Module runtime 

Runtime, memory, boundary, and reservation contracts for native execution.

Values defined in this module govern observable execution behavior across crates. Policy enum string representations, reservation fact keys, and retry fact keys are exposed to tests, telemetry reports, and adapter diagnostics. These identifiers constitute stable diagnostic contracts and are updated in coordination with reporting schemas.

Re-exports

pub use cache::CacheGraphActivity;

pub use cache::CacheSourceActivity;

pub use cache::CacheSourceEventSnapshot;

pub use cache::DeviceCacheReport;

pub use cache::cache_observation_fingerprint;

Modules

cache

GPU-resident cache ownership and lifecycle contracts.

resident

Frontend-neutral GPU-resident data and reusable build handles. Frontend-neutral GPU-resident data, result, and reusable build handles.

Structs

AdmittedQueryContext

Execution context acquired from a QueryHandle.

BackendAdmissionMetricsSnapshot

Ledger-class backend metrics for one fixed device slot.

BackendAttemptMetricsSnapshot

Terminal counters and duration histograms for the complete backend.

BackendDeviceMetricsSnapshot

Metrics for one configured device. The ordinal is the bounded device-label value fixed during backend construction, never an attempt or query id.

BackendDiagnostics

One read of a live service’s state, each part taken under its owner’s lock. The device ledgers are the admission authority; cache, memory failure, and execution-pool reports are diagnostic cross-checks.

BackendHistogramProjection

An OTLP-compatible histogram projection. Bounds and sum use the descriptor-declared unit, while bucket counts are disjoint (including the final implicit +Inf bucket).

BackendHistogramSnapshot

A fixed cumulative histogram snapshot. bucket_counts follows BACKEND_ATTEMPT_DURATION_BUCKETS_NS and ends with its +Inf bucket.

BackendMetricAttribute

One descriptor-ordered attribute of a projected backend metric sample.

BackendMetricSample

One closed-label sample for a canonical backend descriptor.

BackendMetricsSnapshot

Coherent fixed-cardinality snapshot of one backend metrics generation.

CacheFillBytePermit

RAII permit for bytes charged to a cache-fill byte budget.

DeviceCapabilities

Immutable capability metadata for one installed CUDA device.

DeviceCapability

Stable identity of one device-execution capability required by a native attempt.

DeviceFacts

Snapshot of the stable facts algeon needs for one CUDA device.

DeviceLedgerGrantRow

One active grant row in DeviceLedgerReport.

DeviceLedgerReport

Public diagnostic report for one device ledger.

DeviceMemoryFailureReport

Point-in-time device memory allocation-failure report.

DeviceResourceProfile

Frozen capacity and execution-slot bounds for one CUDA device.

DeviceTelemetrySnapshot

Live CUDA and NVML telemetry for one device.

ExecutionCancellation

Shared cooperative cancellation token for native query execution.

ExecutionContext

Immutable execution configuration shared by native query execution.

GpuBackendCloseSnapshot

Two-scope close state. GPU quiescence can become true while observation permits still exist; at that point the device-owner lease has been released.

GpuBackendConfig

Backend construction input for the device set, the execution and queue policy, and the optional attempt completion queue.

GpuBackendCore

One process-owned GPU backend core.

GpuBackendDeviceConfig

Pure per-device construction value binding a profile to deployment ownership.

GpuBackendDeviceInstallationBounds

Explicit installation limits for one device profile.

GpuBackendDeviceProfile

Pure per-device construction profile validated before owner acquisition.

GpuBackendDeviceStartupRecord

Immutable per-device record frozen before backend publication.

GpuBackendSession

Opaque capability installed into session/optimizer state.

GpuDeviceCapacityProbe

Pre-build capacity probe for one CUDA device.

GpuExecutionPoolSnapshot

Point-in-time counters for one device’s GPU execution pool.

GpuExecutionQuerySnapshot

Point-in-time counters shared by a group of GPU execution pools.

GpuExecutorLease

Opaque lifecycle-bound capability for synchronous GPU work.

GpuMemoryFraction

Validated automatic-capacity fraction retained unresolved until post-claim device discovery. The private bit representation makes Eq/Hash sound. NaN, infinities, zero, and out-of-range values cannot be constructed.

IoLifecycle

One revocable I/O generation shared by native execution contexts.

KvikioProcessConfig

KvikIO’s process-global settings: one thread pool and one task size shared by local compatibility-mode reads and remote range reads, across every device in the process.

LakehouseCacheFillConfig

Service-wide resources used only by complete-object lakehouse cache fills.

QueryActivityLease

Opaque RAII token retained until query-service cleanup is complete.

QueryAdmissionRequest

Complete request-scoped metadata for one query-service admission attempt.

QueryExecutionConfig

Execution policy copied into admitted native query contexts.

QueryHandle

RAII admission handle for one native query.

QueryServiceAdmissionSnapshot

Point-in-time service admission state for bounded-queue and overload reports.

QueueBounds

Bounded admission-queue policy.

S3SourceStorageProfile

S3-compatible storage settings for source object-store fetches.

ServiceStreamPool

Service-level CUDA stream pool for short-lived native worker tasks.

SourceSchedulerPolicy

Native source scheduling controls.

SourceStorageProfile

Storage profile used by native remote reads and cache maintenance.

ValidatedGpuBackendConfig

Configuration for GpuBackendCore::new with verified rules, resolved installation bounds, and devices ordered by ordinal.

Enums

BackendAdmissionHealth

Closed admission health projected from the owning ledger.

BackendAttemptDisposition

Closed disposition dimension of terminal backend-attempt metrics.

BackendAttemptDuration

Closed duration dimension for the attempt-duration histogram.

BackendAttemptOutcome

Closed outcome dimension of terminal backend-attempt metrics.

BackendMetricAttributeValue

A typed, closed backend-metric attribute value.

BackendMetricProjectionError

Projection failures that prevent an exporter from emitting a valid descriptor-owned backend metric.

BackendMetricSampleValue

One value produced by a backend descriptor projection.

DeviceTelemetryStatus

Availability of one live device-telemetry substrate.

GpuDeviceMemoryLimit

Soft admission capacity calculation for one device.

GpuExecutionJobKind

Classifies work submitted to the backend-owned GPU execution pool.

GpuMemoryOwnership

Deployment ownership promised for one CUDA device.

KvikioCompatMode

KvikIO compatibility mode, the KVIKIO_COMPAT_MODE setting.

MemoryResourcePolicy

S3SourceStorageCredential

Credential source for an S3 source storage profile.

Constants

BACKEND_ATTEMPT_DURATION_BUCKET_COUNT

Number of cumulative histogram buckets, including the final +Inf bucket.

BACKEND_METRIC_DEVICE_CAPACITY

Maximum number of ordinal-addressable devices in one backend metrics store.

Functions

probe_device_capacity

Queries declarable capacity for one device without building a backend.