ClickHouse

Query Insights is now in preview for ClickHouse Cloud Managed Postgres: every query pattern your database runs, ranked by impact, with the diagnostic picture of why each one is slow.

How ClickStack and Odigos eliminate observability gaps with zero-code eBPF instrumentation and full-fidelity distributed tracing at scale.

Agentic analytics makes query-readiness a write-side cost problem. This post compares Snowflake and ClickHouse under continuous ingest, showing how ClickHouse obtains query-ready data at 22× lower cost and delivers 31× better write-side cost-performance.

We use clickhousectl to spin up multiple ClickHouse versions side by side and benchmark two recent performance improvements.

Learn how Gala migrated to the ClickHouse Cloud data platform on AWS to improve analytics performance and cut costs

How Qonto uses ClickHouse Cloud to power observability at scale — replacing sampling and hour-capped queries with two-week query windows, 99.84% compression on high-cardinality data, and an AI incident companion built on the ClickHouse MCP server.

Backward Incompatible Changes

• The IN operator now uses exact value semantics for Bool type: only 0 and 1 values in the set match Bool values. Previously, numeric values greater than 255 in the IN set were incorrectly clamped to true.
• The H3 library has been updated to v4, which improves the precision of length, area, and other metric calculations. This change is backward incompatible because the new results differ from previous ones.
• Disallows using SELECT as a bareword identifier in a WITH expression list element.
• The merge table will now handle virtuals differently: if the underlying table contains _table or _database, these columns will be read from storage; otherwise, they will be filled after the read step.
• The IN operator now rejects lossy Decimal conversions inside composite types (Tuple, Array, Map).
• Reduced default http_max_fields from 1,000,000 to 1,000 and http_max_field_name_size from 128 KB to 4 KB to limit pre-authentication memory usage by HTTP connections.

New Features

• Automatic spilling to hash and parallel hash joins by converting them to grace hash join when memory limit is reached.
• Arrow Flight SQL support.
• Incremental read support for Paimon table engines with Keeper-backed snapshot progress tracking.
• The stem function is now non-experimental.
• New behavior of max_insert_block_size_rows and related settings under compatibility setting use_strict_insert_block_limits.
• Function arrayAutocorrelation(arr [, max_lag]) for computing normalized autocorrelation of numeric arrays.
• SQL function obfuscateQuery for query obfuscation.
• Support for Map and JSON/Object types as dictionary attributes.
• New MergeTree settings replicated_fetches_min_part_level and replicated_fetches_min_part_level_timeout_seconds to reduce replication overhead.
• MergeTree skip index support for JSON columns using JSONAllPaths.
• The printf function now supports non-constant format strings.
• New projection index commit_order that reorganizes data in insertion order.
• Function highlight that wraps search terms in HTML tags.
• Quotas by normalized query hash to protect public ClickHouse services.
• Support for NATURAL JOIN syntax.
• Support SET TIME ZONE 'tz' as alias for SET session_timezone.
• Parameterized queries in the Web UI with input fields for parameter values.
• SQL standard VALUES clause as a table expression.
• PostgreSQL-compatible units for EXTRACT operator: EPOCH, DOW, DOY, ISODOW, ISOYEAR, WEEK, CENTURY, DECADE, MILLENNIUM.
• SQL-standard compound interval literals with TO range qualifiers.
• Asynchronous metrics for kernel TCP receive and transmit buffer memory.
• Jemalloc profiling web UI for ClickHouse Keeper.
• Command SYSTEM FLUSH OBJECT STORAGE QUEUE db.table PATH 'x'.
• Function JSONAllValues returning all values from JSON columns as Array(String).
• Setting input_format_column_name_matching_mode for different case sensitivities in input formats.
• watch command in clickhouse-keeper-client.
• getChildrenRecursive (ListRecursive) request to ClickHouse Keeper.
• Function arrayTranspose for transposing two-dimensional arrays.
• Auto statistics defaults: auto_statistics_types mergetree defaults to 'minmax, uniq'; materialize_statistics_on_insert defaults to false.
• Setting prefer_dependency_replica for materialized view dependency chains.
• Function hasPhrase (alias matchPhrase) for phrase search.
• Histogram metrics s3_read_request_duration_microseconds and s3_read_request_bytes.
• Date and Date32 values can now be added to Time and Time64 values using the + operator.
• Text index is now GA and stays enabled regardless of the compatibility setting.

Experimental Features

• ALTER TABLE ... EXECUTE remove_orphan_files for Iceberg tables.
• Setting query_plan_optimize_join_order_randomize for randomizing join reordering statistics.
• AI function support: aiGenerate, aiClassify, aiExtract, aiTranslate for calling OpenAI and Anthropic endpoints.
• System table system.histogram_metric_log for periodically snapshotting histogram metrics.

Performance Improvements

• ClickHouse can now prune entire data parts based on min/max statistics.
• Reduced lock contention during readonly operations on ReplicatedMergeTree tables.
• Respect optimize_read_in_order when reading projections.
• Improvements in Hash Join and Concurrent Hash Join.
• Optimize DISTINCT by disabling LowCardinality optimization when input is almost distinct.
• Performance optimization for LIKE queries using text indices.
• Vectorized math functions accelerated on AArch64 and FreeBSD/Darwin.
• Primary key pruning for regexp alternations over literals with common prefixes.
• Generalized ORDER BY ... LIMIT top-k dynamic filtering for Nullable, String, COLLATE types.
• Faster hash join on Int32 and Int64 keys with small range.
• Faster discontinuous queries for LowCardinality columns.
• Faster var*Stable and stddev*Stable functions for Float64 columns.
• Optimized base58 encode/decode operations.
• Binary size reduction and instruction cache utilization via linker optimizations.
• Fixed negative scaling for short queries with aggregation.
• Improved performance with parallel replicas.
• Prefetching support for remote file reads.
• Avoided unnecessary String .size subcolumn computation.
• Less jittery progress bar in clickhouse-client.
• MemoryWorker support in clickhouse-local for userspace page cache.
• LIMIT clause push-down optimization for UNION ALL.
• JIT compilation for String and FixedString column comparisons in ORDER BY.
• Virtual row boundary information emission for improved merge reprioritization.
• Faster Float-to-String conversion using extended itoa and zmij library.
• Faster Int128/UInt128 to string conversion.
• Avoided redundant threads in uniqExact parallel merge; batch parallel merge support.
• Better parallelization of simple views with parallel replicas.
• Parallel replicas support over simple views including UNION ALL views.
• Optimized reading in order with IN filters in full_sorting_merge.
• Cached sampling settings to optimize allocations/deallocations.
• Fixed INSERT performance regression with deduplicate_insert = 'enable'.
• Reduced profiled lock overhead.
• AVX2 and ARM NEON support for arrayDotProduct.
• Improved INSERT VALUES performance for Map, Array, Tuple types.
• Fixed excessive RabbitMQ table engine CPU usage.
• JOIN order optimizer now infers transitive equi-join predicates.
• TRUNCATE DATABASE TABLES LIKE optimization with parallel merge cancellation.
• Monotonicity support for multiply in primary key pruning.
• Cache dictionaries no longer take exclusive lock in hasKeys.
• Inlined VIEW subqueries for more optimizations.
• Optimized cache loading on server startup.
• Lazy column materialization for ReplacingMergeTree with FINAL.
• Re-enabled optimize_rewrite_array_exists_to_has optimization with compatibility fixes.

Improvements

• Improved EXPLAIN PLAN output with better formatting and join information.
• Added MergeTree setting share_nested_offsets for independent nested column handling.
• Multiple authentication methods in users.xml/yaml configuration.
• Auto reload Raft inter-node connections using TLS.
• Extended cast_keep_nullable to work with Dynamic/JSON types.
• Reduced memory footprint from internal ISerialization objects via object pool.
• Support for password and identity fields in keeper-client XML config.
• Improved Iceberg writes for unity catalog.
• Setting finalize_projection_parts_synchronously for synchronous projection finalization.
• Added projections_duration_ms column to system.part_log.
• Improved query cancellation via KILL QUERY and clickhouse-client Ctrl+C.
• Replaced hardcoded source_table_engines with runtime lookup.
• Setting for type mismatch behavior in Variant and Dynamic.
• Improved Iceberg and Spark compatibility with consistent path handling.
• Fixed race condition in IPartitionStrategy::cached_result.
• ClickHouse Interval datatypes can now be written in Arrow Format.
• Native support for UUID data types in Arrow and Parquet formats.
• Support for 7z archives on object storage.
• ProfileEvents for object storage introspection.
• Fixed merge table function with non-present columns.
• Included commit time in mutation execution time metric.
• Write-ahead log for blob objects pending removal.
• Disabled AI SQL generation in embedded client for security.
• Changed Iceberg insert interface with deprecated settings.
• Tabs rendered as 4 spaces in clickhouse-client.
• Avoid scanning remote data lake catalog for table hints when disabled.
• Applied distributed_index_analysis_min_indexes_bytes_to_activate after partition pruning.
• Fixed Parquet bloom filter push-down assertion with empty IN/NOT IN clauses.
• MinMax column statistics now store values as Field with type name serialization.
• Updated cppkafka with Consumer close deadlock fix.
• Iceberg file object information now includes row count and size from manifest.
• Configuration parameter use_separate_cache_arena for cache memory arena separation.
• Native support for Arrow StringView and BinaryView data types.
• Hot-reloaded Keeper settings: max_requests_batch_size, max_requests_batch_bytes_size, etc.
• Incremented profile events MemoryAllocatedWithoutCheck in release build.
• Cgroupv2 memory tracking excludes slab_reclaimable.
• use_partition_pruning = 0 disables MinMax index pruning and count optimization.
• pretty=1 in EXPLAIN prints expressions in human readable format.
• accurateCastOrNull and accurateCastOrDefault support Tuple target types.
• Fixed chart duplication in Play UI theme switching.
• Updated chdig tool to newer versions with perfetto UI and other improvements.
• Trailing comma support in WITH clause before SELECT.
• Setting compress_per_column_in_compact_parts for compressed block organization.
• Improved Play UI with better table info balloon and engine-specific icons.
• Support for Nullable(Tuple) in Arrow, ArrowStream, ORC, Parquet formats.
• TOTALS row displayed as table footer in web UI.
• Multi-query mode in web UI with parallel SELECT execution.
• Fixed column resize in web UI result table.
• Limit on jemalloc profile flushes per time interval.
• Keeper settings hot-reload: nuraft_streaming_mode, nuraft_max_log_gap_in_stream.
• CGroupMemoryUsedWithoutPageCache async metric.
• Parser-level syntactic sugar for SQL standard OVERLAY function.
• Column alias INDEX_LENGTH in information_schema.tables.
• information_schema.tables now ignores inactive table parts.
• ngrams function now rejects invalid ngram lengths.
• FIPS-specific test filtering and improvements.
• Expandable cells in Web UI with three-line height limit.
• Option to force virtual/path style for S3 endpoints.
• restore_replace_external_engines_to_null now skips external engine databases.
• Text index analysis support for hasPhrase function.
• STATISTICS treated as read-only in ColumnDependency.
• system.asynchronous_metric_log creation in keeper-as-server mode.
• Configuration option default_system_log_flush_policy.skip_alias_columns.
• Auto statistics disabled for system tables.
• array tokenizer support for LIKE optimization.
• MemoryAllocatedWithoutCheck sent in release builds.
• Per-thread untracked_memory exposed in system.stack_trace.

Bug Fixes

• Fixed Block structure mismatch in stream error from Lazy materialization.
• Fixed logical error with data masking policy query on CLUSTER.
• Fixed bug when using Unity catalog on GCS.
• DataLakeCatalog respects server's http_forbid_headers configuration.
• Fixed N+1 HeadObject calls for S3 brace-expansion globs.
• Validate setting changes in create queries.
• Fixed ALTER TABLE UPDATE/DELETE with MATERIALIZED columns depending on EPHEMERAL columns.
• Credentials in JDBC, ODBC, NATS connection strings are now masked.
• Fixed parseDateTimeBestEffort incorrectly parsing words with month prefixes.
• Fixed ignoring TABLE_UUID_MISMATCH for non-analyzer.
• Fixed explicit settings ignored when matched server default with compatibility setting.
• Fixed numbers with leading zeros in hive partitioning causing errors.
• Fixed heap-use-after-free when table dropped concurrently with read query.
• Fixed Keeper read request stuck when unrelated session closed at wrong moment.
• Validate column structure before applying patches.
• Fixed vertical merge assertion with Dynamic columns and SYSTEM STOP/START MERGES.
• Fixed incorrect partition pruning for toWeek() function.
• Fixed exception in functions with ColumnReplicated from JOIN.
• Fixed CLEAR COLUMN not rebuilding projections and materialized columns.
• Fixed exception in ConditionSelectivityEstimator with scalar query parameters and statistics.
• Fixed parts with unknown projections marked as lost forever.

Google Cloud Next is one of the biggest moments in the cloud calendar, and this year ClickHouse made a splash. Not just with a booth, house party and some swag, but a wave of product launches, integrations, and a deepening partnership with Google Cloud.

～国内企業のリアルタイムデータ活用を加速させる、パートナーエコシステムを構築～

世界最高速レベルのデータプラットフォームとAIエージェント運用監視を統合、次世代のAIデータスタックを構築