Performance Regression Report — i8g Tablets (Official Releases)

GENERATE-PERF-REPORT.md

Prompt: Generate Performance Regression Report

Copy-paste the following prompt to an AI assistant with access to the Argus CLI to generate a fresh report.

---

Markdown Report Prompt

Using the argus CLI at `cli/bin/argus`, generate a performance regression report in markdown.

**Parameters** (adjust as needed):
- Run ID (any recent run from the test): 24b36865-65ba-4fdc-8697-879029a49428
- Argus URL: https://argus.scylladb.com
- Time window: last 3 months (use `--after $(date -d '90 days ago' +%s)`)
- Output file: perf-report-i8g-tablets-releases.md
- Filter: Official releases only (version matches `^\d{4}\.\d+\.\d+$`, no ~dev/~rc)

**Data collection:**
1. Resolve test-id: `./bin/argus run get --run-id <RUN_ID> --url https://argus.scylladb.com`
2. List runs: `./bin/argus run list --test-id <TEST_ID> --after <TIMESTAMP> --limit 200 --full --url https://argus.scylladb.com`
3. Filter to official releases only (scylla_version matches `^\d{4}\.\d+\.\d+$`)
4. Fetch results per run: `./bin/argus run results --run-id <RUN_ID> --url https://argus.scylladb.com`
5. Group by workload (read_disk_only, read, write, mixed) and step. Average across runs of same version.

**Report structure:**

  # Performance Regression Report — i8g Tablets
  (test name, period, run count)

  | Version | Date | Passed | Failed | Test Error |
  |---------|------|--------|--------|------------|

  ## Table of Contents
  (links to each workload + Key Observations)

  ## <Workload Name>

  | Step | Samples | P90 | P99 | Throughput | Best Release |
  |------|---------|-----|-----|------------|--------------|

  ### P99 Latency by Step (ASCII bar chart)

  110K  ▓▓▓ 1.66  ░░░ 1.70  ███ 1.65  ▒▒▒ 1.77
  220K  ▓▓▓ 1.99  ░░░ 2.02  ███ 2.01  ▒▒▒▒ 2.80
  330K  ▓▓▓ 2.55  ░░░ 2.58  ███ 2.59  ▒▒▒▒▒▒▒▒▒ 14.79
  400K  ▓▓ 4.93   ░░░░░░░░░░░░░░░░░░░░ 292.29  ███████ 92.14  ▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒ 797.44

  ▓ 2026.1.1  ░ 2025.4.6  █ 2025.4.7  ▒ 2026.1.2

  ### Unthrottled Throughput

  2026.1.1  ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓ 546,380
  2025.4.6  ░░░░░░░░░░░░░░░░░░░  515,489
  2025.4.7  ████████████████████  543,609
  2026.1.2  ▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒  488,160

  (repeat per workload)

  ## Key Observations
  (numbered findings)

**Chart rules:**
- Use log-scale bar widths when max/min > 50× (mark with "log scale" note)
- Always show exact value at end of each bar
- Use consistent block chars per release: ▓ ░ █ ▒
- Wrap charts in ``` code blocks for monospace alignment

**DO NOT include:**
- Mermaid or image-based charts
- Separate per-step charts
- Run-over-run charts
- "Achieved Throughput by Step" aggregate charts

Save to `perf-report-i8g-tablets-releases.md`.

---

Quick one-liner (if CLI is already built)

# Adjust RUN_ID or TEST_ID and period as needed
TEST_ID="d6ebf1a5-135f-43fc-a7ba-0716b60dfa94"
AFTER=$(date -d '90 days ago' +%s)

# List runs with version info
./bin/argus run list --test-id $TEST_ID --after $AFTER --limit 200 --full --url https://argus.scylladb.com | \
  python3 -c "
import sys, json, re
for r in json.load(sys.stdin):
    v = r.get('scylla_version','')
    if re.match(r'^\d{4}\.\d+\.\d+$', v):
        print(f'{r[\"id\"]}  {v}  {r[\"status\"]}')
"

# Then feed run IDs into: ./bin/argus run results --run-id <ID> --url https://argus.scylladb.com

---

HTML Report Prompt (Chart.js)

Using the argus CLI at `cli/bin/argus`, generate a performance regression report as a self-contained HTML file with Chart.js graphs.

**Parameters** (adjust as needed):
- Run ID (any recent run from the test): 24b36865-65ba-4fdc-8697-879029a49428
- Argus URL: https://argus.scylladb.com
- Time window: last 3 months (use `--after $(date -d '90 days ago' +%s)`)
- Output file: perf-report-i8g-tablets-releases.html
- Filter: Official releases only (version matches pattern `^\d{4}\.\d+\.\d+$`, excludes ~dev and ~rc)

**Data collection steps** are identical to the Markdown prompt above (steps 1–5).

**HTML report structure:**
- Single self-contained `.html` file, no build step
- Load Chart.js and chartjs-plugin-datalabels from CDN:
  ```html
  <script src="https://cdn.jsdelivr.net/npm/chart.js@4"></script>
  <script src="https://cdn.jsdelivr.net/npm/chartjs-plugin-datalabels@2"></script>

• Dark theme (background: #1a1a2e, surface: #16213e, text: #e0e0e0)
• Header with test name, period, run counts
• Version summary table
• Table of Contents with anchor links to each workload section
• Per-workload section with:
  • HTML summary table with "Best Release" column (green highlight)
  • P99 latency grouped bar chart: x-axis = step values, one dataset per release version
  • Unthrottled throughput bar chart: x-axis = version labels, one dataset per release (use null for other positions so each version gets its own color)
  • Both chart types show data labels on top of each bar (formatted: ≥1M → "1.5M", ≥1K → "89.8K", else decimal)
• Key Observations section

Consistent per-release colors across ALL charts:

• 2026.1.1 → #4CAF50 (green)
• 2025.4.6 → #2196F3 (blue)
• 2025.4.7 → #FF9800 (orange)
• 2026.1.2 → #F44336 (red) Adjust version list as needed but keep one consistent color per release.

Auto-logarithmic scale: When the max/min ratio of values in a chart exceeds 50×, use logarithmic y-axis. Append " (log scale)" to the y-axis title when log is active. Otherwise use linear scale.

DO NOT include:

• Separate per-step charts (one chart per step) — combine all into one graph
• "Achieved Throughput by Step" aggregate charts
• Run-over-run charts
• "Unthrottled P99 by Release" charts

Save to perf-report-i8g-tablets-releases.html.

---

## For a different test/view

Replace the run-id with any run from the target test. The script will auto-resolve the test-id and discover all workload types from the result tables.

To find run IDs for other tests in the view `scylla-enterprise-performance-regression-tests-i8g-tablets`, inspect network requests on https://argus.scylladb.com/view/scylla-enterprise-performance-regression-tests-i8g-tablets — the API returns test UUIDs you can use with `--test-id`.

perf-report-i8g-tablets-releases.html

<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Performance Regression Report — i8g Tablets (Official Releases)</title>
<script src="https://cdn.jsdelivr.net/npm/chart.js@4"></script>
<script src="https://cdn.jsdelivr.net/npm/chartjs-plugin-datalabels@2"></script>
<style>
  :root { --bg: #1a1a2e; --surface: #16213e; --text: #e0e0e0; --muted: #999; }
  body { font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif; background: var(--bg); color: var(--text); margin: 0; padding: 2rem; line-height: 1.6; }
  h1 { color: #fff; margin-bottom: 0.25rem; }
  h2 { color: #90caf9; border-bottom: 1px solid #333; padding-bottom: 0.5rem; margin-top: 3rem; }
  h3 { color: #ce93d8; }
  .meta { color: var(--muted); margin-bottom: 2rem; }
  table { border-collapse: collapse; width: 100%; margin: 1rem 0 2rem; }
  th, td { border: 1px solid #333; padding: 0.5rem 0.75rem; text-align: right; }
  th { background: var(--surface); color: #90caf9; }
  td:first-child, th:first-child { text-align: left; }
  .best { color: #4caf50; font-weight: bold; }
  .chart-container { background: var(--surface); border-radius: 8px; padding: 1rem; margin: 1.5rem 0; max-width: 900px; }
  canvas { max-height: 400px; }
  #toc { background: var(--surface); border-radius: 8px; padding: 1rem 1.5rem; margin: 1.5rem 0; display: inline-block; }
  #toc a { color: #90caf9; text-decoration: none; }
  #toc a:hover { text-decoration: underline; }
  #toc ul { list-style: none; padding-left: 0; margin: 0.5rem 0; }
  #toc li { margin: 0.4rem 0; }
  .legend { display: flex; gap: 1.5rem; flex-wrap: wrap; margin: 0.5rem 0 1rem; }
  .legend-item { display: flex; align-items: center; gap: 0.4rem; font-size: 0.9rem; }
  .legend-swatch { width: 14px; height: 14px; border-radius: 3px; }
  .observations { background: var(--surface); border-radius: 8px; padding: 1.5rem; margin-top: 2rem; }
  .observations li { margin: 0.5rem 0; }
</style>
</head>
<body>

<h1>Performance Regression Report — i8g Tablets</h1>
<p class="meta">
  <strong>Test:</strong> scylla-enterprise-perf-regression-predefined-throughput-steps-i8g-tablets<br>
  <strong>Period:</strong> February 12 – May 12, 2026 (3 months)<br>
  <strong>Runs analyzed:</strong> 16 official release runs
</p>

<table>
  <tr><th>Version</th><th>Date</th><th>Passed</th><th>Failed</th><th>Test Error</th></tr>
  <tr><td>2026.1.1</td><td>2026-04-05</td><td>3</td><td>1</td><td>0</td></tr>
  <tr><td>2025.4.6</td><td>2026-04-10</td><td>2</td><td>2</td><td>0</td></tr>
  <tr><td>2025.4.7</td><td>2026-04-17</td><td>1</td><td>2</td><td>1</td></tr>
  <tr><td>2026.1.2</td><td>2026-05-04</td><td>0</td><td>4</td><td>0</td></tr>
</table>

<nav id="toc">
  <strong>Table of Contents</strong>
  <ul>
    <li><a href="#read-disk-only">Read (Disk-Only)</a></li>
    <li><a href="#read-cache">Read (Cache)</a></li>
    <li><a href="#write">Write</a></li>
    <li><a href="#mixed">Mixed 50/50 R/W</a></li>
    <li><a href="#observations">Key Observations</a></li>
  </ul>
</nav>

<!-- ============ Read (Disk-Only) ============ -->
<h2 id="read-disk-only">Read (Disk-Only)</h2>

<table>
  <tr><th>Step</th><th>Samples</th><th>P90 Read (ms)</th><th>P99 Read (ms)</th><th>Throughput (op/s)</th><th>Best Release</th></tr>
  <tr><td>110K</td><td>4</td><td>1.06</td><td>1.69</td><td>109,963</td><td class="best">2025.4.7 (1.65 ms)</td></tr>
  <tr><td>220K</td><td>4</td><td>1.30</td><td>2.21</td><td>219,928</td><td class="best">2026.1.1 (1.99 ms)</td></tr>
  <tr><td>330K</td><td>4</td><td>1.62</td><td>5.63</td><td>329,829</td><td class="best">2026.1.1 (2.55 ms)</td></tr>
  <tr><td>400K</td><td>4</td><td>2.14</td><td>296.70</td><td>399,854</td><td class="best">2026.1.1 (4.93 ms)</td></tr>
  <tr><td>unthrottled</td><td>4</td><td>3.80</td><td>49.33</td><td>523,409</td><td class="best">2026.1.1 (40.14 ms)</td></tr>
</table>

<h3>P99 Read Latency by Step</h3>
<div class="chart-container"><canvas id="chart-read-disk-p99"></canvas></div>

<h3>Unthrottled Throughput</h3>
<div class="chart-container"><canvas id="chart-read-disk-thr"></canvas></div>

<!-- ============ Read (Cache) ============ -->
<h2 id="read-cache">Read (Cache)</h2>

<table>
  <tr><th>Step</th><th>Samples</th><th>P90 Read (ms)</th><th>P99 Read (ms)</th><th>Throughput (op/s)</th><th>Best Release</th></tr>
  <tr><td>500K</td><td>4</td><td>0.40</td><td>0.70</td><td>499,794</td><td class="best">2026.1.1 (0.56 ms)</td></tr>
  <tr><td>900K</td><td>4</td><td>0.62</td><td>21.35</td><td>899,594</td><td class="best">2026.1.2 (2.58 ms)</td></tr>
  <tr><td>1200K</td><td>4</td><td>0.89</td><td>216.04</td><td>1,199,538</td><td class="best">2026.1.2 (4.29 ms)</td></tr>
  <tr><td>1500K</td><td>4</td><td>1.44</td><td>11.14</td><td>1,499,691</td><td class="best">2025.4.6 (6.50 ms)</td></tr>
  <tr><td>unthrottled</td><td>4</td><td>1.53</td><td>13.40</td><td>1,617,045</td><td class="best">2026.1.1 (2.68 ms)</td></tr>
</table>

<h3>P99 Read Latency by Step</h3>
<div class="chart-container"><canvas id="chart-read-cache-p99"></canvas></div>

<h3>Unthrottled Throughput</h3>
<div class="chart-container"><canvas id="chart-read-cache-thr"></canvas></div>

<!-- ============ Write ============ -->
<h2 id="write">Write</h2>

<table>
  <tr><th>Step</th><th>Samples</th><th>P90 Write (ms)</th><th>P99 Write (ms)</th><th>Throughput (op/s)</th><th>Best Release</th></tr>
  <tr><td>350K</td><td>4</td><td>1.63</td><td>2.27</td><td>349,761</td><td class="best">2025.4.6 (2.15 ms)</td></tr>
  <tr><td>600K</td><td>4</td><td>17097.55</td><td>22450.35</td><td>595,657</td><td class="best">2025.4.6 (3.02 ms)</td></tr>
  <tr><td>unthrottled</td><td>4</td><td>2.79</td><td>12.77</td><td>706,547</td><td class="best">2025.4.6 (6.34 ms)</td></tr>
</table>

<h3>P99 Write Latency by Step</h3>
<div class="chart-container"><canvas id="chart-write-p99"></canvas></div>

<h3>Unthrottled Throughput</h3>
<div class="chart-container"><canvas id="chart-write-thr"></canvas></div>

<!-- ============ Mixed ============ -->
<h2 id="mixed">Mixed 50/50 R/W</h2>

<table>
  <tr><th>Step</th><th>Samples</th><th>P90 W (ms)</th><th>P90 R (ms)</th><th>P99 W (ms)</th><th>P99 R (ms)</th><th>Thr W (op/s)</th><th>Thr R (op/s)</th><th>Best Release</th></tr>
  <tr><td>250K</td><td>3</td><td>0.40</td><td>1.00</td><td>2.13</td><td>2.38</td><td>124,914</td><td>124,903</td><td class="best">2025.4.6 (2.14 ms)</td></tr>
  <tr><td>480K</td><td>3</td><td>1.81</td><td>1.96</td><td>4.68</td><td>5.18</td><td>239,908</td><td>239,928</td><td class="best">2026.1.2 (3.67 ms)</td></tr>
  <tr><td>600K</td><td>3</td><td>1.99</td><td>2.26</td><td>6991.25</td><td>6977.41</td><td>299,832</td><td>299,846</td><td class="best">2026.1.2 (6.85 ms)</td></tr>
  <tr><td>750K</td><td>3</td><td>1426.16</td><td>1428.62</td><td>19374.89</td><td>19374.89</td><td>374,816</td><td>374,799</td><td class="best">2026.1.2 (11282.68 ms)</td></tr>
  <tr><td>unthrottled</td><td>3</td><td>3.68</td><td>27.60</td><td>10.35</td><td>186.60</td><td>470,417</td><td>470,372</td><td class="best">2026.1.2 (170.26 ms)</td></tr>
</table>

<h3>P99 Read Latency by Step</h3>
<div class="chart-container"><canvas id="chart-mixed-p99"></canvas></div>

<h3>Unthrottled Throughput</h3>
<div class="chart-container"><canvas id="chart-mixed-thr"></canvas></div>

<!-- ============ Key Observations ============ -->
<div class="observations" id="observations">
  <h2 style="margin-top:0; border:none;">Key Observations</h2>
  <ol>
    <li><strong>2026.1.2</strong> (latest): All 4 runs failed — limited data. Requires investigation.</li>
    <li><strong>2025.4.7</strong>: Only 1 of 4 runs passed. Limited data available.</li>
    <li><strong>2025.4.6</strong>: 2 of 4 passed. Stable baseline.</li>
    <li><strong>2026.1.1</strong>: 3 of 4 passed. Best coverage across workloads.</li>
    <li><strong>Pass rate</strong>: 6/16 (37.5%) — indicates infrastructure or test stability issues beyond code regressions.</li>
  </ol>
</div>

<p style="color: var(--muted); margin-top: 3rem; font-size: 0.85rem;">
  Generated from Argus (argus.scylladb.com) — Test ID: d6ebf1a5-135f-43fc-a7ba-0716b60dfa94
</p>

<script>
const COLORS = {
  '2026.1.1': { bg: 'rgba(76, 175, 80, 0.8)',  border: '#4CAF50' },
  '2025.4.6': { bg: 'rgba(33, 150, 243, 0.8)', border: '#2196F3' },
  '2025.4.7': { bg: 'rgba(255, 152, 0, 0.8)',  border: '#FF9800' },
  '2026.1.2': { bg: 'rgba(244, 67, 54, 0.8)',  border: '#F44336' },
};

const VERSIONS_4 = ['2026.1.1', '2025.4.6', '2025.4.7', '2026.1.2'];
const VERSIONS_3 = ['2026.1.1', '2025.4.6', '2026.1.2']; // mixed has no 2025.4.7

function makeDatasets(versions, dataPerVersion) {
  return versions.map(v => ({
    label: v,
    data: dataPerVersion[v],
    backgroundColor: COLORS[v].bg,
    borderColor: COLORS[v].border,
    borderWidth: 1,
  }));
}

Chart.register(ChartDataLabels);

function barChart(id, labels, versions, dataPerVersion, yLabel, useLog) {
  const allValues = Object.values(dataPerVersion).flat().filter(v => v != null && v > 0);
  const max = Math.max(...allValues);
  const min = Math.min(...allValues);
  const needsLog = useLog !== false && (max / min > 50);

  new Chart(document.getElementById(id), {
    type: 'bar',
    data: { labels, datasets: makeDatasets(versions, dataPerVersion) },
    options: {
      responsive: true,
      plugins: {
        legend: { labels: { color: '#e0e0e0' } },
        datalabels: {
          anchor: 'end',
          align: 'end',
          color: '#e0e0e0',
          font: { size: 10 },
          formatter: (v) => {
            if (v == null) return '';
            if (v >= 1000000) return (v / 1000000).toFixed(1) + 'M';
            if (v >= 1000) return (v / 1000).toFixed(1) + 'K';
            if (v >= 100) return v.toFixed(0);
            if (v >= 10) return v.toFixed(1);
            return v.toFixed(2);
          },
        },
      },
      scales: {
        x: { ticks: { color: '#ccc', maxRotation: 45 }, grid: { color: '#333' } },
        y: {
          type: needsLog ? 'logarithmic' : 'linear',
          ticks: { color: '#ccc' },
          grid: { color: '#333' },
          title: { display: true, text: yLabel + (needsLog ? ' (log scale)' : ''), color: '#ccc' },
        }
      }
    }
  });
}

// --- Read (Disk-Only) ---
barChart('chart-read-disk-p99', ['110K', '220K', '330K', '400K'], VERSIONS_4, {
  '2026.1.1': [1.66, 1.99, 2.55, 4.93],
  '2025.4.6': [1.70, 2.02, 2.58, 292.29],
  '2025.4.7': [1.65, 2.01, 2.59, 92.14],
  '2026.1.2': [1.77, 2.80, 14.79, 797.44],
}, 'P99 read (ms)');

barChart('chart-read-disk-thr', ['2026.1.1', '2025.4.6', '2025.4.7', '2026.1.2'], VERSIONS_4, {
  '2026.1.1': [546380, null, null, null],
  '2025.4.6': [null, 515489, null, null],
  '2025.4.7': [null, null, 543609, null],
  '2026.1.2': [null, null, null, 488160],
}, 'Throughput (op/s)');

// --- Read (Cache) ---
barChart('chart-read-cache-p99', ['500K', '900K', '1200K', '1500K'], VERSIONS_4, {
  '2026.1.1': [0.56, 8.27, 4.95, 7.40],
  '2025.4.6': [0.73, 3.44, 6.61, 6.50],
  '2025.4.7': [0.65, 71.11, 848.30, 8.29],
  '2026.1.2': [0.87, 2.58, 4.29, 22.36],
}, 'P99 read (ms)');

barChart('chart-read-cache-thr', ['2026.1.1', '2025.4.6', '2025.4.7', '2026.1.2'], VERSIONS_4, {
  '2026.1.1': [1768372, null, null, null],
  '2025.4.6': [null, 1628350, null, null],
  '2025.4.7': [null, null, 1561155, null],
  '2026.1.2': [null, null, null, 1510305],
}, 'Throughput (op/s)');

// --- Write ---
barChart('chart-write-p99', ['350K', '600K'], VERSIONS_4, {
  '2026.1.1': [2.23, 3.51],
  '2025.4.6': [2.15, 3.02],
  '2025.4.7': [2.21, 3.21],
  '2026.1.2': [2.51, 89791.66],
}, 'P99 write (ms)');

barChart('chart-write-thr', ['2026.1.1', '2025.4.6', '2025.4.7', '2026.1.2'], VERSIONS_4, {
  '2026.1.1': [702614, null, null, null],
  '2025.4.6': [null, 742218, null, null],
  '2025.4.7': [null, null, 720203, null],
  '2026.1.2': [null, null, null, 661154],
}, 'Throughput (op/s)');

// --- Mixed ---
barChart('chart-mixed-p99', ['250K', '480K', '600K', '750K'], VERSIONS_3, {
  '2026.1.1': [2.55, 5.67, 16064.18, 28739.37],
  '2025.4.6': [2.14, 6.20, 4861.20, 18102.62],
  '2026.1.2': [2.46, 3.67, 6.85, 11282.68],
}, 'P99 read (ms)');

barChart('chart-mixed-thr', ['2026.1.1', '2025.4.6', '2026.1.2'], VERSIONS_3, {
  '2026.1.1': [422946, null, null],
  '2025.4.6': [null, 512838, null],
  '2026.1.2': [null, null, 475469],
}, 'Throughput (op/s)');
</script>
</body>
</html>

perf-report-i8g-tablets-releases.md

Performance Regression Report — i8g Tablets (Official Releases Only)

Test: scylla-enterprise-perf-regression-predefined-throughput-steps-i8g-tablets Period: February 12 – May 12, 2026 (3 months) Runs analyzed: 16 official release runs

Version	Color	Date	Passed	Failed	Test Error
2026.1.1	🟩	2026-04-05	3	1	0
2025.4.6	🟦	2026-04-10	2	2	0
2025.4.7	🟧	2026-04-17	1	2	1
2026.1.2	🟥	2026-05-04	0	4	0

---

Table of Contents

• Read (Disk-Only)
• Read (Cache)
• Write
• Mixed 50/50 R/W
• Key Observations

---

Read (Disk-Only)

Step	Samples	P90 Read (ms)	P99 Read (ms)	Throughput (op/s)	Best Release
110K	4	1.06	1.69	109,963	🟧 2025.4.7 (1.65 ms)
220K	4	1.30	2.21	219,928	🟩 2026.1.1 (1.99 ms)
330K	4	1.62	5.63	329,829	🟩 2026.1.1 (2.55 ms)
400K	4	2.14	296.70	399,854	🟩 2026.1.1 (4.93 ms)
unthrottled	4	3.80	49.33	523,409	🟩 2026.1.1 (40.14 ms)

P99 Read Latency by Step (log scale)

Step	Version	P99 (ms)	Bar
110K	🟩	1.66	🟩
110K	🟦	1.70	🟦
110K	🟧	1.65	🟧
110K	🟥	1.77	🟥
220K	🟩	1.99	🟩
220K	🟦	2.02	🟦
220K	🟧	2.01	🟧
220K	🟥	2.80	🟥🟥
330K	🟩	2.55	🟩
330K	🟦	2.58	🟦
330K	🟧	2.59	🟧
330K	🟥	14.79	🟥🟥🟥🟥🟥🟥🟥
400K	🟩	4.93	🟩🟩🟩🟩
400K	🟦	292.29	🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦
400K	🟧	92.14	🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧
400K	🟥	797.44	🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥

Unthrottled Throughput

Version	Throughput (op/s)	Bar
🟩 2026.1.1	546,380	🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩
🟦 2025.4.6	515,489	🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦
🟧 2025.4.7	543,609	🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧
🟥 2026.1.2	488,160	🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥

---

Read (Cache)

Step	Samples	P90 Read (ms)	P99 Read (ms)	Throughput (op/s)	Best Release
500K	4	0.40	0.70	499,794	🟩 2026.1.1 (0.56 ms)
900K	4	0.62	21.35	899,594	🟥 2026.1.2 (2.58 ms)
1200K	4	0.89	216.04	1,199,538	🟥 2026.1.2 (4.29 ms)
1500K	4	1.44	11.14	1,499,691	🟦 2025.4.6 (6.50 ms)
unthrottled	4	1.53	13.40	1,617,045	🟩 2026.1.1 (2.68 ms)

P99 Read Latency by Step (log scale)

Step	Version	P99 (ms)	Bar
500K	🟩	0.56	🟩
500K	🟦	0.73	🟦
500K	🟧	0.65	🟧
500K	🟥	0.87	🟥
900K	🟩	8.27	🟩🟩🟩🟩🟩🟩🟩
900K	🟦	3.44	🟦🟦🟦🟦🟦
900K	🟧	71.11	🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧
900K	🟥	2.58	🟥🟥🟥🟥
1200K	🟩	4.95	🟩🟩🟩🟩🟩🟩
1200K	🟦	6.61	🟦🟦🟦🟦🟦🟦🟦
1200K	🟧	848.30	🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧
1200K	🟥	4.29	🟥🟥🟥🟥🟥🟥
1500K	🟩	7.40	🟩🟩🟩🟩🟩🟩🟩
1500K	🟦	6.50	🟦🟦🟦🟦🟦🟦🟦
1500K	🟧	8.29	🟧🟧🟧🟧🟧🟧🟧
1500K	🟥	22.36	🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥

Unthrottled Throughput

Version	Throughput (op/s)	Bar
🟩 2026.1.1	1,768,372	🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩
🟦 2025.4.6	1,628,350	🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦
🟧 2025.4.7	1,561,155	🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧
🟥 2026.1.2	1,510,305	🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥

---

Write

Step	Samples	P90 Write (ms)	P99 Write (ms)	Throughput (op/s)	Best Release
350K	4	1.63	2.27	349,761	🟦 2025.4.6 (2.15 ms)
600K	4	17097.55	22450.35	595,657	🟦 2025.4.6 (3.02 ms)
unthrottled	4	2.79	12.77	706,547	🟦 2025.4.6 (6.34 ms)

P99 Write Latency by Step (log scale)

Step	Version	P99 (ms)	Bar
350K	🟩	2.23	🟩
350K	🟦	2.15	🟦
350K	🟧	2.21	🟧
350K	🟥	2.51	🟥
600K	🟩	3.51	🟩
600K	🟦	3.02	🟦
600K	🟧	3.21	🟧
600K	🟥	89,791.66	🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥

Unthrottled Throughput

Version	Throughput (op/s)	Bar
🟩 2026.1.1	702,614	🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩
🟦 2025.4.6	742,218	🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦
🟧 2025.4.7	720,203	🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧🟧
🟥 2026.1.2	661,154	🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥

---

Mixed 50/50 R/W

Step	Samples	P90 W (ms)	P90 R (ms)	P99 W (ms)	P99 R (ms)	Thr W (op/s)	Thr R (op/s)	Best Release
250K	3	0.40	1.00	2.13	2.38	124,914	124,903	🟦 2025.4.6 (2.14 ms)
480K	3	1.81	1.96	4.68	5.18	239,908	239,928	🟥 2026.1.2 (3.67 ms)
600K	3	1.99	2.26	6991.25	6977.41	299,832	299,846	🟥 2026.1.2 (6.85 ms)
750K	3	1426.16	1428.62	19374.89	19374.89	374,816	374,799	🟥 2026.1.2 (11282.68 ms)
unthrottled	3	3.68	27.60	10.35	186.60	470,417	470,372	🟥 2026.1.2 (170.26 ms)

P99 Read Latency by Step (log scale)

Step	Version	P99 (ms)	Bar
250K	🟩	2.55	🟩
250K	🟦	2.14	🟦
250K	🟥	2.46	🟥
480K	🟩	5.67	🟩🟩
480K	🟦	6.20	🟦🟦
480K	🟥	3.67	🟥
600K	🟩	16,064.18	🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩
600K	🟦	4,861.20	🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦
600K	🟥	6.85	🟥🟥
750K	🟩	28,739.37	🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩
750K	🟦	18,102.62	🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦
750K	🟥	11,282.68	🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥

⚠️ No data for 🟧 2025.4.7 in mixed workload

Unthrottled Throughput

Version	Throughput (op/s)	Bar
🟩 2026.1.1	422,946	🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩🟩
🟦 2025.4.6	512,838	🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦🟦
🟥 2026.1.2	475,469	🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥

---

Key Observations

1. 🟥 2026.1.2 (latest release tested): All 4 runs failed — limited data. Requires investigation.
2. 🟧 2025.4.7: Only 1 of 4 runs passed. Limited data available.
3. 🟦 2025.4.6: 2 of 4 passed. Stable baseline. Best write performance across all steps.
4. 🟩 2026.1.1: 3 of 4 passed. Best read (disk & cache) performance.
5. Pass rate for official releases: 6/16 (37.5%) — indicates infrastructure or test stability issues beyond code regressions.
6. Bar widths use logarithmic scale — a bar twice as long means ~10× higher latency, not 2×.

---

Generated from Argus (argus.scylladb.com) — Test ID: d6ebf1a5-135f-43fc-a7ba-0716b60dfa94