[Logging] add time stamp logging + test

apps/grpo/main.py

@@ -319,7 +319,7 @@ async def main(cfg: DictConfig):
             ProvisionerConfig(launcher_config=LauncherConfig(**cfg.provisioner))
         )
     metric_logging_cfg = cfg.get("metric_logging", {"console": {"log_per_rank": False}})
-    mlogger = await get_or_create_metric_logger()
+    mlogger = await get_or_create_metric_logger(process_name="Controller")
     await mlogger.init_backends.call_one(metric_logging_cfg)
     await ts.initialize(strategy=ts.ControllerStorageVolumes())
 

apps/grpo/qwen3_1_7b.yaml

@@ -16,11 +16,12 @@ rollout_threads: 1   # Recommended to set equal to policy.num_replicas
 # Observability configuration
 metric_logging:
   wandb:
-    project: "grpo-training"
-    group: "grpo_exp_${oc.env:USER}"
-    reduce_across_ranks: True
+    project: grpo-training
+    group: grpo_exp_${oc.env:USER}
+    logging_mode: global_reduce # global_reduce, per_rank_reduce, per_rank_no_reduce
+    per_rank_share_run: False
   console:
-    reduce_across_ranks: True
+    logging_mode: global_reduce
 
 # Dataset configuration
 dataset:

apps/grpo/qwen3_32b.yaml

@@ -19,11 +19,12 @@ rollout_threads: 1   # Recommended to set equal to policy.num_replicas
 # Observability configuration
 metric_logging:
   wandb:
-    project: "grpo-training"
-    group: "grpo_exp_${oc.env:USER}"
-    reduce_across_ranks: True
+    project: grpo-training
+    group: grpo_exp_${oc.env:USER}
+    logging_mode: global_reduce # global_reduce, per_rank_reduce, per_rank_no_reduce
+    per_rank_share_run: False
   console:
-    reduce_across_ranks: True
+    logging_mode: global_reduce
 
 # Dataset configuration
 dataset:

apps/grpo/qwen3_8b.yaml

@@ -12,11 +12,12 @@ off_by_n: 1 # Off by one by default
 # Observability configuration
 metric_logging:
   wandb:
-    project: "grpo-training"
-    group: "grpo_exp_${oc.env:USER}"
-    reduce_across_ranks: True
+    project: grpo-training
+    group: grpo_exp_${oc.env:USER}
+    logging_mode: global_reduce # global_reduce, per_rank_reduce, per_rank_no_reduce
+    per_rank_share_run: False
   console:
-    reduce_across_ranks: True
+    logging_mode: global_reduce
 
 # Dataset configuration
 dataset:

src/forge/controller/provisioner.py

@@ -21,6 +21,7 @@
 from forge.controller.launcher import BaseLauncher, get_launcher
 
 from forge.observability.metric_actors import get_or_create_metric_logger
+from forge.observability.utils import detect_actor_name_from_call_stack
 
 from forge.types import ProcessConfig, ProvisionerConfig
 
@@ -262,8 +263,10 @@ def bootstrap(env: dict[str, str]):
 
             self._proc_host_map[procs] = host_mesh
 
-        # Spawn local logging actor on each process and register with global logger
-        _ = await get_or_create_metric_logger(procs)
+        # Detect actor name and spawn local logging actor on each process
+        process_name = detect_actor_name_from_call_stack()
+        _ = await get_or_create_metric_logger(procs, process_name=process_name)
+
         return procs
 
     async def host_mesh_from_proc(self, proc_mesh: ProcMesh):

src/forge/env_constants.py

@@ -11,7 +11,7 @@
 
 # Force all timing methods in forge.observability.perf_tracker.py to use
 # CPU timer if False or GPU timer if True. If unset, defaults to the assigned value to the function.
-METRIC_TIMER_USES_CUDA = "METRIC_TIMER_USES_CUDA"
+METRIC_TIMER_USES_GPU = "METRIC_TIMER_USES_GPU"
 
 # Makes forge.observability.metrics.record_metric a no-op
 FORGE_DISABLE_METRICS = "FORGE_DISABLE_METRICS"

src/forge/observability/__init__.py

@@ -10,15 +10,15 @@
     LocalFetcherActor,
 )
 from .metrics import (
+    BackendRole,
     ConsoleBackend,
-    # Utility functions
     get_actor_name_with_rank,
     get_logger_backend_class,
-    # Backend classes
     LoggerBackend,
+    LoggingMode,
     MaxAccumulator,
     MeanAccumulator,
-    # Accumulator classes
+    Metric,
     MetricAccumulator,
     MetricCollector,
     MinAccumulator,
@@ -41,8 +41,12 @@
     # Performance tracking
     "Tracer",
     "trace",
+    # Data classes
+    "Metric",
+    "BackendRole",
     # Enums
     "Reduce",
+    "LoggingMode",
     # Actor classes
     "GlobalLoggingActor",
     "LocalFetcherActor",