ClickHouse’s pgclickhouse 0.11.1 release adds pgvector support, vector-search pushdown and PostgreSQL-style UPDATE, DELETE and TRUNCATE for MergeTree tables. It also brings a clutch of compatibility improvements that could make the PostgreSQL-to-ClickHouse bridge more useful in real workloads.
ClickHouse Watch analysis
What happened
The new version maps ClickHouse Array(Float32) and Array(Float64) columns to PostgreSQL vector columns. It pushes supported pgvector comparisons and distance calculations down to ClickHouse, and can detect and report when a vector-similarity index is available for a query.
The release also adds UPDATE, DELETE and TRUNCATE support for MergeTree-family tables, plus pushdown for linear-regression aggregates on ClickHouse 26.9 or later. Other changes improve handling of NULL-returning aggregates, JSONB subscripts, time zones and unconstrained NUMERIC values.
Our top picks
- pgvector mapping and search pushdown
PostgreSQL vector queries can run against ClickHouse arrays, with supported operators and distance functions pushed to the ClickHouse side. - Vector-index awareness
Query plans can report when a ClickHouse vector-similarity index is available, rather than leaving users to guess what the search can use. - UPDATE, DELETE and TRUNCATE
PostgreSQL commands now work on MergeTree-family tables, with a useful distinction: UPDATE and DELETE are designed for smaller changes, while TRUNCATE is the option for clearing a table. - More faithful aggregate results
NULL-returning aggregates now return NULL for empty input instead of a ClickHouse default such as zero, matching PostgreSQL behaviour. - JSONB and time-zone improvements
JSONB subscripts can be pushed down, and session time-zone handling is aligned with PostgreSQL on ClickHouse 23.6 and later.
Why it matters
The release makes a wider set of PostgreSQL workflows usable through ClickHouse without asking developers to abandon familiar types and operators. That matters especially for vector search: the connector can translate supported pgvector operations and pass them to ClickHouse, rather than pulling the work back to the PostgreSQL side.
There are a couple of details worth keeping close. UPDATE requires ClickHouse settings to be enabled, and PostgreSQL permits UPDATE and DELETE without a WHERE clause, so an omitted filter can affect every row. ClickHouse says the minor release reaches new and existing Managed Postgres servers automatically; installations on v0.10 or earlier need to run ALTER EXTENSION pgclickhouse UPDATE.
Our read
This is a genuinely useful connector release, not a version-number shrug. Vector search is the headline, but the quieter compatibility work may save users just as much friction. Before upgrading or changing a workload, check the version-specific requirements and be especially deliberate with destructive SQL. The database will not admire your confidence if you truncate the wrong table.
What to watch
- Which supported pgvector operations and index types work in users’ own query plans.
- Whether UPDATE and DELETE behave as expected under real workloads and configured settings.
- How quickly older installations move from v0.10 or earlier to the new extension.
Discussion spark: Would you run vector search and row-level changes through this PostgreSQL connector in production, or keep those workloads on separate paths until you have tested the query plans and failure modes?
Sources and evidence
- Source update (19 November 2026, 11:52 UTC)
- Source update (19 November 2026, 11:52 UTC)
Independent WittyWires tracker for public updates about ClickHouse. Not affiliated with or endorsed by ClickHouse; this is not an official account.