htranx commited on
Commit
4e8b91e
·
1 Parent(s): 987ad28

one image size cap for every division, 200 GB (#18)

Browse files

- one image size cap for every division, 200 GB (1647060331406bab8765d9203faabd04784917fe)

Files changed (2) hide show
  1. config.py +11 -10
  2. tests/test_submit_test.py +3 -1
config.py CHANGED
@@ -108,16 +108,17 @@ MAX_TEST_SUBMISSIONS_PER_TRACK = 3
108
  # count_prior_submissions() in the organizers' test_phase_worker.py.
109
  TEST_SLOT_STATUSES = ("PENDING", "FINISHED", "FAILED")
110
 
111
- # Per-division image size ceiling, enforced here at registration and again by the
112
- # worker from the ECR manifest before any pull. Keep in step with
113
- # DEFAULT_SIZE_CAPS_GB in test_phase_worker.py and participant/validate_image.sh.
114
- TEST_IMAGE_SIZE_CAPS_GB = {"small": 10, "large": 200}
115
-
116
- # The size we publish to participants: one number for both divisions, stated as
117
- # the compressed size in the registry (the number `docker push` reports), not the
118
- # uncompressed size on disk. Over it, a team asks for an exception rather than
119
- # being blocked by the form.
120
- TEST_IMAGE_ADVERTISED_GB = 100
 
121
 
122
 
123
  # A registry reference we are willing to pull from. Anchored on purpose: the
 
108
  # count_prior_submissions() in the organizers' test_phase_worker.py.
109
  TEST_SLOT_STATUSES = ("PENDING", "FINISHED", "FAILED")
110
 
111
+ # The image size ceiling, one number for every division, stated as the compressed
112
+ # size in the registry (the number `docker push` reports), not the uncompressed
113
+ # size on disk. Over it, a team asks for an exception rather than being blocked
114
+ # by the form. Nothing in this Space enforces it; the organizer worker rejects
115
+ # from the ECR manifest before any pull.
116
+ #
117
+ # It used to be two numbers. A per-division dict said 10 GB for small while this
118
+ # published 100 GB to everyone, and the dict was never actually read here, so a
119
+ # small-division team could follow the published figure and still be rejected at
120
+ # scoring time. One number now, with no per-division split to drift back into.
121
+ TEST_IMAGE_ADVERTISED_GB = 200
122
 
123
 
124
  # A registry reference we are willing to pull from. Anchored on purpose: the
tests/test_submit_test.py CHANGED
@@ -157,7 +157,9 @@ class ValidateOnlyTest(unittest.TestCase):
157
 
158
  The organizers publish ONE number for both divisions and the form does
159
  not refuse an over-size image, so the message must state the advertised
160
- size rather than TEST_IMAGE_SIZE_CAPS_GB[division]. It must also say
 
 
161
  WHICH quantity that number is: the participant side measures the
162
  uncompressed image on disk and the organizer side reads the compressed
163
  size from the registry, and the two differ by enough to mislead.
 
157
 
158
  The organizers publish ONE number for both divisions and the form does
159
  not refuse an over-size image, so the message must state the advertised
160
+ size rather than a per-division one. That per-division dict is gone as
161
+ of 2026-08-08: it read 10 GB for small while this published 100 to
162
+ everyone, and nothing here ever read it. It must also say
163
  WHICH quantity that number is: the participant side measures the
164
  uncompressed image on disk and the organizer side reads the compressed
165
  size from the registry, and the two differ by enough to mislead.