Module cache

Expand description

Cache for the engine.

Modules§

cache_size 🔒: Cache size of different cache value.
file_cache 🔒: A cache for files.
index 🔒
manifest_cache 🔒: A cache for manifest files.
parquet_page 🔒: Fixed byte blocks shared across row groups of an immutable Parquet file.
range_cache 🔒: Coordinates range-result cache loads and SST retirement. No I/O or decoding runs under the coordinator lock; loaders use the shared query runtime.
write_cache 🔒: A write-through cache for remote object stores.

CacheManager: Manages cached data for the engine.
CacheManagerBuilder: Builder to construct a CacheManager.
CachedSstMeta 🔒: Cached SST metadata combines the parquet footer with the decoded region metadata.
CompactSstMeta 🔒: Compact, authoritative form of one SST’s metadata.
PrefilterKey 🔒: Key for a cached prefilter result.
PrefilterRowSelector 🔒
PreparedSstMeta 🔒: Both forms produced by decoding metadata after a cache miss.
RangeResultMemoryLimiter 🔒
SelectorResultKey: Cache key for time series row selector result.
SelectorResultValue: Cached result for time series row selector.
SstDecodedMetaKey 🔒: Separates loads that require different page-index guarantees.
SstMetaKey 🔒: Cache key (region id, file id) for SST meta.
SstMetaLoadError 🔒
SstMetaLoadResult 🔒: Metadata returned by a shared load. Cache hits ignore the original load’s miss flag.
SstMetaResolution 🔒: Resolved cache source, not physical I/O. Also returned when the subsequent load fails. A caller that observed a memory miss can find a completed value at the native recheck; it reuses that value’s source without executing its initializer. This does not prove that the caller waited for an in-flight load or performed a file-cache lookup. Ordinary memory hits return before this resolution is applied. A default value means no file-cache resolution was needed (or known).

CacheStrategy: Cache strategies that may only enable a subset of caches.
SelectorResult: Result stored in the selector result cache.
SstMetaObservation 🔒: A caller’s memory observation. Holding the entry avoids a second lookup after coordination.
SstMetaPreparation 🔒: Result of decoding SST metadata and attempting to encode its compact cache entry.

FILE_TYPE 🔒: Metrics type key for files on the local store.
INDEX_TYPE 🔒: Metrics type key for index files (puffin) on the local store.
PREFILTER_RESULT_TYPE 🔒: Metrics type key for prefilter result cache.
RANGE_RESULT_CONCAT_MEMORY_LIMIT 🔒
RANGE_RESULT_CONCAT_MEMORY_PERMIT 🔒
RANGE_RESULT_TYPE 🔒: Metrics type key for range scan result cache.
SELECTOR_RESULT_TYPE 🔒: Metrics type key for selector result cache.
SST_META_DECODED_TYPE 🔒: Metrics type key for the optional decoded SST metadata acceleration tier.
SST_META_TYPE 🔒: Metrics type key for sst meta.
VECTOR_TYPE 🔒: Metrics type key for vector.

WriteCacheUploadStoreWrapper: Wraps the remote object store used by write cache uploads.

decode_sst_meta 🔒: Decodes SST metadata on the given blocking runtime without preparing a compact cache entry.
decoded_meta_cache_weight 🔒
encode_compact_sst_meta 🔒
find_shared_page_source 🔒: Finds a typed shared error through the range-reader I/O wrappers.
finish_sst_meta_preparation 🔒
infer_loaded_page_index_policy 🔒
meta_cache_weight 🔒
new_prefilter_result_cache 🔒
prefilter_result_cache_weight 🔒
prepare_sst_meta 🔒: Decodes SST metadata and attempts to encode both cache representations on the given blocking runtime.
prepare_sst_meta_sync 🔒: Synchronously prepares SST metadata. Callers must run this on a blocking runtime.
range_result_cache_weight 🔒
removal_cause_str 🔒
satisfies_page_index_policy 🔒
selector_result_cache_hit: Increases selector cache hit metrics.
selector_result_cache_miss: Increases selector cache miss metrics.
selector_result_cache_weight 🔒
strip_column_indexes 🔒
strip_region_metadata_from_parquet 🔒
update_hit_miss 🔒: Updates cache hit/miss metrics.
vector_cache_weight 🔒

CacheManagerRef
PrefilterResultCache 🔒
RangeResultCacheRef 🔒: Maps partition-range scan key to cached flat batches.
SelectorResultCache 🔒: Maps (file id, row group id, time series row selector) to SelectorResultValue.
SstDecodedMetaCache 🔒: Maps (region id, file id) to decoded SST metadata retained as an acceleration tier.
SstMetaCache 🔒: Maps (region id, file id) to compact SST metadata.
VectorCache 🔒: Maps [Value] to a vector that holds this value repeatedly.
WriteCacheUploadStoreWrapperRef