Tessera

Dynoxide 1.0.0 is out 🎉 Vector indexes are the big one: SearchVectors does exact KNN over cosine, euclidean or dot product, with filtering, and the browser build gets it too. Plus a long run of conformance fixes! github.com/nubo-db/dyno...

GitHub - nubo-db/dynoxide: A DynamoDB emulator in Rust, backed by SQLite. Starts in milliseconds as a single static binary, and is verified against real AWS by a public conformance suite. Runs as an H...

A DynamoDB emulator in Rust, backed by SQLite. Starts in milliseconds as a single static binary, and is verified against real AWS by a public conformance suite. Runs as an HTTP server, an MCP serve...

github.com

My conformance suite measured index capacity on reads for months. The write side, where a GSI actually costs money, went unmeasured until someone filed an issue. Fourteen tests later, every number pinned. https://martinhicks.dev/articles/what-a-write-to-an-indexed-dynamodb-table-costs

What a write to an indexed DynamoDB table costs

What a write to a DynamoDB table carrying GSIs and LSIs actually costs, measured against real DynamoDB in eu-west-2 rather than lifted from the docs.

martinhicks.dev

AWS shipped DynamoDB vector search straight to GA. Its docs get readiness wrong three ways: the tutorial contradicts three other pages, the state machine names a state the index never enters, and the ready signal comes early. https://martinhicks.dev/articles/dynamodb-vector-search-docs-get-wrong

When a DynamoDB vector index is actually ready

AWS says a DynamoDB vector index is ready when IndexStatus is ACTIVE and Backfilling is false. I measured it against real DynamoDB, and that state never happens.

martinhicks.dev

Every DynamoDB emulator on my conformance board now carries a letter grade, A to F, read from two figures: how much of DynamoDB it supports, and how much of that it gets wrong. Nothing on the board holds an A+. #dynamodb #aws https://martinhicks.dev/articles/parity-suite-300-release-notes

Parity Suite 3.0.0: measuring the wrong thing

Parity Suite 3.0.0 splits the old conformance percentage into divergence and coverage, grades every DynamoDB emulator from A to F, and takes the suite to 1,054 tests with the first coverage of DynamoDB's new vector search.

martinhicks.dev

Dynoxide 0.12.0 is out. PartiQL's RETURNING clause has been parsed and then thrown away since PartiQL landed - it actually returns something now. The browser build also lands on the conformance board for the first time, at 100% on the… https://martinhicks.dev/articles/dynoxide-0120-release-notes

Dynoxide 0.12.0: PartiQL RETURNING, wasm conformance

Dynoxide 0.12.0 makes PartiQL honour the RETURNING clause it used to parse and throw away, and puts the browser build through the conformance suite for the first time as its own scored target. Plus a stack of fixes, and one breaking change for Rust crate consumers.

martinhicks.dev

My own DynamoDB emulator started failing a conformance test it used to pass. I hadn't touched the code. AWS changed a validation rule in eu-west-2, then reverted it weeks later - my suite trusted that one region as ground truth. martinhicks.dev/articles/you...

You can't pin DynamoDB to one region

A conformance suite is only as good as its ground truth. Mine used one AWS region, until DynamoDB changed its behaviour there, then reverted it, and I had to rethink what correct meant.

martinhicks.dev

Dynoxide 0.11.3 is out. I went looking at a Windows socket option for TIME_WAIT parity and found the opposite problem: the ports could be hijacked by any same-user process using SO_REUSEADDR. Both listeners now bind with… https://martinhicks.dev/articles/dynoxide-0113-windows-listener-hardening

Dynoxide 0.11.3: closing a Windows port hijack

Dynoxide 0.11.3 sets SO_EXCLUSIVEADDRUSE on the DynamoDB and MCP listeners on Windows, closing a same-user port hijack that an investigation into TIME_WAIT parity wasn't looking for, and runs the unit tests on Windows in CI for the first time.

martinhicks.dev

Dynoxide 0.11.2 is out. Last week I pinned what a conditional DynamoDB transaction costs against real AWS; this release makes the engine report the same numbers, idempotent-replay read cost and all. Plus an UpdateTable bug that… https://martinhicks.dev/articles/dynoxide-0112-conformance-fixes

Dynoxide 0.11.2: a stack of conformance fixes

Dynoxide 0.11.2 makes the engine report the transaction capacity numbers I pinned against real DynamoDB last week - the read/write split, a ConditionCheck billed as a write, an idempotent replay that recomputes a real read - plus an UpdateTable bug that dropped table keys, and a batch of projection and validation fixes.

martinhicks.dev

Does a conditional DynamoDB transaction cost read capacity on top of the write? The docs aren't clear, so I ran it against real DynamoDB. An idempotent replay flips the bill from write units to read. https://martinhicks.dev/articles/what-a-conditional-dynamodb-transaction-costs

What a conditional DynamoDB transaction actually costs

The AWS docs aren't clear on what a conditional TransactWriteItems costs in capacity units, so I ran it against real DynamoDB and pinned the numbers.

martinhicks.dev

accesspatterns.dev runs a DynamoDB engine in the browser tab, with nothing behind it. That's dynoxide 0.11, compiled to WebAssembly. The how: martinhicks.dev/articles/introducing-accesspatterns-dev #Rust

accesspatterns.dev: learn DynamoDB by running real queries

Learn DynamoDB access patterns by running real queries in your browser and seeing what each one costs. A real DynamoDB-compatible engine, backed by SQLite, runs in the tab.

accesspatterns.dev

DynamoDB is a brilliant database but can be a hard one to learn - it punishes your SQL instincts. So I built accesspatterns.dev: a real DynamoDB engine in a browser tab, with a course, access-pattern models and a full console. Learn it by running it, no account needed. #dynamodb #aws #dynoxide

accesspatterns.dev: a DynamoDB engine in your browser

A real DynamoDB-compatible engine runs in your browser, backed by SQLite. Run queries, scans and writes to see how DynamoDB behaves and which rows each one returns.

accesspatterns.dev

Dynoxide 0.11.0: the browser engine is an npm package now. @dynoxide/wasm-engine - npm install and run a DynamoDB-compatible engine client-side, no Web Worker wiring. Plus a stack of correctness fixes the conformance suite turned up. Still… https://martinhicks.dev/articles/dynoxide-0110-wasm-on-npm

Dynoxide 0.11.0: wasm on npm, plus a stack of fixes

Dynoxide 0.11.0 ships the browser engine as an npm package, @dynoxide/wasm-engine, and lands a stack of correctness fixes: secondary-index validation on write, sparse-index behaviour, and error messages brought back in line with real DynamoDB.

martinhicks.dev

My DynamoDB conformance suite went red this week and the test that failed was real AWS itself. It's reworded a chunk of its validation errors - and the change hasn't reached every region yet. Two had it, two didn't. I caught a… https://martinhicks.dev/articles/dynamodb-changed-its-validation-errors

I watched DynamoDB change under my conformance suite

My DynamoDB conformance suite went red and the test that failed was real AWS. It had reworded its validation errors, and the change hasn't reached every region yet.

martinhicks.dev

Which frontier model you build on matters. Just as important is where the capability lives in your business once it works. Not in any single function. Build the squad. Own the platform. Let the org chart catch up. https://martinhicks.dev/articles/build-the-squad-own-the-platform

Build the squad. Own the platform. The org chart will catch up.

Deploying AI agents is an organisational problem, not just a technical one. My answer: build a deployment squad and own the platform layer.

martinhicks.dev

Dynoxide 0.10.0 is out. The same DynamoDB engine now runs in a browser tab, on OPFS, no server. Plus an official ~5 MB Docker image, a drop-in for amazon/dynamodb-local. First breaking release too. #rustlang #dynamodb https://martinhicks.dev/articles/dynoxide-0100-browser-and-docker

Dynoxide 0.10.0: it runs in the browser now

Dynoxide 0.10.0 compiles to WebAssembly and runs in the browser on OPFS, ships an official ~5 MB Docker image, and adds a storage-backend trait. Plus what breaks.

martinhicks.dev

iai-callgrind needs Valgrind, and Valgrind has zero Apple Silicon support - so your instruction-count benchmarks only ever run in CI. Here's the native arm64 Docker workaround I use to run them locally. https://martinhicks.dev/articles/running-iai-callgrind-on-apple-silicon

Running iai-callgrind on Apple Silicon

iai-callgrind needs Valgrind, and Valgrind doesn't run on Apple Silicon. Here's how to run aarch64 Callgrind benchmarks in a native arm64 Docker container.

martinhicks.dev

Open-source DynamoDB emulators are closing on the real thing fast. Floci's gone from ~61% to low 90s in a month. ExtendDB landed near the top from a standing start. Wrote it up: martinhicks.dev/articles/dyn...

How close is your DynamoDB emulator to AWS?

The DynamoDB conformance suite now has a home: eight emulators and 684 tests scored against live AWS DynamoDB, with results published on every run.

martinhicks.dev

My DynamoDB conformance suite has a proper home now: dynamodb-conformance.org Eight emulators. 684 tests. One independent suite scoring all of them against DynamoDB's real-world behaviour. Runs weekly. Matrix, run history, per-target pages. Every figure pulled straight from each run.

Filed my first GitHub Security Advisory yesterday after a DNS rebinding CVE turned up in one of dynoxide's transitive deps. The fix was the easy bit. The GHSA was new ground and less daunting than I'd built it up to be. Wrote up what I learned: martinhicks.dev/articles/fil...

Filing my first security advisory

I filed my first GitHub Security Advisory for dynoxide after a DNS rebinding CVE landed via rmcp. Notes on the form, the order of operations, and the bug.

martinhicks.dev

Just shipped dynoxide 0.9.13. Closes a DNS rebinding CVE in rmcp (which powers the MCP HTTP transport) plus a related cross-origin CSRF gap. If you're running `dynoxide mcp --http`, you'll want to upgrade. github.com/nubo-db/dyno...

Release v0.9.13 · nubo-db/dynoxide

Security Close a DNS rebinding vulnerability in the MCP HTTP transport (GHSA-fvh2-gm75-j4j7 / CVE-2026-42559) by upgrading rmcp from 1.1.1 to 1.6.0 in both lockfiles. A malicious page could make ...

github.com