feat: add explicit experiment type to experiment config to ease analysis

benchmarks/codex/config.py

@@ -61,6 +61,9 @@ CodexDisseminationExperiment = IteratedExperiment[
 
 
 class CodexExperimentConfig(ExperimentBuilder[CodexDisseminationExperiment]):
+    # Having this here makes it easier to analyse mixed logs later
+    experiment_type: str = "codex_static_dissemination"
+
     experiment_set_id: str = Field(
         description="Identifies the group of experiment repetitions", default="unnamed"
     )

benchmarks/deluge/config.py

@@ -62,6 +62,9 @@ DelugeDisseminationExperiment = IteratedExperiment[
 
 
 class DelugeExperimentConfig(ExperimentBuilder[DelugeDisseminationExperiment]):
+    # Having this here makes it easier to analyse mixed logs later
+    experiment_type: str = "deluge_static_dissemination"
+
     experiment_set_id: str = Field(
         description="Identifies the group of experiment repetitions", default="unnamed"
     )

benchmarks/logging/sources/sources.py

@@ -58,7 +58,13 @@ class OutputManager(AbstractContextManager):
 
 
 class FSOutputManager(OutputManager):
-    """Simple :class:`OutputManager` which writes directly into the file system."""
+    """Simple :class:`OutputManager` which writes directly into the file system.
+
+    The current implementation is very simple, and might end up keeping a lot of open files.
+    Since there's locality in th logs, we can make this better without making it slower by just
+    bounding the number of open files and juggling them as an LRU cache of sorts,
+    but for now just make sure your limits are big enough.
+    """
 
     def __init__(self, root: Path) -> None:
         self.root = root