We are live with the 26.10 ClickHouse Release call!
Join us, on youtube or LinkedIn
youtube.com/watch?v=k30ikP…
(a thread)
Join us, on youtube or LinkedIn
youtube.com/watch?v=k30ikP…
(a thread)
1
2
9
1,878
We start, as always, with small and nice features.
Kicking off with syntactic sugar via a function composition operator.
Kicking off with syntactic sugar via a function composition operator.
1
0
1
205
Are you quantizing for vector search? We've added another tool to your toolbox.
Oct 22, 2026 · 3:11 PM UTC
1
0
0
121
SQL-compatible integer division?
If I divide 7/2 in ClickHouse, what do I get?
If I divide 7/2 in PostgresSQL, what do I get?
(This works up to Int256; overflow and division by zero throw)
'SET sql_compatible_division_for_integers'
If I divide 7/2 in ClickHouse, what do I get?
If I divide 7/2 in PostgresSQL, what do I get?
(This works up to Int256; overflow and division by zero throw)
'SET sql_compatible_division_for_integers'
1
0
0
88
This feature is even less easy to explain than many others...but we've use 2x less memory for hash tables (useful with GROUP BY and uniqExact)
1
0
1
111
GROUP BY k ORDER BY k LIMIT n and GROUP BY k LIMIT n keep a top-K heap. Now its boundary is sent to the table scan:
SELECT UserID, count(), any(URL)
FROM hits
GROUP BY UserID
ORDER BY UserID
LIMIT 10;
Example: read 9.45 GB → 144 MB.
SELECT UserID, count(), any(URL)
FROM hits
GROUP BY UserID
ORDER BY UserID
LIMIT 10;
Example: read 9.45 GB → 144 MB.
1
0
1
77
@ApacheParquet @ApacheIceberg This is one of Alexey's favourite optimizations.
So I'm not going to try and give commentary...You should watch it live.
So I'm not going to try and give commentary...You should watch it live.
1
0
1
82
@ApacheParquet @ApacheIceberg And now onto "something interesting" (as if everything before wasn't already)
1
0
0
78
@ApacheParquet @ApacheIceberg We've added two entirely new database engine - 'Cluster' and 'Overlay'
Imagine a cluster for product and for staging. On staging, create a Cluster engine connecting to prod and shadow these tables using Overlay.
Imagine a cluster for product and for staging. On staging, create a Cluster engine connecting to prod and shadow these tables using Overlay.
1
0
0
121
@ApacheParquet @ApacheIceberg We promised JOIN improvements in every release.
Now Alexey speaks about supporting JOINs with arbitrary ON conditions.
Now Alexey speaks about supporting JOINs with arbitrary ON conditions.
1
0
0
123
@ApacheParquet @ApacheIceberg Are you aware that ClickHouse supports recursive CTE?
We also have Materalized CTE.
Now you can nest them.
Super useful for graph traversal and tree structure traversal.
We also have Materalized CTE.
Now you can nest them.
Super useful for graph traversal and tree structure traversal.
1
0
0
104
@ApacheParquet @ApacheIceberg SQLite was already supported as a database engine, the ability to attach a table, and use a table function. This has existed for 5 years.
In 26.10 you can now use SQLite as a data format as well.
In 26.10 you can now use SQLite as a data format as well.
1
0
0
90
@ApacheParquet @ApacheIceberg Pushing the boundaries requires building experimental features.
This talk is about to get very good.
This talk is about to get very good.
1
0
0
83
@ApacheParquet @ApacheIceberg We start with support for LATERAL JOINs!
It translates the lateral join into a correlated subquery, decorrelates and everything works.
It translates the lateral join into a correlated subquery, decorrelates and everything works.
1
0
0
114
@ApacheParquet @ApacheIceberg We are also adding support for the PCO compression codec.
For the right data it can be 15% smaller and 2x faster than DoubleDelta.
For the right data it can be 15% smaller and 2x faster than DoubleDelta.
1
0
0
89
@ApacheParquet @ApacheIceberg Want to hear about someone using it already?
Read this - tomalard.github.io/posts/pcodec-i…
Read this - tomalard.github.io/posts/pcodec-i…
1
0
2
82
@ApacheParquet @ApacheIceberg Alexey is now live-demoing usage of XGBoost dictionaries (train a gradient-boosted model at dictionary load)
1
0
1
91
@ApacheParquet @ApacheIceberg We aren't stopping there.
A new query language for logs! LogsQL (the log query language of VictoriaLogs)
A new query language for logs! LogsQL (the log query language of VictoriaLogs)
1
0
4
109
@ApacheParquet @ApacheIceberg Integration with Elasticsearch as a table engine and a table function?
You can read documents from an index as a JSON column. Read-only...for now.
You can read documents from an index as a JSON column. Read-only...for now.
1
0
1
108
@ApacheParquet @ApacheIceberg We continue with some features for data lakes.
As a 'native' option, you can use the type: s3_plain_rewritable
As a 'native' option, you can use the type: s3_plain_rewritable
1
0
0
84
@ApacheParquet @ApacheIceberg What else is new in our support for @ApacheIceberg
Parquet VARIANT support, and a new implementation of the unity catalog type.
But that's not all, ClickHouse now applies deletion vectors when reading.
Parquet VARIANT support, and a new implementation of the unity catalog type.
But that's not all, ClickHouse now applies deletion vectors when reading.
1
0
0
94
@ApacheParquet @ApacheIceberg And now some bonus features...
I won't share them all.
You should watch the recording.
But, as a teaser, the start of this section is that Trino and Kusto dialects are moved into beta!
I won't share them all.
You should watch the recording.
But, as a teaser, the start of this section is that Trino and Kusto dialects are moved into beta!
1
0
0
75
@ApacheParquet @ApacheIceberg Related, do you love using PromQL?
Maybe you can just use ClickHouse as Prometheus? ;)
Maybe you can just use ClickHouse as Prometheus? ;)
1
0
0
68
@ApacheParquet @ApacheIceberg We are officially out of time? But how do you ignore the integrations!
1
0
0
47
@ApacheParquet @ApacheIceberg Also, if you would rather we run ClickHouse for you instead of running it on your own....there are a number of new amazing features on ClickHouse Cloud.
1
0
0
98