Caching

Apache Doris provides multi-layer caching to accelerate queries: result-level caching reuses query results for identical SQL, segment-level caching reuses computation results of filter conditions, and file-level caching offloads remote lakehouse data, and internal table data in compute-storage decoupled mode, to local disks. Choose the appropriate caching strategy based on your workload characteristics.

Result and Filter Caches

SQL Cache

Caches complete query results based on SQL text and table/partition versions. Suitable for T+1 or low-frequency update scenarios. When a cache hit occurs, the entire computation process can be skipped.

Condition Cache

Caches the results of filter conditions on segments. Repeated queries with the same conditions can directly reuse the cached results, avoiding redundant scans and filter computations.

External Table File Cache

Data Cache

Caches files from HDFS and object storage to local disks, accelerating queries on Hive, Iceberg, Hudi, Paimon, and other external tables. Supports cache warmup, quotas, and admission control.

Compute-Storage Decoupled File Cache

File Cache (Compute-Storage Decoupled)

Caches internal table data from object storage on BE local disks in compute-storage decoupled mode. Covers the multi-queue LRU, cache warmup, cache quotas, and TTL policies.

Peer Cache Read

On a local File Cache miss, read the block from another BE (including other compute groups) before falling back to object storage. Speeds up scaling and cross-group cold reads.