Expand description
Observability glue: install [atomr_telemetry::TelemetryExtension]
on a host ActorSystem and expose a small set of GPU-specific
probes that callers feed from kernel actors / placement actors /
stream allocators.
Probes are designed to be callable from anywhere: the helpers
here just look up the installed extension via
TelemetryExtension::from_system(...) and update an internal
counter. When telemetry isn’t installed the call short-circuits.
The dashboard at atomr-dashboard/ (in the atomr workspace)
consumes the resulting [atomr_telemetry::dto::NodeSnapshot] over
WebSocket — point it at any atomr-accel-cuda host and the GPU probes
show up automatically alongside the standard actor / cluster /
sharding panels.
Structs§
- GpuProbe
State - GpuProbes
- In-memory counters for the GPU-specific probes. The probes are
passive: kernel actors / stream allocators bump the counters
directly; the dashboard polls via
GpuProbes::snapshot.
Functions§
- install
- Convenience helper: construct + install a telemetry extension with
sensible defaults (1024-deep broadcast bus). Returns the shared
Arcso the caller can register exporters or read snapshots.