[PD] Improve disaggregation metrics output: update the metrics to keep reflecting real stats (#7317)
This commit is contained in:
@@ -142,7 +142,7 @@ class SchedulerStats:
|
||||
spec_accept_length: float = 0.0
|
||||
avg_request_queue_latency: float = 0.0
|
||||
num_prefill_prealloc_queue_reqs: int = 0
|
||||
num_prefill_infight_queue_reqs: int = 0
|
||||
num_prefill_inflight_queue_reqs: int = 0
|
||||
num_decode_prealloc_queue_reqs: int = 0
|
||||
num_decode_transfer_queue_reqs: int = 0
|
||||
total_retracted_reqs: int = 0
|
||||
@@ -235,9 +235,9 @@ class SchedulerMetricsCollector:
|
||||
multiprocess_mode="mostrecent",
|
||||
)
|
||||
|
||||
self.num_prefill_infight_queue_reqs = Gauge(
|
||||
name="sglang:num_prefill_infight_queue_reqs",
|
||||
documentation="The number of requests in the prefill infight queue.",
|
||||
self.num_prefill_inflight_queue_reqs = Gauge(
|
||||
name="sglang:num_prefill_inflight_queue_reqs",
|
||||
documentation="The number of requests in the prefill inflight queue.",
|
||||
labelnames=labels.keys(),
|
||||
multiprocess_mode="mostrecent",
|
||||
)
|
||||
@@ -294,7 +294,7 @@ class SchedulerMetricsCollector:
|
||||
self.num_prefill_prealloc_queue_reqs, stats.num_prefill_prealloc_queue_reqs
|
||||
)
|
||||
self._log_gauge(
|
||||
self.num_prefill_infight_queue_reqs, stats.num_prefill_infight_queue_reqs
|
||||
self.num_prefill_inflight_queue_reqs, stats.num_prefill_inflight_queue_reqs
|
||||
)
|
||||
self._log_gauge(
|
||||
self.num_decode_prealloc_queue_reqs, stats.num_decode_prealloc_queue_reqs
|
||||
|
||||
Reference in New Issue
Block a user