[AMD][CI] Name the ROCm Image That Actually Ran in AMD Job Names (#35686)

Co-authored-by: Cursor <cursoragent@cursor.com>

scripts/ci_monitor/ci_failures_analysis.py

@@ -27,6 +27,51 @@ from typing import Dict, List, Optional, Tuple
 import requests
 
 
+def _filter_legacy_amd_job_rows(job_data: Dict[str, Dict]) -> Dict[str, Dict]:
+    """Drop pre-cutover AMD names without changing the shared analyzer."""
+
+    filtered = {}
+    for full_name, data in job_data.items():
+        # This caller was renamed by the AMD job-name cutover. Other outer
+        # callers, including AITER's *-rocm720 callers, are still current.
+        name_parts = full_name.split(" / ")
+        if "call-pr-test-amd-extra-rocm720" in name_parts[:-1]:
+            continue
+
+        leaf_name = name_parts[-1]
+        if leaf_name.startswith(("wait-for-stage-a-amd", "wait-for-stage-b-amd")):
+            continue
+        if leaf_name in {
+            "call-gate",
+            "call-pr-test-amd-extra",
+            "check-all-jobs",
+            "check-changes",
+            "pr-gate",
+            "pr-test-amd-extra-finish",
+            "pr-test-amd-finish",
+            "pr-test-amd-rocm720-finish",
+        }:
+            continue
+
+        # Old display-name stems ended in -rocm<digits>. Their version was not
+        # reliable enough to map forward: a -rocm720 PR job could run rocm724.
+        stem = leaf_name.split(" (", 1)[0]
+        _, separator, version = stem.rpartition("-rocm")
+        if separator and version.isdigit():
+            continue
+
+        # An intermediate nightly schema showed only the ROCm flavor. Current
+        # names always include both the flavor and runner inside parentheses.
+        if leaf_name.endswith(")") and " (" in leaf_name:
+            details = leaf_name.rsplit(" (", 1)[1][:-1]
+            if details.startswith("rocm") and details[4:].isdigit():
+                continue
+
+        filtered[full_name] = data
+
+    return filtered
+
+
 class SGLangFailuresAnalyzer:
     """Analyzes consecutive failures in GitHub Actions workflows."""
 
@@ -2648,6 +2693,21 @@ def main():
             else ({}, {})
         )
 
+        # AMD renamed its display names to carry the actual ROCm flavor. Reset
+        # only AMD history at that boundary so legacy failures cannot remain
+        # "current" under names that no longer exist. Other platforms continue
+        # to use the unmodified shared analyzer results. General AMD reports
+        # intentionally omit old-schema rows from branches that have not moved
+        # to the new names; their ROCm flavor cannot be mapped reliably.
+        pr_test_amd_scheduled_data = _filter_legacy_amd_job_rows(
+            pr_test_amd_scheduled_data
+        )
+        nightly_amd_scheduled_data = _filter_legacy_amd_job_rows(
+            nightly_amd_scheduled_data
+        )
+        pr_test_amd_general_data = _filter_legacy_amd_job_rows(pr_test_amd_general_data)
+        nightly_amd_general_data = _filter_legacy_amd_job_rows(nightly_amd_general_data)
+
         # Analyze runner health and consecutive failures on all runs
         (
             runner_stats,

scripts/ci_monitor/test_ci_failures_analysis.py

@@ -0,0 +1,57 @@
+"""AMD job-name cutover tests.
+
+Run with:
+    python -m unittest discover -s scripts/ci_monitor -p 'test_ci_failures_analysis.py'
+"""
+
+import os
+import sys
+import unittest
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+from ci_failures_analysis import _filter_legacy_amd_job_rows  # noqa: E402
+
+
+class TestFilterLegacyAmdJobRows(unittest.TestCase):
+    def test_drops_legacy_names_and_nested_utilities(self):
+        rows = {
+            "stage-b-test-1-gpu-small-amd-rocm720 (linux-mi300-1gpu-sglang, 0)": {},
+            "nightly-accuracy-2-gpu-rocm720 (rocm724)": {},
+            "nightly-accuracy-2-gpu-rocm724": {},
+            "nightly-test-1-gpu-unit (rocm724)": {},
+            "call-pr-test-amd-rocm720 / call-pr-test-amd-extra-rocm720 / extra-a-test-1-gpu-small-amd (linux-mi300-1gpu-sglang)": {},
+            "wait-for-stage-a-amd": {},
+            "call-pr-test-amd-extra / pr-test-amd-extra-finish": {},
+            "call-pr-test-amd-extra / call-gate / pr-gate": {},
+        }
+
+        self.assertEqual(_filter_legacy_amd_job_rows(rows), {})
+
+    def test_keeps_current_flavors_and_nested_callers_separate(self):
+        new_success = {"current_streak": 0}
+        rows = {
+            "stage-b-test-1-gpu-small-amd (rocm724, linux-mi300-1gpu-sglang, 0)": new_success,
+            "nightly-accuracy-2-gpu (rocm720, linux-mi300-2gpu-sglang)": {
+                "current_streak": 1
+            },
+            "call-pr-test-amd-rocm720 / stage-c-test-4-gpu-amd (rocm724, linux-mi300-4gpu-sglang, 0)": {
+                "current_streak": 0
+            },
+            "call-pr-test-amd-rocm720 / call-pr-test-amd-extra / extra-a-test-1-gpu-small-amd (rocm724, linux-mi300-1gpu-sglang)": {
+                "current_streak": 0
+            },
+        }
+
+        filtered = _filter_legacy_amd_job_rows(rows)
+
+        self.assertEqual(set(filtered), set(rows))
+        self.assertIs(
+            filtered[
+                "stage-b-test-1-gpu-small-amd (rocm724, linux-mi300-1gpu-sglang, 0)"
+            ],
+            new_success,
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()