20 ans de Bigtable : la base qui a fait naître tout l'écosystème NoSQL moderne (Cassandra, DynamoDB, HBase...). Toujours l'un des plus gros systèmes de données internes chez Google. Deux décennies de choix d'archi racontées par l'équipe. go.thibaud.dev/EYHo #Databases #Infra #nosql #sql

Twenty years of Bigtable

Our teams advance the state of the art through research, systems engineering, and collaboration across Google.

research.google

This is great! DynamoDB now has full native support for vector search. I’m still holding out for vector support in DSQL though, so I can finally drop Aurora Serverless. aws-news.com/article/2026...

Amazon DynamoDB now supports real-time vector search at any scale

Amazon DynamoDB now supports native vector search with single-digit millisecond latency, enabling semantic search on operational data without requiring a…

aws-news.com

Cleared the wrong filter on a big table? Rebuilding it by hand is the usual cost. ⌘Z now restores it and re-runs the query. The same key still reopens a closed tab, so whichever you did last is what it undoes. https://dynotable.com/changelog?utm_source=bluesky&utm_medium=social&utm_campaign=drip

DynamoDB is a poor fit for analytics, ad-hoc queries, and join-heavy relational models. Knowing where it stops is half the skill. What workload made you regret choosing it? https://dynotable.com/dynamodb-faq/when-not-to-use-dynamodb?utm_source=bluesky&utm_medium=social&utm_campaign=drip

When should you not use DynamoDB?

Skip DynamoDB for analytical workloads, unknown access patterns, heavy ad-hoc querying, and normalized relational schemas that depend on joins.

dynotable.com

DynamoDB stores timestamps as epoch numbers. Most viewers show them in UTC and leave you doing the math. DynoTable now lets you set the time zone and hour format once, and read any column as relative time. https://dynotable.com/changelog?utm_source=bluesky&utm_medium=social&utm_campaign=drip

BatchWriteItem rejects the whole request past 25 items: one ValidationException, nothing written. Chunk to 25 and retry any UnprocessedItems the response hands back. https://dynotable.com/dynamodb-errors/batchwriteitem-too-many-items?utm_source=bluesky&utm_medium=social&utm_campaign=drip

On-demand or provisioned? Spiky traffic favors on-demand; steady load runs cheaper provisioned. Which carries your prod tables? https://dynotable.com/learn/dynamodb-on-demand-vs-provisioned?utm_source=bluesky&utm_medium=social&utm_campaign=drip&utm_content=diagram

ADD views :one in an update expression turns any DynamoDB item into an atomic counter: no read first, no race, concurrent increments all land. Runnable CLI version: https://dynotable.com/code-examples/updateitem-cli?utm_source=bluesky&utm_medium=social&utm_campaign=drip

One table, several entity types, one Query to hydrate a whole view. Single-table design in one picture. https://dynotable.com/learn/single-table-design?utm_source=bluesky&utm_medium=social&utm_campaign=drip&utm_content=diagram

Closed the wrong tab? ⌘Z brings it back. ⌘⇧T reopens them in order. Closing is reversible now, so DynoTable no longer stops to ask if you meant it. New too: ⌘-click a URL in a cell or the item editor and it opens in your browser. https://dynotable.com/changelog

A DynamoDB client with real SQL: JOIN, GROUP BY and aggregates compiled into Query and Scan. An AI agent on your own Bedrock keys, schema autocomplete, fast profile switching, and an MCP server for your coding agent. We're on Product Hunt today. https://www.producthunt.com/products/dynotable

A DynamoDB item caps at 400 KB, and attribute names count toward it in UTF-8 bytes. Long names cost space on every single item, and the limit cannot be raised. https://dynotable.com/dynamodb-faq/what-is-the-max-item-size-in-dynamodb?utm_source=bluesky&utm_medium=social&utm_campaign=drip

DynamoDB rounds capacity up: reads bill per 4 KB, writes per 1 KB, minimum one unit each. A 100-byte write costs the same as a 1 KB write. Item size is a pricing lever. https://dynotable.com/tools/dynamodb-pricing-calculator?utm_source=bluesky&utm_medium=social&utm_campaign=drip

The TTL mistake everyone makes: milliseconds. DynamoDB TTL expects epoch seconds. A millisecond value reads as a date thousands of years away, so the item never expires. https://dynotable.com/dynamodb-errors/ttl-attribute-not-number?utm_source=bluesky&utm_medium=social&utm_campaign=drip

DynamoDB can't join, so every cross-table question turns into a scratch file of copied JSON and a one-off script. DynoTable joins on your machine and runs real SQL over your tables. 95 seconds of it working: dynotable.com

One eval scenario sat at 0% on the cheapest model we support. We rewrote a single validator error message and it went to 100%. The model was never the bottleneck. https://dynotable.com/blog/tool-validators-evals-cheap-models?utm_source=bluesky&utm_medium=social&utm_campaign=drip

git push --force-with-lease, for DynamoDB. Each commit pins every attribute to the value you reviewed, so a concurrent edit becomes a conflict, never a silent overwrite. https://dynotable.com/blog/database-gui-staged-writes?utm_source=bluesky&utm_medium=social&utm_campaign=drip

SQL makes scans look free. DynoTable's Workbench now shows the cost before a query runs: Query or Scan, which index, an honest read estimate, and a warning when your SQL would scan the whole table. https://dynotable.com/changelog?utm_source=bluesky&utm_medium=social&utm_campaign=drip

Does DynamoDB support SQL? PartiQL looks like SQL and stops at JOIN, GROUP BY, and COUNT. We built a local engine that runs the real thing. How: https://dynotable.com/blog/local-sql-engine-for-dynamodb?utm_source=bluesky&utm_medium=social&utm_campaign=drip

AWS Config now supports 191 additional managed rules https://aws.amazon.com/about-aws/whats-new/2026/07/aws-config-additional-managed-rules

AWS Glue SAP OData connector and zero-ETL integrations are now available in AWS GovCloud (US) regions https://aws.amazon.com/about-aws/whats-new/2026/07/aws-glue-sap-zero-etl-govcloud

Amazon GameLift Streams now supports IAM role credentials for stream sessions https://aws.amazon.com/about-aws/whats-new/2026/07/amazon-gamelift-streams-iam/

Do you know one DynamoDB partition key can hold a whole sorted collection? Add a sort key and related items live together, ordered, readable in a single Query. https://dynotable.com/learn/dynamodb-composite-primary-key?utm_source=bluesky&utm_medium=social&utm_campaign=drip&utm_content=diagram

An MCP server holding AWS credentials is a different problem than a toy stdio server. Safe meant OAuth inside a desktop app, scoped consent, a region pinned at approval. https://dynotable.com/blog/building-dynamodb-mcp-server?utm_source=bluesky&utm_medium=social&utm_campaign=drip

AWS open sourced a Bulk Executor for DynamoDB: count, find, delete, update, copy, or load a whole table from one CLI. It drives Glue behind the scenes, so no custom job code. https://aws.amazon.com/blogs/database/introducing-open-source-bulk-executor-for-amazon-dynamodb/

A DynamoDB Stream is an ordered log of every insert, update, and delete on a table, kept 24 hours. Point a Lambda at it for event-driven work with no polling. The change feed is the integration point. https://dynotable.com/learn/dynamodb-streams?utm_source=bluesky&utm_medium=social&utm_campaign=drip

A DynamoDB gotcha: sort keys sort as strings, so 10 comes before 2. If a sort key is numeric, zero-pad it to a fixed width (0002, 0010) so string order matches numeric order. https://dynotable.com/learn/dynamodb-zero-padding-sort?utm_source=bluesky&utm_medium=social&utm_campaign=drip

An item collection is every item sharing one partition key, the unit a single Query reads. It emerges from your key schema, you don't declare it. Design the sort key, design the Query. https://dynotable.com/learn/dynamodb-item-collections?utm_source=bluesky&utm_medium=social&utm_campaign=drip

A GSI is a second copy of your table keyed differently, kept in sync for you. It answers queries the base key can't, for extra storage and write cost. Project only the attributes each pattern needs. https://dynotable.com/learn/gsi-vs-lsi?utm_source=bluesky&utm_medium=social&utm_campaign=drip

A DynamoDB hot partition: one key takes more traffic than its share of throughput, so it throttles while the rest of the table sits idle. The fix is key design: spread load across more keys. https://dynotable.com/learn/dynamodb-hot-partition?utm_source=bluesky&utm_medium=social&utm_campaign=drip