Al Brown

Product @ ClickHouse
Thinking about agentic experience & identity

[Al Brown @sdairs]

4/4 allow_block_nested_loop_join is enabled by default. EXPLAIN identifies BlockNestedLoopJoin.

[Al Brown @sdairs]

1/4 ClickHouse 26.10 (@ClickhouseDB) can run an outer join without an equality key between its inputs. LEFT JOIN products ON editDistance(q, name) <= 2 can keep unmatched typo rows while finding nearby product names.

[Al Brown @sdairs]

3/4 Work can still grow with the product of both input sizes, so filter candidate inputs before a large fuzzy match. INNER JOIN already had a cross-join-and-filter route; supporting the outer-join shape is the useful addition.

[Al Brown @sdairs]

As much as I'm enjoying @superlogical rex...I'm not using it as much as I thought I would. I'm quite enjoying the agent desktop app experience these days

[Al Brown @sdairs]

your genz aura consultant is steering you poorly @RamaswmySridhar

[Kaushik Iska @iskakaushik]

asked the Postgres team at our berlin offsite what their favorite feature is. WalShadow won with 7 votes, Philip got a vote, and i got filed as a bug 🤣

my answer is still the team.

[Kaushik Iska @iskakaushik]

What does it take to make a Postgres service GA in 10 months? For ClickHouse Managed Postgres it came down to seven things, and most of them have very little to do with Postgres itself. Here's what it took:

[Andy Pavlo (@andypavlo.bsky.social) @andy_pavlo]

Most of ClickHouse's Postgres customers are migrating from existing databases. They already use the ClickHouse DBMS, so running their Postgres instances on the same platform makes it easier to get data into the OLAP engine.
Source: @saisrirampur

[Andy Pavlo (@andypavlo.bsky.social) @andy_pavlo]

Part of why I joined @ClickHouseDB is because they do what I would do if I were running a database company.
✅Kept flagship OLAP DBMS open source. No enshitification.
✅Add verticals (agents, observability) on top of OLAP DBMS.
✅Launch Postgres service 1yr after buying PeerDB.

[ClickHouse @ClickHouseDB]

ClickHouse Managed Postgres is GA: up to 10× faster on local NVMe, sub-second CDC into ClickHouse via WalShadow, and hundreds of production customers, including @clay.

We’re also acquiring @xata, which brings agentic capabilities to the platform: instant branches and clones, automatic idling, and scale-to-zero.

Full announcement:
https://t.co/JDFBhqAj2B

[Al Brown @sdairs]

5/5 Nondeterministic functions such as rand are rejected inside the lateral subquery. The final join still consumes memory, and grouping/sorting limits cover the combined decorrelated work.

[Al Brown @sdairs]

1/5 ClickHouse 26.10 (@ClickhouseDB) adds experimental LATERAL JOIN. A subquery on the right can refer to columns on its left, making patterns such as the latest two invoices per company natural to express.

[Al Brown @sdairs]

4/5 Enable allow_experimental_lateral_join = 1 for INNER/LEFT with ALL semantics and ON true, with the correlation filter inside the subquery. CROSS, RIGHT, FULL and LATERAL unnest(...) are unsupported.

[Al Brown @sdairs]

@marcklingen @NathanFlurry @steventey @maxdeichmann hah its fun to see you explored clickhouse through that tinybird KB (I created it)

[sridhar @RamaswmySridhar]

in a world where data interoperability is a must, it's absurd to see the games dbx plays...

[Al Brown @sdairs]

5/5 The tests shift dictionary order across parts and force external aggregation to check what makes this safe. Dictionary positions are a shortcut within a block, while serialized values still define groups across blocks.

[Al Brown @sdairs]

1/5 ClickHouse 26.10 (@ClickhouseDB) speeds up GROUP BY over repeated LowCardinality dimensions. One change avoids expanding eligible dictionary keys into full columns before aggregation, saving a copy of each key column.

[Al Brown @sdairs]

4/5 The cache needs all keys to be non-nullable LowCardinality, with a dictionary-size product at most 65,536. Its array is bounded to 512 KiB per thread. An active top-K heap and lone count() use other paths.