nemo_rl.models.generation.dynamo.metrics#

Prometheus polling lifecycle for managed Dynamo workers.

Module Contents#

Classes#

DynamoMetricsSampler

Own polling thread, samples, and deterministic sampler shutdown.

Functions#

_http_get_text

parse_prometheus_metrics

Parse Prometheus text exposition into summed scalar values.

Data#

API#

nemo_rl.models.generation.dynamo.metrics.LOGGER#

‘getLogger(…)’

nemo_rl.models.generation.dynamo.metrics.DEFAULT_METRICS_EXCLUDE_PREFIXES#

(‘python_’, ‘process_’)

nemo_rl.models.generation.dynamo.metrics.CURATED_METRICS_INCLUDE_PREFIXES#

(‘dynamo_component_gpu_cache_usage’, ‘dynamo_component_inflight_requests’, ‘dynamo_work_handler_queu…

nemo_rl.models.generation.dynamo.metrics.CANONICAL_LOGGER_ALIASES#

None

nemo_rl.models.generation.dynamo.metrics._http_get_text(url: str, timeout_s: float) str | None#
nemo_rl.models.generation.dynamo.metrics.parse_prometheus_metrics(
text: str,
include_prefixes: tuple[str, ...] | None = None,
exclude_prefixes: tuple[str, ...] = DEFAULT_METRICS_EXCLUDE_PREFIXES,
) dict[str, float]#

Parse Prometheus text exposition into summed scalar values.

class nemo_rl.models.generation.dynamo.metrics.DynamoMetricsSampler(
workers: collections.abc.Sequence[dict[str, Any] | nemo_rl.models.generation.dynamo.refit.DynamoWorkerEndpoint],
*,
interval_s: float,
include_prefixes: list[str] | None,
exclude_prefixes: list[str] | None,
)#

Own polling thread, samples, and deterministic sampler shutdown.

Initialization

start() None#
_run() None#
snapshot() dict[str, Any]#
clear() None#
shutdown() None#