vllm.distributed.ec_transfer.ec_connector.metrics
¶
Classes:
-
ECConnectorLogging– -
ECConnectorProm–Support for registering per-connector Prometheus metrics, and
-
ECConnectorPromMetrics–A base class for per-connector Prometheus metric registration
-
ECConnectorStats–Base class for EC Connector Stats, a container for transfer performance
ECConnectorLogging
¶
Methods:
-
log–Log transfer metrics periodically, similar to throughput logging
Source code in vllm/distributed/ec_transfer/ec_connector/metrics.py
log(log_fn=logger.info)
¶
Log transfer metrics periodically, similar to throughput logging
Source code in vllm/distributed/ec_transfer/ec_connector/metrics.py
ECConnectorProm
¶
Support for registering per-connector Prometheus metrics, and recording transfer statistics to those metrics. Uses ECConnectorBase.build_prom_metrics().
Source code in vllm/distributed/ec_transfer/ec_connector/metrics.py
ECConnectorPromMetrics
¶
A base class for per-connector Prometheus metric registration and recording.
Methods:
-
observe–Record the supplied transfer statistics to Prometheus metrics. These
Source code in vllm/distributed/ec_transfer/ec_connector/metrics.py
observe(transfer_stats_data, engine_idx=0)
¶
Record the supplied transfer statistics to Prometheus metrics. These statistics are engine-specific, and should be recorded to a metric with the appropriate 'engine' label. These metric instances can be created using the create_metric_per_engine() helper method.
Source code in vllm/distributed/ec_transfer/ec_connector/metrics.py
ECConnectorStats
dataclass
¶
Base class for EC Connector Stats, a container for transfer performance metrics or otherwise important telemetry from the connector. All sub-classes need to be serializable as stats are sent from worker to logger process.
Methods:
-
aggregate–Aggregate stats with another
ECConnectorStatsobject. -
is_empty–Return True if the stats are empty.
-
reduce–Reduce the observations collected during a time interval to one or
-
reset–Reset the stats, clear the state.
Source code in vllm/distributed/ec_transfer/ec_connector/metrics.py
aggregate(other)
¶
is_empty()
¶
reduce()
¶
Reduce the observations collected during a time interval to one or more representative values (eg avg/median/sum of the series). This is meant to be called by the logger to produce a summary of the stats for the last time interval.