[ch] optimize master_commit_red_percent query (#6580)

perf:
```
./clickhouse_query_perf.py --query master_commit_red_percent  --perf --times 1 --base HEAD

+------+----------+-----------+-------------+---------------+-----------+------------+-------------+--------------+
| Test | Avg Time | Base Time | Time Change | % Time Change |  Avg Mem  |  Base Mem  |  Mem Change | % Mem Change |
+------+----------+-----------+-------------+---------------+-----------+------------+-------------+--------------+
|  0   |   1164   |    5585   |    -4421    |      -79      | 160555724 | 7544036697 | -7383480973 |     -98      |
|  1   |   2992   |    5483   |    -2491    |      -45      | 516092615 | 7858494950 | -7342402335 |     -93      |
+------+----------+-----------+-------------+---------------+-----------+------------+-------------+--------------+
```

Removes expensive join, improves the pre-filtering utilizing a
materialized view.



### Testing

The new query in some cases returns slightly different results for
`broken_trunk_red` metrics (although results for `all_red` metric are
identical), which leads me to conclude that partitioning:

```
 ROW_NUMBER() OVER(
            PARTITION BY j.name,
            j.head_sha
            ORDER BY
                j.run_attempt DESC
        ) AS row_num
```
is not stable (missing workflow_name? each row can have multiple
pushes?)

but overall the results are close enough (ε < 0.01) and further
debugging and fixing could be made in a separate PR.

Author: izaitsevfb

torchci/clickhouse_queries/master_commit_red_percent/params.json

@@ -5,5 +5,24 @@
     "stopTime": "DateTime64(3)",
     "workflowNames": "Array(String)"
   },
-  "tests": []
+  "tests": [
+    {
+      "granularity": "day",
+      "startTime": "2025-03-18T21:09:47.987",
+      "stopTime": "2025-03-25T21:09:47.987",
+      "workflowNames": ["lint", "pull", "trunk"]
+    },
+    {
+      "granularity": "day",
+      "startTime": "2025-03-18T21:09:47.987",
+      "stopTime": "2025-05-25T21:09:47.987",
+      "workflowNames": ["lint", "pull", "trunk"]
+    },
+    {
+      "granularity": "day",
+      "startTime": "2025-03-24T00:00:00.000",
+      "stopTime": "2025-03-25T00:00:00.000",
+      "workflowNames": ["pull"]
+    }
+  ]
 }

torchci/clickhouse_queries/master_commit_red_percent/query.sql

@@ -1,45 +1,43 @@
--- TODO (huydhn): This query tracks the number of red commits on HUD KPIs page. This
--- is not the most efficient query around both in term of speed and memory usage. So,
--- a good BE task is to re-write this in a more efficient way, HUD code is also
--- subjected to change if need be
-WITH join_with_workflow_run AS (
-    -- Do the join with workflow_run, then workflow_job to avoid OOM
-    SELECT
-        w.id AS id,
+-- huydhn: This query tracks the number of red commits on HUD KPIs page.
+WITH pushes AS ( -- very selective
+    select
         p.head_commit. 'timestamp' AS time,
         p.head_commit. 'id' AS sha
-    FROM
+    from
         default .push p FINAL
-        JOIN default .workflow_run w FINAL ON w.head_commit. 'id' = p.head_commit. 'id'
-    WHERE
-        (
-            -- Limit it to workflows which block viable/strict upgrades
-            has({workflowNames: Array(String) }, lower(w.name))
-            OR w.name like 'linux-binary%'
-        )
-        AND w.event != 'workflow_run' -- Filter out worflow_run-triggered jobs, which have nothing to do with the SHA
-        AND p.ref = 'refs/heads/main'
-        AND p.repository. 'owner'.'name' = 'pytorch'
-        AND p.repository. 'name' = 'pytorch'
-        AND p.head_commit. 'timestamp' >= {startTime: DateTime64(3) }
-        AND p.head_commit. 'timestamp' < {stopTime: DateTime64(3) }
+    where
+        p.ref = 'refs/heads/main'
+        and p.repository. 'owner'.'name' = 'pytorch'
+        and p.repository. 'name' = 'pytorch'
+        and p.head_commit. 'timestamp' >= {startTime: DateTime64(3) }
+        and p.head_commit. 'timestamp' < {stopTime: DateTime64(3) }
 ),
 all_jobs AS (
     SELECT
-        w.time AS time,
+        p.time AS time,
         j.conclusion AS conclusion,
-        w.sha AS sha,
+        j.head_sha AS sha,
         ROW_NUMBER() OVER(
             PARTITION BY j.name,
-            w.sha
+            j.head_sha
             ORDER BY
                 j.run_attempt DESC
         ) AS row_num
     FROM
-        join_with_workflow_run w
-        JOIN default .workflow_job j FINAL ON w.id = j.run_id
+        default .workflow_job j FINAL
+        join pushes p FINAL on j.head_sha = p.sha
     WHERE
-        j.name != 'ciflow_should_run'
+        j.id in (
+            SELECT id FROM materialized_views.workflow_job_by_head_sha
+            WHERE head_sha in (SELECT distinct p.sha FROM pushes p)
+        )
+        AND j.workflow_event != 'workflow_run' -- Filter out worflow_run-triggered jobs, which have nothing to do with the SHA
+        AND (
+            -- Limit it to jobs which block viable/strict upgrades
+            has({workflowNames: Array(String) }, lower(j.workflow_name))
+            OR j.workflow_name like 'linux-binary%'
+        )
+        AND j.name != 'ciflow_should_run'
         AND j.name != 'generate-test-matrix'
         AND j.name NOT LIKE '%rerun_disabled_tests%'
         AND j.name NOT LIKE '%unstable%'