Fix duplicate leaderboard entries and clipped display names

kernelboard/api/leaderboard.py

@@ -251,7 +251,7 @@ def _get_query():
             SELECT r.*
             FROM scoped_submissions s
             JOIN LATERAL (
-                SELECT submission_id, runner, score, secret, passed
+                SELECT id, submission_id, runner, score, secret, passed
                 FROM leaderboard.runs
                 WHERE submission_id = s.id
                 OFFSET 0
@@ -284,7 +284,10 @@ def _get_query():
                 s.file_name AS file_name,
                 r.submission_id AS submission_id,
                 COALESCE(sc.submission_count, 0) AS submission_count,
-                RANK() OVER (PARTITION BY r.runner, u.id ORDER BY r.score ASC) AS rank
+                ROW_NUMBER() OVER (
+                    PARTITION BY r.runner, s.user_id
+                    ORDER BY r.score ASC, s.submission_time ASC, s.id ASC, r.id ASC
+                ) AS rank
             FROM scoped_runs r
                 JOIN scoped_submissions s ON r.submission_id = s.id
                 LEFT JOIN leaderboard.user_info u ON s.user_id = u.id

kernelboard/api/leaderboard_summaries.py

@@ -16,7 +16,7 @@
 leaderboard_summaries_bp = Blueprint("leaderboard_summaries_bp", __name__, url_prefix="/leaderboard-summaries")
 
 # Redis cache key prefix for ended leaderboard top_users
-CACHE_KEY_PREFIX = "lb_top_users:v2:"
+CACHE_KEY_PREFIX = "lb_top_users:v3:"
 
 
 # =============================================================================
@@ -354,9 +354,9 @@ def _get_query_for_ids():
                 s.user_id,
                 u.user_name,
                 r.score,
-                RANK() OVER (
+                ROW_NUMBER() OVER (
                     PARTITION BY s.leaderboard_id, r.runner, s.user_id
-                    ORDER BY r.score ASC
+                    ORDER BY r.score ASC, s.submission_time ASC, s.id ASC, r.id ASC
                 ) AS personal_submission_rank
             FROM leaderboard.runs r
             JOIN leaderboard.submission s ON r.submission_id = s.id
@@ -475,9 +475,9 @@ def _get_query():
                 s.user_id,
                 u.user_name,
                 r.score,
-                RANK() OVER (
+                ROW_NUMBER() OVER (
                     PARTITION BY s.leaderboard_id, r.runner, s.user_id
-                    ORDER BY r.score ASC
+                    ORDER BY r.score ASC, s.submission_time ASC, s.id ASC, r.id ASC
                 ) AS personal_submission_rank
             FROM leaderboard.runs r
             JOIN leaderboard.submission s ON r.submission_id = s.id

kernelboard/index.py

@@ -75,8 +75,8 @@ def index():
                 s.leaderboard_id AS leaderboard_id,
                 u.user_name AS user_name,
                 r.score AS score,
-                RANK() OVER (PARTITION BY s.leaderboard_id, r.runner, u.id
-                ORDER BY r.score ASC) AS personal_submission_rank
+                ROW_NUMBER() OVER (PARTITION BY s.leaderboard_id, r.runner, s.user_id
+                ORDER BY r.score ASC, s.submission_time ASC, s.id ASC, r.id ASC) AS personal_submission_rank
             FROM leaderboard.runs r
                 JOIN leaderboard.submission s ON r.submission_id = s.id
                 JOIN active_leaderboards a ON s.leaderboard_id = a.id

kernelboard/leaderboard.py

@@ -45,7 +45,10 @@ def leaderboard(leaderboard_id: int):
                 r.score AS score,
                 s.submission_time AS submission_time,
                 s.file_name AS file_name,
-                RANK() OVER (PARTITION BY r.runner, u.id ORDER BY r.score ASC) AS rank
+                ROW_NUMBER() OVER (
+                    PARTITION BY r.runner, s.user_id
+                    ORDER BY r.score ASC, s.submission_time ASC, s.id ASC, r.id ASC
+                ) AS rank
             FROM leaderboard.runs r
                 JOIN leaderboard.submission s ON r.submission_id = s.id
                 LEFT JOIN leaderboard.user_info u ON s.user_id = u.id

ranking_worker.py

@@ -117,9 +117,9 @@ def ensure_snapshot_table(conn):
         s.user_id,
         u.user_name,
         r.score,
-        RANK() OVER (
+        ROW_NUMBER() OVER (
             PARTITION BY s.leaderboard_id, r.runner, s.user_id
-            ORDER BY r.score ASC
+            ORDER BY r.score ASC, s.submission_time ASC, s.id ASC, r.id ASC
         ) AS personal_submission_rank
     FROM leaderboard.runs r
     JOIN leaderboard.submission s ON r.submission_id = s.id

tests/api/test_leaderboard_api.py

@@ -430,7 +430,7 @@ def test_hacked_submissions_are_hidden_from_public_leaderboard(client, app):
     assert "visible_clean_status.py" in ranked_files
 
 
-def test_equal_best_runs_preserve_ties_and_distinct_submission_count(client, app):
+def test_equal_best_runs_keep_one_entry_and_distinct_submission_count(client, app):
     before = client.get("/api/leaderboard/339").get_json()["data"]["rankings"]
     gpu, entries = next(iter(before.items()))
     winner = entries[0]
@@ -455,6 +455,53 @@ def test_equal_best_runs_preserve_ties_and_distinct_submission_count(client, app
         conn.commit()
     after = client.get("/api/leaderboard/339").get_json()["data"]["rankings"][gpu]
     tied = [item for item in after if item["submission_id"] == winner["submission_id"]]
-    assert len(tied) == 2
+    assert len(tied) == 1
     assert all(item["score"] == winner["score"] for item in tied)
     assert all(item["submission_count"] == winner["submission_count"] for item in tied)
+
+
+def test_tied_submissions_keep_earliest_per_account_and_preserve_other_users(client, app):
+    before = client.get("/api/leaderboard/339").get_json()["data"]["rankings"]
+    gpu, entries = next(iter(before.items()))
+    winner = entries[0]
+    with app.app_context():
+        conn = get_db_connection()
+        with conn.cursor() as cur:
+            cur.execute(
+                "INSERT INTO leaderboard.user_info (id, user_name) VALUES (%s, %s)",
+                ("tie-regression-user", winner["user_name"]),
+            )
+            for submission_id, new_user in [(990101, False), (990102, False), (990103, True)]:
+                cur.execute(
+                    """
+                    INSERT INTO leaderboard.submission (
+                        id, leaderboard_id, file_name, user_id, code_id, submission_time, done, status
+                    )
+                    SELECT %s, leaderboard_id, 'tie_regression.py',
+                           CASE WHEN %s THEN 'tie-regression-user' ELSE user_id END,
+                           code_id, submission_time - INTERVAL '1 day', done, status
+                    FROM leaderboard.submission WHERE id = %s
+                    """,
+                    (submission_id, new_user, winner["submission_id"]),
+                )
+                cur.execute(
+                    """
+                    INSERT INTO leaderboard.runs (
+                        id, submission_id, start_time, end_time, mode, secret,
+                        runner, score, passed, compilation, meta, result, system_info
+                    )
+                    SELECT %s + ROW_NUMBER() OVER (), %s, start_time, end_time, mode, secret,
+                           runner, score, passed, compilation, meta, result, system_info
+                    FROM leaderboard.runs
+                    WHERE submission_id = %s AND runner = %s
+                    """,
+                    (submission_id * 100, submission_id, winner["submission_id"], gpu),
+                )
+        conn.commit()
+    after = client.get("/api/leaderboard/339").get_json()["data"]["rankings"][gpu]
+    matches = [row for row in after if row["user_name"] == winner["user_name"]]
+    assert {row["submission_id"] for row in matches} == {990101, 990103}
+    assert len(matches) == 2
+    assert all(row["score"] == winner["score"] for row in matches)
+    original = next(row for row in matches if row["submission_id"] == 990101)
+    assert original["submission_count"] == winner["submission_count"] + 2