[autorevert] correctly fetch ghstack commits (#7291)

izaitsevfb · web-flow · commit 433cd568cfd4 · 2025-10-21T14:24:51.000-07:00
Ghstack commits are nested inside a single push event. Fortunately, the webhook has this information, we just need to unroll the list. Testing: [2025-10-03T19-16-15.103579-00-00.html](https://github.com/user-attachments/files/22686819/2025-10-03T19-16-15.103579-00-00.html) see this stack there: ``` Tue 10:43 am 5ed4672 [dynamo, 3.14] fix _detect_and_normalize_assert_statement for 3.14 (#164005) #164005 williamwen42 Tue 10:43 am 2600f8b [dynamo, 3.14] fix tracing typing.Union (#164004) #164004 williamwen42 Tue 10:42 am 9ce31e4 [3.14] make unbacked_sym[int/float]_counter integers (#163920) #163920 williamwen42 Tue 10:42 am 0657de9 [dynamo, 3.14] support LOAD_COMMON_CONSTANT (#163919) #163919 williamwen42 Tue 10:42 am 4ead8eb [dynamo, 3.14] fix BUILD_TUPLE with 0 args (#163818) #163818 williamwen42 Tue 10:42 am d4b785a [dynamo, 3.14] fix stack ref copy error (#163796) #163796 williamwen42 Tue 10:42 am 9278b18 [dynamo, 3.14] fix WITH_EXCEPT_START (#163292) #163292 williamwen42 Tue 10:42 am 008b0a9 [dynamo, 3.14] fix inactive ctx handling in resume functions (#163191) #163191 williamwen42 Tue 10:42 am 44677ad [dynamo, 3.14] support LOAD_CONST on slice, codegen LOAD_CONST slice instead of BINARY/STORE_SLICE (#163110) #163110 williamwen42 Tue 10:42 am 1c9987f [dynamo, 3.14] fix context managers (#163109) #163109 williamwen42 Tue 10:41 am 7cbc011 [dynamo, 3.14] support some bytecodes, fix CALL_FUNCTION_EX (#163009) #163009 williamwen42 Tue 10:41 am 09c7741 [dynamo, 3.14] Python dynamo changes to get basic programs working (#161839) #161839 williamwen42 Tue 10:41 am 763ab2a [dynamo, 3.14] compile actual code in C dynamo (#161555) #161555 williamwen42 Tue 10:41 am 4b8fe79 [dynamo] format cpython_defs.c (#161838) #161838 williamwen42 ```
diff --git a/aws/lambda/pytorch-auto-revert/pytorch_auto_revert/signal_extraction_datasource.py b/aws/lambda/pytorch-auto-revert/pytorch_auto_revert/signal_extraction_datasource.py
@@ -30,17 +30,26 @@ def fetch_commits_in_time_range(
         """
         Fetch all commits pushed to main within the lookback window.
         Returns list of (sha, timestamp) tuples ordered newest → older.
+
+        Note: Uses ARRAY JOIN to extract all commits from ghstack pushes,
+        which can contain multiple commits in a single push event.
+        For commits with identical timestamps (from the same ghstack push),
+        orders by array index descending to match HUD/torchci ordering.
         """
         lookback_time = datetime.now() - timedelta(hours=lookback_hours)
 
         query = """
-        SELECT head_commit.id AS sha, max(head_commit.timestamp) AS ts
+        SELECT
+            commit.id AS sha,
+            max(commit.timestamp) AS ts,
+            arrayMax(groupArray(arrayFirstIndex(c -> c.'id' = commit.id, commits))) as array_index
         FROM default.push
+        ARRAY JOIN commits as commit, commits as c
         WHERE head_commit.timestamp >= {lookback_time:DateTime}
           AND ref = 'refs/heads/main'
           AND dynamoKey like {repo:String}
         GROUP BY sha
-        ORDER BY ts DESC
+        ORDER BY ts DESC, array_index DESC
         """
 
         params = {
