fix(settings): make DATA_UPLOAD_MAX_MEMORY_SIZE env-configurable

.envs/.local/.django

@@ -15,6 +15,10 @@ REDIS_URL=redis://redis:6379/0
 # NATS
 NATS_URL=nats://nats:4222
 
+# Upload size limit (integer, binary MiB — multiplied by 1024*1024). 100 MiB default
+# is fine for local dev. Raise only if testing large ML result payloads locally.
+# DJANGO_DATA_UPLOAD_MAX_MEMORY_MB=100
+
 # Celery / Flower
 CELERY_FLOWER_USER=QSocnxapfMvzLqJXSsXtnEZqRkBtsmKT
 CELERY_FLOWER_PASSWORD=BEQgmCtgyrFieKNoGTsux9YIye0I7P5Q7vEgfJD2C4jxmtHDetFaE2jhS7K7rxaf

.envs/.production/.django-example

@@ -27,6 +27,17 @@ REDIS_URL=redis://redis:6379/0
 # Tune higher only after confirming postgres connection-pool headroom.
 CELERY_WORKER_CONCURRENCY=16
 
+# Django
+# ------------------------------------------------------------------------------
+
+# Maximum in-memory request body size (integer, binary MiB — multiplied by 1024*1024).
+# Default is 100 MiB. Raise if ADC workers posting large ML result batches (detection +
+# classification payloads) return HTTP 413. global_moths_2024 batches have been
+# observed at 139–321 MiB on staging. Must be raised in lockstep with nginx client_max_body_size
+# on the fronting proxy — Django will never see a request nginx rejects.
+# See RolnickLab/antenna#1223 for the longer-term fix (incremental result posting).
+DJANGO_DATA_UPLOAD_MAX_MEMORY_MB=512
+
 # Flower
 CELERY_FLOWER_USER=
 CELERY_FLOWER_PASSWORD=

config/settings/base.py

@@ -458,8 +458,16 @@ def _celery_result_backend_url(redis_url):
 CELERY_BROKER_CONNECTION_MAX_RETRIES = None  # Retry forever
 
 
-# Allow large request bodies from ML workers posting classification results
-DATA_UPLOAD_MAX_MEMORY_SIZE = 100 * 1024 * 1024  # 100MB (default 2.5MB)
+# Maximum in-memory request body size for multipart form data and request.body access.
+# ML detection+classification payloads for a single batch can exceed the Django
+# default (2.5 MB). Configurable via env (integer, binary MiB — multiplied by
+# 1024*1024) so operators can tune without a code change. See RolnickLab/antenna#1223
+# for the longer-term fix (worker-side incremental result posting).
+#
+# Note: this setting does NOT apply to DRF JSON bodies — DRF parsers read from the
+# raw WSGI stream, bypassing request.body where Django enforces this limit.
+# nginx client_max_body_size is the hard cap for all request types.
+DATA_UPLOAD_MAX_MEMORY_SIZE = env.int("DJANGO_DATA_UPLOAD_MAX_MEMORY_MB", default=100) * 1024 * 1024
 
 # django-rest-framework
 # -------------------------------------------------------------------------------