Module resource

Structs§

ByteLruCache: Byte-limited LRU cache with an optional entry cap.
CgroupMemoryAvailability: Reclaim-aware headroom under the binding finite cgroup ancestor.
CgroupMemoryProbeFailure: Fail-closed evidence from an active cgroup controller probe.
Governed: A value whose live memory is coupled to a process-wide reservation.
MaterializationPolicy
MemoryAvailability: Provenance-preserving memory availability for the current process.
MemoryGovernor: Process-wide byte-accounting governor for large allocations.
MemoryReservation: RAII guard for bytes reserved on a MemoryGovernor ledger; dropping it releases the reservation. Hold it exactly as long as the accounted allocation is live.
ProblemHints: Hints that flip strict mode on regardless of n/p — used when a code path is structurally operator-only and any dense fallback would be a bug.
RayonSafeOnce: Lazy-init cache safe to call from inside rayon par_iter.
ResourcePolicy

CgroupMemoryLimit: The exact syntax of one cgroup-v2 memory.max value.
CgroupMemoryObservation: The process’ typed cgroup memory provenance.
CgroupMemoryProbeFailureKind: Why a live cgroup memory hierarchy could not be observed safely.
DerivativeStorageMode
MatrixMaterializationError
MemoryAvailabilitySource: Which observation is the binding ceiling on memory available to this process.
MemoryReservationError: Typed refusal from MemoryGovernor::try_reserve.

LIBRARY_ROW_CHUNK_TARGET_BYTES: The library-default streamed row-chunk target (8 MiB), shared as a const so compile-time consumers (e.g. device tile geometry) stay in lockstep with ResourcePolicy::default_library without a runtime policy query.
OWNED_DATA_CACHE_MAX_ENTRIES

dense_f64_bytes: Checked byte footprint of a dense nrows × ncols f64 matrix.
detect_memory_availability: Refresh and return the OS and cgroup observations that govern process memory admission. This is the single system-memory authority shared by the global governor and specialized runtime planners.
prediction_chunk_rows: Select the row count for prediction-time covariance work.
rows_for_target_bytes: Returns how many rows to stream per chunk so that each chunk uses approximately target_bytes given a row width of cols f64 entries.