax, a merge that evaporates on first write, and a timing oracle across tenant boundaries. Here's how ksmd actually works, and why sharing at the snapshot layer beats scanning for duplicates you should never have made.

ksmmemoryfirecrackerAjay Kumar
10 min read

overlayfs Inside the Guest: Read-Only Rootfs, Writable Upper Layer

Mount the template read-only, stack a writable layer on top, let the kernel merge them: overlayfs is three lines of mount options and a surprising number of sharp edges. Whole-file copy_up, whiteouts, metacopy, inode weirdness — and why PandaStack does its copy-on-write below the guest instead.

overlayfsfirecrackerrootfsAjay Kumar
9 min read

Runloop vs E2B: Choosing a Sandbox for Coding Agents

Runloop and E2B are both credible places to put a coding agent, but they're built around different units of work — a repo-shaped devbox versus a general-purpose execution sandbox. A fair head-to-head.

comparisonai-sandboxrunloopAjay Kumar