[ch] optimize master_commit_red_percent query

[izaitsevfb]

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.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Updated (UTC)
torchci	✅ Ready (Inspect)	Visit Preview	Apr 25, 2025 11:39pm