[platinumhamburg]

Introduce a TabletServer-level shared LRUCache for all KV tablets, controlled by kv.rocksdb.shared-block-cache.size.

• Add kv.rocksdb.shared-block-cache.enabled and .size config options
• KvManager creates/manages shared cache lifecycle
• RocksDBResourceContainer accepts optional external cache
• Adapt RocksDBStatistics to avoid double-counting shared cache in metrics
• TabletServerMetricGroup includes shared cache usage once in total
• Update config and metrics documentation

Purpose

Linked issue: close #1319

Brief change log

Tests

API and Format

Documentation

[Copilot]

Pull request overview

This PR adds an optional TabletServer-scoped shared RocksDB LRUCache to be used as the block cache across all KV tablets, aiming to better control/optimize memory usage as table count grows (Issue #1319). It also updates metrics aggregation to avoid double-counting when a cache is shared, and documents the new behavior.

Changes:

• Introduce a TabletServer-level shared RocksDB block cache created/owned by KvManager and passed down into KvTablet/RocksDBResourceContainer when enabled via config.
• Extend server-level metrics to report shared block cache usage/capacity/pinned usage once, and adjust total memory aggregation accordingly.
• Add/extend tests and documentation for the shared-cache configuration and the updated metrics semantics.

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
website/docs/maintenance/observability/monitor-metrics.md	Document new server-level shared block cache metrics and clarify total-memory aggregation semantics.
website/docs/maintenance/configuration.md	Document kv.rocksdb.shared-block-cache.size and clarify interaction with per-instance kv.rocksdb.block.cache-size.
fluss-server/src/test/java/org/apache/fluss/server/metrics/group/TabletServerMetricGroupTest.java	Add unit test ensuring server total memory counts shared cache once and shared-cache metrics are exposed.
fluss-server/src/test/java/org/apache/fluss/server/kv/rocksdb/RocksDBResourceContainerTest.java	Add tests for shared cache wiring/lifecycle and independent cache closure behavior.
fluss-server/src/test/java/org/apache/fluss/server/kv/KvTabletTest.java	Update test construction signature and strengthen RocksDB memory-metric consistency assertions.
fluss-server/src/test/java/org/apache/fluss/server/kv/KvManagerTest.java	Add test verifying shared cache is used across multiple KV tablets when configured.
fluss-server/src/main/java/org/apache/fluss/server/metrics/group/TabletServerMetricGroup.java	Add shared-cache gauges and incorporate shared cache usage into server total memory metric.
fluss-server/src/main/java/org/apache/fluss/server/kv/rocksdb/RocksDBStatistics.java	Avoid double-counting in total memory calculation and clarify semantics when cache is shared/not attributable.
fluss-server/src/main/java/org/apache/fluss/server/kv/rocksdb/RocksDBResourceContainer.java	Accept optional external/shared cache and avoid closing it from per-tablet containers.
fluss-server/src/main/java/org/apache/fluss/server/kv/KvTablet.java	Thread shared cache into RocksDB construction and omit cache from per-tablet RocksDBStatistics when shared.
fluss-server/src/main/java/org/apache/fluss/server/kv/KvManager.java	Create/own shared cache lifecycle, wire it into tablets, and register shared-cache metric suppliers.
fluss-common/src/main/java/org/apache/fluss/metrics/MetricNames.java	Add metric name constants for shared block cache usage/capacity/pinned usage.
fluss-common/src/main/java/org/apache/fluss/config/ConfigOptions.java	Add kv.rocksdb.shared-block-cache.size config option and clarify kv.rocksdb.block.cache-size behavior with shared cache.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[luoyuxia]

+1