add --grouping_key for monitor (#476)

Co-authored-by: baishihao <baishihao@sensetime.com>
Co-authored-by: shihaobai <42648726+shihaobai@users.noreply.github.com>
Co-authored-by: hiworldwzj <30762946+hiworldwzj@users.noreply.github.com>

lightllm/server/api_server.py

@@ -433,6 +433,9 @@ def main():
     )
     parser.add_argument("--metric_gateway", type=str, default=None, help="address for collecting monitoring metrics")
     parser.add_argument("--job_name", type=str, default="lightllm", help="job name for monitor")
+    parser.add_argument(
+        "--grouping_key", action="append", default=[], help="grouping_key for the monitor in the form key=value"
+    )
     parser.add_argument("--push_interval", type=int, default=10, help="interval of pushing monitoring metrics")
     parser.add_argument(
         "--enable_monitor_auth", action="store_true", help="Whether to open authentication for push_gateway"

lightllm/server/metrics/metrics.py

@@ -50,6 +50,11 @@ def __init__(self, args):
         self.gateway_url = args.metric_gateway
         self.registry = CollectorRegistry()
         self.job_name = args.job_name
+        self.grouping_key = {}
+        if args.grouping_key:
+            for item in args.grouping_key:
+                key, value = item.split("=")
+                self.grouping_key[key] = value
         self.auth = args.enable_monitor_auth
         self.init_metrics(args)
 
@@ -127,6 +132,14 @@ def gauge_set(self, name, value):
     def push_metrices(self):
         if self.gateway_url is not None:
             if self.auth:
-                push_to_gateway(self.gateway_url, job=self.job_name, registry=self.registry, handler=my_auth_handler)
+                push_to_gateway(
+                    self.gateway_url,
+                    job=self.job_name,
+                    grouping_key=self.grouping_key,
+                    registry=self.registry,
+                    handler=my_auth_handler,
+                )
             else:
-                push_to_gateway(self.gateway_url, job=self.job_name, registry=self.registry)
+                push_to_gateway(
+                    self.gateway_url, job=self.job_name, grouping_key=self.grouping_key, registry=self.registry
+                )