models
Pydantic v2 models for the Marianne system profiler.
Defines configuration, per-process metrics, GPU metrics, system snapshots, process lifecycle events, anomalies, and resource estimates used by the profiler collector, storage layer, anomaly detector, and correlation analyzer.
Classes¶
Bases: str, Enum
Severity level for detected anomalies.
Bases: str, Enum
Types of resource anomalies the detector can identify.
Attributes¶
class-attribute
instance-attribute
¶
RSS increased >threshold in recent window.
class-attribute
instance-attribute
¶
Child process consuming >threshold CPU for extended duration.
class-attribute
instance-attribute
¶
One or more zombie child processes found.
class-attribute
instance-attribute
¶
Process approaching file descriptor limits.
Bases: str, Enum
Process lifecycle event types.
Bases: BaseModel
Resource metrics for a single process in a snapshot.
Bases: BaseModel
Per-GPU resource snapshot at a point in time.
Bases: BaseModel
Progress snapshot for a single running job.
Populated from CheckpointState in _live_states during
collect_snapshot(). Consumed by the TUI jobs panel for
accurate progress bars.
Bases: BaseModel
Point-in-time system resource snapshot.
Collected periodically by ProfilerCollector, stored in SQLite + JSONL, and consumed by AnomalyDetector and CorrelationAnalyzer.
Bases: BaseModel
Lifecycle event for a child process (spawn, exit, signal, kill, oom).
Bases: BaseModel
A detected resource anomaly.
Produced by AnomalyDetector when heuristic thresholds are exceeded.
Published to EventBus as monitor.anomaly events and stored as
RESOURCE_ANOMALY patterns in the learning system.
Bases: BaseModel
Scheduling hint based on learned resource correlations.
Returned by BackpressureController.estimate_job_resource_needs() to inform job admission and scheduling decisions.
Bases: BaseModel
Data retention policy for profiler storage.
Bases: BaseModel
Thresholds for anomaly detection.
Bases: BaseModel
Configuration for the periodic correlation analyzer.
Bases: BaseModel
Top-level profiler configuration for DaemonConfig.profiler.
Controls data collection, storage, anomaly detection thresholds, and correlation analysis frequency.