Better name for the setting as this is sort of decoupled from overflo…

…w (but also not really)

posthog/api/capture.py

@@ -624,7 +624,7 @@ def capture_internal(
     candidate_partition_key = f"{token}:{distinct_id}"
     if (
         not historical
-        and settings.CAPTURE_OVERFLOW_ENABLED
+        and settings.CAPTURE_ALLOW_RANDOM_PARTITIONING
         and (distinct_id.lower() in LIKELY_ANONYMOUS_IDS or is_randomly_partitioned(candidate_partition_key))
     ):
         kafka_partition_key = None

posthog/settings/ingestion.py

@@ -11,10 +11,12 @@
 # KEEP IN SYNC WITH plugin-server/src/config/config.ts
 BUFFER_CONVERSION_SECONDS = get_from_env("BUFFER_CONVERSION_SECONDS", default=60, type_cast=int)
 
-# Whether or not overflow (random partitioning) should be enabled *at all*.
-# Note that this setting takes precedence over other overflow-related settings
-# below, if disabled.
-CAPTURE_OVERFLOW_ENABLED = get_from_env("CAPTURE_OVERFLOW_ENABLED", True, type_cast=str_to_bool)
+# Whether or not random partitioning (i.e. overflow routing) should be enabled
+# at all. Enabling this setting this does not cause messages to be randomly
+# partitioned, but only allows them to be be randomly partitioned if other
+# criteria are met. Note that this setting takes precedence over other
+# partition-related settings below, if disabled.
+CAPTURE_ALLOW_RANDOM_PARTITIONING = get_from_env("CAPTURE_ALLOW_RANDOM_PARTITIONING", True, type_cast=str_to_bool)
 
 # A list of <team_id:distinct_id> pairs (in the format 2:myLovelyId) that we should use
 # random partitioning for when producing events to the Kafka topic consumed by the plugin server.