210 lines
6.7 KiB
Python
210 lines
6.7 KiB
Python
# Copyright 2023-2024 SGLang Team
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
# ==============================================================================
|
|
"""Utilities for Prometheus Metrics Collection."""
|
|
|
|
from dataclasses import dataclass
|
|
from typing import Dict, Union
|
|
|
|
|
|
@dataclass
|
|
class SchedulerStats:
|
|
num_running_reqs: int = 0
|
|
num_used_tokens: int = 0
|
|
token_usage: float = 0.0
|
|
gen_throughput: float = 0.0
|
|
num_queue_reqs: int = 0
|
|
cache_hit_rate: float = 0.0
|
|
|
|
|
|
class SchedulerMetricsCollector:
|
|
|
|
def __init__(self, labels: Dict[str, str]) -> None:
|
|
# We need to import prometheus_client after setting the env variable `PROMETHEUS_MULTIPROC_DIR`
|
|
from prometheus_client import Gauge
|
|
|
|
self.labels = labels
|
|
|
|
self.num_running_reqs = Gauge(
|
|
name="sglang:num_running_reqs",
|
|
documentation="The number of running requests",
|
|
labelnames=labels.keys(),
|
|
multiprocess_mode="sum",
|
|
)
|
|
|
|
self.num_used_tokens = Gauge(
|
|
name="sglang:num_used_tokens",
|
|
documentation="The number of used tokens",
|
|
labelnames=labels.keys(),
|
|
multiprocess_mode="sum",
|
|
)
|
|
|
|
self.token_usage = Gauge(
|
|
name="sglang:token_usage",
|
|
documentation="The token usage",
|
|
labelnames=labels.keys(),
|
|
multiprocess_mode="mostrecent",
|
|
)
|
|
|
|
self.gen_throughput = Gauge(
|
|
name="sglang:gen_throughput",
|
|
documentation="The generate throughput (token/s)",
|
|
labelnames=labels.keys(),
|
|
multiprocess_mode="sum",
|
|
)
|
|
|
|
self.num_queue_reqs = Gauge(
|
|
name="sglang:num_queue_reqs",
|
|
documentation="The number of requests in the waiting queue",
|
|
labelnames=labels.keys(),
|
|
multiprocess_mode="sum",
|
|
)
|
|
|
|
self.cache_hit_rate = Gauge(
|
|
name="sglang:cache_hit_rate",
|
|
documentation="The cache hit rate",
|
|
labelnames=labels.keys(),
|
|
multiprocess_mode="mostrecent",
|
|
)
|
|
|
|
def _log_gauge(self, gauge, data: Union[int, float]) -> None:
|
|
# Convenience function for logging to gauge.
|
|
gauge.labels(**self.labels).set(data)
|
|
|
|
def log_stats(self, stats: SchedulerStats) -> None:
|
|
self._log_gauge(self.num_running_reqs, stats.num_running_reqs)
|
|
self._log_gauge(self.num_used_tokens, stats.num_used_tokens)
|
|
self._log_gauge(self.token_usage, stats.token_usage)
|
|
self._log_gauge(self.gen_throughput, stats.gen_throughput)
|
|
self._log_gauge(self.num_queue_reqs, stats.num_queue_reqs)
|
|
self._log_gauge(self.cache_hit_rate, stats.cache_hit_rate)
|
|
|
|
|
|
class TokenizerMetricsCollector:
|
|
def __init__(self, labels: Dict[str, str]) -> None:
|
|
# We need to import prometheus_client after setting the env variable `PROMETHEUS_MULTIPROC_DIR`
|
|
from prometheus_client import Counter, Histogram
|
|
|
|
self.labels = labels
|
|
|
|
self.prompt_tokens_total = Counter(
|
|
name="sglang:prompt_tokens_total",
|
|
documentation="Number of prefill tokens processed.",
|
|
labelnames=labels.keys(),
|
|
)
|
|
|
|
self.generation_tokens_total = Counter(
|
|
name="sglang:generation_tokens_total",
|
|
documentation="Number of generation tokens processed.",
|
|
labelnames=labels.keys(),
|
|
)
|
|
|
|
self.histogram_time_to_first_token = Histogram(
|
|
name="sglang:time_to_first_token_seconds",
|
|
documentation="Histogram of time to first token in seconds.",
|
|
labelnames=labels.keys(),
|
|
buckets=[
|
|
0.001,
|
|
0.005,
|
|
0.01,
|
|
0.02,
|
|
0.04,
|
|
0.06,
|
|
0.08,
|
|
0.1,
|
|
0.25,
|
|
0.5,
|
|
0.75,
|
|
1.0,
|
|
2.5,
|
|
5.0,
|
|
7.5,
|
|
10.0,
|
|
15.0,
|
|
20.0,
|
|
25.0,
|
|
30.0,
|
|
],
|
|
)
|
|
|
|
self.histogram_time_per_output_token = Histogram(
|
|
name="sglang:time_per_output_token_seconds",
|
|
documentation="Histogram of time per output token in seconds.",
|
|
labelnames=labels.keys(),
|
|
buckets=[
|
|
0.005,
|
|
0.01,
|
|
0.015,
|
|
0.02,
|
|
0.025,
|
|
0.03,
|
|
0.04,
|
|
0.05,
|
|
0.075,
|
|
0.1,
|
|
0.15,
|
|
0.2,
|
|
0.3,
|
|
0.4,
|
|
0.5,
|
|
0.75,
|
|
1.0,
|
|
2.5,
|
|
],
|
|
)
|
|
|
|
self.histogram_e2e_request_latency = Histogram(
|
|
name="sglang:e2e_request_latency_seconds",
|
|
documentation="Histogram of End-to-end request latency in seconds",
|
|
labelnames=labels.keys(),
|
|
buckets=[
|
|
0.3,
|
|
0.5,
|
|
0.8,
|
|
1.0,
|
|
1.5,
|
|
2.0,
|
|
2.5,
|
|
5.0,
|
|
10.0,
|
|
15.0,
|
|
20.0,
|
|
30.0,
|
|
40.0,
|
|
50.0,
|
|
60.0,
|
|
],
|
|
)
|
|
|
|
def _log_histogram(self, histogram, data: Union[int, float]) -> None:
|
|
histogram.labels(**self.labels).observe(data)
|
|
|
|
def _log_counter(self, counter, data: Union[int, float]) -> None:
|
|
# Convenience function for logging to counter.
|
|
counter.labels(**self.labels).inc(data)
|
|
|
|
def inc_prompt_tokens(self, value: int):
|
|
self._log_counter(self.prompt_tokens_total, value)
|
|
|
|
def inc_generation_tokens(self, value: int):
|
|
self._log_counter(self.generation_tokens_total, value)
|
|
|
|
def observe_time_to_first_token(self, value: Union[float, int]):
|
|
self._log_histogram(self.histogram_time_to_first_token, value)
|
|
|
|
def observe_time_per_output_token(self, value: Union[float, int]):
|
|
self._log_histogram(self.histogram_time_per_output_token, value)
|
|
|
|
def observe_e2e_request_latency(self, value: Union[float, int]):
|
|
self._log_histogram(self.histogram_e2e_request_latency, value)
|