Vllm build metrics and tabs

torchci/clickhouse_queries/vllm/docker_build_runtime/params.json

@@ -0,0 +1,14 @@
+{
+  "params": {
+    "repo": "String",
+    "startTime": "DateTime64(3)",
+    "stopTime": "DateTime64(3)"
+  },
+  "tests": [
+    {
+      "repo": "https://github.com/vllm-project/vllm.git",
+      "startTime": "2025-10-01T00:00:00.000",
+      "stopTime": "2025-11-01T00:00:00.000"
+    }
+  ]
+}

torchci/clickhouse_queries/vllm/docker_build_runtime/query.sql

@@ -0,0 +1,32 @@
+-- vLLM Docker Build Image Runtime Trends (main branch only)
+-- Tracks runtime for the ":docker: build image" job specifically
+-- This is a critical job for build speed monitoring
+
+WITH jobs AS (
+    SELECT
+        tupleElement(job, 'name') AS job_name,
+        tupleElement(job, 'started_at') AS job_started_at,
+        tupleElement(job, 'finished_at') AS job_finished_at,
+        tupleElement(job, 'state') AS job_state,
+        tupleElement(build, 'number') AS build_number
+    FROM vllm.vllm_buildkite_jobs
+    WHERE
+        tupleElement(pipeline, 'repository') = {repo: String }
+        AND tupleElement(build, 'branch') = 'main'
+        AND tupleElement(job, 'name') = ':docker: build image'
+        AND tupleElement(job, 'started_at') IS NOT NULL
+        AND tupleElement(job, 'finished_at') IS NOT NULL
+        AND tupleElement(job, 'started_at') >= {startTime: DateTime64(3) }
+        AND tupleElement(job, 'started_at') < {stopTime: DateTime64(3) }
+        AND lowerUTF8(tupleElement(job, 'state')) IN (
+            'passed', 'finished', 'success', 'failed'
+        )
+)
+
+SELECT
+    job_started_at AS timestamp,
+    build_number,
+    round(dateDiff('second', job_started_at, job_finished_at) / 60.0, 2)
+        AS runtime_minutes
+FROM jobs
+ORDER BY job_started_at ASC

torchci/clickhouse_queries/vllm/job_runtime_trends/params.json

@@ -0,0 +1,16 @@
+{
+  "params": {
+    "repo": "String",
+    "startTime": "DateTime64(3)",
+    "stopTime": "DateTime64(3)",
+    "jobGroups": "Array(String)"
+  },
+  "tests": [
+    {
+      "repo": "https://github.com/vllm-project/vllm.git",
+      "startTime": "2025-10-01T00:00:00.000",
+      "stopTime": "2025-10-08T00:00:00.000",
+      "jobGroups": ["main", "amd", "torch_nightly"]
+    }
+  ]
+}

torchci/clickhouse_queries/vllm/job_runtime_trends/query.sql

@@ -0,0 +1,66 @@
+-- vLLM Job Runtime Trends (main branch only)
+-- Aggregates per-job runtime statistics by day
+-- Shows count, mean, p90, and max runtime for each job per day
+-- Supports filtering by job groups: AMD, Torch Nightly, or Main
+
+WITH jobs AS (
+    SELECT
+        tupleElement(job, 'name') AS job_name,
+        tupleElement(job, 'started_at') AS job_started_at,
+        tupleElement(job, 'finished_at') AS job_finished_at,
+        tupleElement(job, 'state') AS job_state,
+        tupleElement(build, 'branch') AS branch
+    FROM vllm.vllm_buildkite_jobs
+    WHERE
+        tupleElement(pipeline, 'repository') = {repo: String }
+        AND tupleElement(build, 'branch') = 'main'
+        AND tupleElement(job, 'started_at') IS NOT NULL
+        AND tupleElement(job, 'finished_at') IS NOT NULL
+        AND tupleElement(job, 'started_at') >= {startTime: DateTime64(3) }
+        AND tupleElement(job, 'started_at') < {stopTime: DateTime64(3) }
+        AND lowerUTF8(tupleElement(job, 'state')) IN (
+            'passed', 'finished', 'success', 'failed'
+        )
+        -- Job group filtering: AMD, Torch Nightly, or Main
+        AND (
+            (
+                has({jobGroups: Array(String)}, 'amd')
+                AND positionCaseInsensitive(tupleElement(job, 'name'), 'AMD')
+                > 0
+            )
+            OR (
+                has({jobGroups: Array(String)}, 'torch_nightly')
+                AND positionCaseInsensitive(
+                    tupleElement(job, 'name'), 'Torch Nightly'
+                )
+                > 0
+            )
+            OR (
+                has({jobGroups: Array(String)}, 'main')
+                AND positionCaseInsensitive(tupleElement(job, 'name'), 'AMD')
+                = 0
+                AND positionCaseInsensitive(
+                    tupleElement(job, 'name'), 'Torch Nightly'
+                )
+                = 0
+            )
+        )
+)
+
+SELECT
+    job_name,
+    toDate(job_started_at) AS date,
+    count() AS count,
+    round(avg(dateDiff('second', job_started_at, job_finished_at) / 60.0), 2)
+        AS mean_runtime_minutes,
+    round(
+        quantile(0.9) (
+            dateDiff('second', job_started_at, job_finished_at) / 60.0
+        ),
+        2
+    ) AS p90_runtime_minutes,
+    round(max(dateDiff('second', job_started_at, job_finished_at) / 60.0), 2)
+        AS max_runtime_minutes
+FROM jobs
+GROUP BY job_name, date
+ORDER BY job_name ASC, date ASC

torchci/components/metrics/vllm/CiDurationsPanel.tsx

@@ -230,28 +230,11 @@ export default function CiDurationsPanel({
       ...getLineSeries(dailyMeanSuccess, dailyMeanNonCanceled),
       ...getScatterSeriesByState(source),
     ],
-    dataZoom: [
-      {
-        type: "slider",
-        show: true,
-        xAxisIndex: 0,
-        bottom: 0,
-        start: 0,
-        end: 100,
-        height: 25,
-      },
-      {
-        type: "inside",
-        xAxisIndex: 0,
-        start: 0,
-        end: 100,
-      },
-    ],
   };
 
   return (
     <ChartPaper
-      tooltip="Main branch CI runtimes over time. Green line = mean runtime for successful builds, Pink line = mean including failures. Scatter points = individual builds (click to view in Buildkite). Use slider or scroll to zoom."
+      tooltip="Main branch CI runtimes over time. Green line = mean runtime for successful builds, Pink line = mean including failures. Scatter points = individual builds (click to view in Buildkite)."
       option={options}
       onEvents={{
         click: handleBuildClick,

torchci/components/metrics/vllm/DockerBuildRuntimePanel.tsx

@@ -0,0 +1,164 @@
+import dayjs from "dayjs";
+import { EChartsOption } from "echarts";
+import { useDarkMode } from "lib/DarkModeContext";
+import _ from "lodash";
+import { ChartPaper } from "./chartUtils";
+import { COLOR_SUCCESS, COLOR_WARNING } from "./constants";
+
+interface DockerBuildData {
+  timestamp: string;
+  build_number: number;
+  runtime_minutes: number;
+}
+
+// Helper function to format tooltip
+function formatTooltip(params: any): string {
+  if (!params || !params.data) return "";
+
+  const data = params.data;
+
+  // Handle both scatter (array) and line (object) series
+  let timestamp, runtime, buildNumber;
+
+  if (Array.isArray(data)) {
+    timestamp = data[0];
+    runtime = data[1];
+    buildNumber = data[2];
+  } else {
+    // For line series (daily average)
+    timestamp = data.day;
+    runtime = data.value;
+    buildNumber = null;
+  }
+
+  if (!timestamp || runtime === undefined) return "";
+
+  const formattedTime = dayjs(timestamp).format("M/D/YY h:mm A");
+
+  let result = buildNumber
+    ? `<b>Build #${buildNumber}</b><br/>`
+    : `<b>Daily Average</b><br/>`;
+  result += `Time: ${formattedTime}<br/>`;
+  result += `Runtime: <b>${runtime.toFixed(1)} min</b>`;
+
+  return result;
+}
+
+// Helper function to handle click events
+function handleBuildClick(params: any) {
+  if (params?.componentType === "series") {
+    const data = Array.isArray(params.data) ? params.data : [params.data];
+    const buildNumber = data[2];
+    if (buildNumber !== undefined && buildNumber !== null) {
+      const url = `https://buildkite.com/vllm/ci/builds/${buildNumber}/`;
+      if (typeof window !== "undefined") {
+        window.open(url, "_blank");
+      }
+    }
+  }
+}
+
+export default function DockerBuildRuntimePanel({
+  data,
+}: {
+  data: DockerBuildData[] | undefined;
+}) {
+  const { darkMode } = useDarkMode();
+
+  // Process data for chart
+  const chartData = (data || []).map((d) => [
+    dayjs(d.timestamp).toISOString(),
+    d.runtime_minutes,
+    d.build_number,
+  ]);
+
+  // Calculate daily average for trend line
+  const groupedByDay = _.groupBy(data || [], (d) =>
+    dayjs(d.timestamp).format("YYYY-MM-DD")
+  );
+
+  const dailyAvg = Object.entries(groupedByDay)
+    .map(([day, records]) => {
+      const avgRuntime = _.meanBy(records, "runtime_minutes");
+      return {
+        day,
+        value: Number(avgRuntime.toFixed(1)),
+      };
+    })
+    .sort((a, b) => (a.day < b.day ? -1 : 1));
+
+  // Calculate statistics
+  const runtimes = (data || []).map((d) => d.runtime_minutes);
+  const avgRuntime = runtimes.length ? _.mean(runtimes).toFixed(1) : "N/A";
+  const p90Runtime = runtimes.length
+    ? runtimes
+        .sort((a, b) => a - b)
+        [Math.floor(runtimes.length * 0.9)].toFixed(1)
+    : "N/A";
+
+  const options: EChartsOption = {
+    title: {
+      text: "Docker Build Image Runtime",
+      subtext: `Avg: ${avgRuntime}m  |  P90: ${p90Runtime}m  |  Total builds: ${runtimes.length}`,
+      textStyle: {
+        fontSize: 14,
+      },
+    },
+    legend: {
+      top: 24,
+      data: ["Individual Builds", "Daily Average"],
+    },
+    grid: { top: 60, right: 20, bottom: 80, left: 60 },
+    dataset: [{ source: chartData }, { source: dailyAvg }],
+    xAxis: {
+      type: "time",
+      axisLabel: {
+        hideOverlap: true,
+        formatter: (value: number) => dayjs(value).format("M/D"),
+      },
+    },
+    yAxis: {
+      type: "value",
+      name: "Runtime (minutes)",
+      nameLocation: "middle",
+      nameGap: 45,
+      nameRotate: 90,
+      axisLabel: {
+        formatter: (value: number) => `${value}m`,
+      },
+    },
+    series: [
+      {
+        name: "Individual Builds",
+        type: "scatter",
+        datasetIndex: 0,
+        symbolSize: 6,
+        itemStyle: { color: COLOR_SUCCESS, opacity: 0.6 },
+      },
+      {
+        name: "Daily Average",
+        type: "line",
+        datasetIndex: 1,
+        smooth: true,
+        encode: { x: "day", y: "value" },
+        lineStyle: { color: COLOR_WARNING, width: 2 },
+        itemStyle: { color: COLOR_WARNING },
+        showSymbol: true,
+        symbolSize: 4,
+      },
+    ],
+    tooltip: {
+      trigger: "item",
+      formatter: formatTooltip,
+    },
+  };
+
+  return (
+    <ChartPaper
+      tooltip="Docker build image runtime over time. Each point is an individual build (click to open in Buildkite). Green line shows daily average trend."
+      option={options}
+      onEvents={{ click: handleBuildClick }}
+      darkMode={darkMode}
+    />
+  );
+}

torchci/components/metrics/vllm/DurationDistributionPanel.tsx

@@ -164,28 +164,11 @@ export default function DurationDistributionPanel({
       axisPointer: { type: "shadow" },
       formatter: formatDistributionTooltip,
     },
-    dataZoom: [
-      {
-        type: "slider",
-        show: true,
-        xAxisIndex: 0,
-        bottom: 0,
-        start: 0,
-        end: 100,
-        height: 25,
-      },
-      {
-        type: "inside",
-        xAxisIndex: 0,
-        start: 0,
-        end: 100,
-      },
-    ],
   };
 
   return (
     <ChartPaper
-      tooltip="Histogram showing distribution of main branch CI runtimes (how long builds take to complete). Green = successful builds, Red = failed builds, Gray = canceled builds. Use slider or scroll to zoom."
+      tooltip="Histogram showing distribution of main branch CI runtimes (how long builds take to complete). Green = successful builds, Red = failed builds, Gray = canceled builds."
       option={options}
       darkMode={darkMode}
     />