Named Scalars

[filimonov]

Changelog category (leave one):

• Experimental Feature

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Adds named scalars - server-side cached values that you define once and reuse across queries instead of recomputing or storing them in a one-row table. Use CREATE [LOCAL|SHARED] NAMED SCALAR <name> [REFRESH EVERY N {SECOND|MINUTE|HOUR|DAY}] AS SELECT ... to define a scalar, getNamedScalar('<name>') (or getNamedScalarOrDefault('<name>', default)) to read it, and SYSTEM REFRESH NAMED SCALAR <name> to force an out-of-schedule refresh. LOCAL scalars are cached per server; SHARED scalars are coordinated cluster-wide via Keeper so every replica sees the same value and exactly one replica refreshes per tick. Refresh bodies run under SQL SECURITY DEFINER, are visible in system.processes and system.query_log (is_internal = 1), and can be interrupted with KILL QUERY. Inspect state via system.named_scalars. Gated behind the experimental setting allow_experimental_named_scalars.

Documentation entry for user-facing changes

Add server-side named, refreshable scalar values backed by either a local on-disk cache or a shared Keeper-backed cache, accessed via getNamedScalar / getNamedScalarOrDefault and surfaced in system.named_scalars.

Surface:

• DDL: CREATE [OR REPLACE] [LOCAL|SHARED] NAMED SCALAR [IF NOT EXISTS] [ON CLUSTER ...] [DEFINER = ...] [SQL SECURITY DEFINER] [REFRESH EVERY ] AS <SELECT ...> DROP NAMED SCALAR [IF EXISTS] [ON CLUSTER ...]
• Functions: getNamedScalar(name), getNamedScalarOrDefault(name, default).
• SYSTEM commands: REFRESH NAMED SCALAR ; {STOP|START} NAMED SCALAR REFRESHES [].
• system.named_scalars table (two-tier: value tier via getNamedScalar grant, operator tier via SHOW_NAMED_SCALARS).
• Profile-events / metrics: NamedScalarRefresh{Attempts,Successes, Failures,SkippedByPeer,DurationMicroseconds}; BackgroundNamedScalarRefreshPool{Task,Size}.
• Server settings: background_named_scalar_refresh_pool_size, named_scalar_definitions_path, named_scalar_definitions_zookeeper_path, named_scalar_local_cache_path, default_named_scalar_cache, max_named_scalars, named_scalar_max_value_size.
• User setting: allow_experimental_named_scalars (experimental gate).
• Access: CREATE_NAMED_SCALAR, DROP_NAMED_SCALAR, SHOW_NAMED_SCALARS, SYSTEM_REFRESH_NAMED_SCALAR, SYSTEM_NAMED_SCALAR_REFRESHES, getNamedScalar (function-execute, with getNamedScalarOrDefault alias).
• Error codes 766–771 (NAMED_SCALAR_NOT_FOUND, NAMED_SCALAR_ALREADY_EXISTS, SHARED_NAMED_SCALARS_NOT_CONFIGURED, NAMED_SCALAR_NOT_REFRESHABLE, NAMED_SCALAR_VALUE_TOO_LARGE, NAMED_SCALAR_HAS_NO_VALUE).

Architecture:

• Definitions are immutable (UUID-identified parsed records). Local scalars persist their definition on disk; shared scalars publish to Keeper. The manager dispatches reads through tryGetScalar; refreshes run on a dedicated BackgroundSchedulePool thread per server.
• Refresh bodies execute via executeQuery({.internal=true}) so they appear in system.processes (killable via KILL QUERY) and system.query_log; DROP / OR REPLACE / shutdown cancel the in-flight body via QueryStatus::cancelQuery.
• DEFINER privileges are used during refresh; setUser(definer_id) applies the definer's profile (max_execution_time, max_memory_usage, etc.) so resource limits inherit the standard policy without a bespoke cap.
• Shared scalars are coordinated by SharedNamedScalarsWatcher: a Keeper child-watch on the definitions root drives reconcile; ephemeral leases serialise refresh evaluation across replicas.
• system.named_scalars column names align with system.view_refreshes (last_refresh_time, last_success_time, next_refresh_time, exception).
• Two-tier disclosure: value-tier columns are non-Nullable / always populated for getNamedScalar grantees; operator-tier columns are NULL unless the caller holds SHOW_NAMED_SCALARS.

Tests:

• 17 stateless tests (03800–03816) covering CRUD, refresh, persistent cadence, OR REPLACE under refresh, definer database resolution, reload from disk, no-Keeper fallback, query_log/processes visibility, KILL QUERY interruption, two-tier access matrix, and orphan cleanup.
• A 712-line integration suite (test_shared_named_scalars_cluster) covering cross-node discovery, shared refresh failover, ZK session loss, restart-during-refresh, OR REPLACE racing the watcher, and drop-while-discovery-in-flight.

Documentation:

• docs/en/sql-reference/statements/create/named-scalar.md (full DDL, cache kinds, OR REPLACE, examples, when-to-use patterns, access).
• docs/en/sql-reference/functions/named-scalar-functions.md (getNamedScalar, getNamedScalarOrDefault).
• docs/en/operations/system-tables/named_scalars.md (column reference, operational signals query, refresh visibility & cancellation).
• docs/en/sql-reference/statements/system.md (SYSTEM REFRESH/STOP/START NAMED SCALAR REFRESHES).

Regression jobs to run:

• Fast suites (mostly <1h)
• Aggregate Functions (2h)
• Alter (1.5h)
• Benchmark (30m)
• ClickHouse Keeper (1h)
• Iceberg (2h)
• LDAP (1h)
• Parquet (1.5h)
• RBAC (1.5h)
• SSL Server (1h)
• S3 (2h)
• S3 Export (2h)
• Swarms (30m)
• Tiered Storage (2h)

[github-actions]

Workflow [PR], commit [d03b265]