Module range_cache 

Structs

BudgetStats

A fetch-permit budget’s current occupancy (one PriorityBudget’s worth), for the saturation sampler (object-cache-srv/src/saturation_monitor.rs).

FetchBudgetStats

Both of FetchScheduler‘s budgets’ occupancy: count (one permit per in-flight run, bounding origin-GET parallelism) and bytes (one permit per in-flight byte, bounding transient fetch memory).

RangeCache

Range-aware read cache over an origin object store.

Enums

RangeError

Errors returned by super::RangeCache that callers may want to handle distinctly.

StreamRangesCaller

Which caller is invoking RangeCache::stream_ranges, selecting which of the two distinct range_cache_get_range_error / range_cache_get_ranges_error counters is emitted on an upfront validation failure or a mid-stream fetch error. stream_ranges is the single fill path behind both get_range/ get_range_handler (Range) and get_ranges/post_ranges_handler (Ranges), so the tag keeps the two metric names emitting exactly as they did before those call sites shared one implementation.

Constants

DEFAULT_BLOCK_SIZE

DEFAULT_DEMAND_RESERVED_FETCH_PERMITS

Default number of DEFAULT_TOTAL_FETCH_PERMITS slots reserved for demand reads (never consumed by prefetch).

DEFAULT_FETCH_MEMORY_BUDGET_BYTES

Default fetch-budget cap on transient origin-GET buffer memory (256 MiB): the smallest budget at which DEFAULT_TOTAL_FETCH_PERMITS runs of DEFAULT_MAX_COALESCED_GET_BYTES each can all be in flight, so at the defaults a cold contiguous scan is throttled by parallelism, not memory.

DEFAULT_MAX_COALESCED_GET_BYTES

Default max byte span of one coalesced run GET.

DEFAULT_PROMOTE_WHOLE_BATCH

Default promotion granularity: promote only the run(s) covering a demanded block, not the whole prefetch batch.

DEFAULT_TOTAL_FETCH_PERMITS

Default total number of origin GETs allowed to run concurrently. See RangeCache::new.

DEMAND_WINDOW_BLOCKS

Number of blocks per streamed fetch window used by stream_ranges. At the default 1 MiB block size, 8 blocks (8 MiB) matches one coalesced origin GET run (DEFAULT_MAX_COALESCED_GET_BYTES), with buffered(2) giving modest pipeline overlap. This bounds peak in-flight memory per stream to roughly 2 * DEMAND_WINDOW_BLOCKS * block_size, independent of how large the requested range is.

MAX_PLAUSIBLE_OBJECT_SIZE

Upper bound on a plausible cached object size. No micromegas lake object (parquet partition or blob) approaches this; a decoded size above it means a corrupt/misdecoded cache entry, which is treated as a miss and re-resolved from origin rather than driving a catastrophic allocation.