Set up the process collector during metrics __init__; that way all split-process workers have it

synapse/app/homeserver.py

@@ -52,7 +52,6 @@ from synapse.config.homeserver import HomeServerConfig
 from synapse.crypto import context_factory
 from synapse.util.logcontext import LoggingContext
 from synapse.metrics import register_memory_metrics, get_metrics_for
-from synapse.metrics.process_collector import register_process_collector
 from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
 from synapse.replication.resource import ReplicationResource, REPLICATION_PREFIX
 from synapse.federation.transport.server import TransportLayerServer
@@ -338,7 +337,6 @@ def setup(config_options):
         hs.get_replication_layer().start_get_pdu_cache()
 
         register_memory_metrics(hs)
-        register_process_collector(get_metrics_for("process"))
 
     reactor.callWhenRunning(start)
 

synapse/metrics/__init__.py

@@ -24,6 +24,7 @@ from .metric import (
     CounterMetric, CallbackMetric, DistributionMetric, CacheMetric,
     MemoryUsageMetric,
 )
+from .process_collector import register_process_collector
 
 
 logger = logging.getLogger(__name__)
@@ -121,6 +122,8 @@ reactor_metrics.register_callback(
     "gc_counts", lambda: {(i,): v for i, v in enumerate(gc.get_count())}, labels=["gen"]
 )
 
+register_process_collector(get_metrics_for("process"))
+
 
 def runUntilCurrentTimer(func):