[Benchmarks] Archive old data

PatKamin · PatKamin · commit d68f88396189 · 2025-06-25T19:26:16.000Z
Archive old benchmark data so it is not loaded in the dashboard.
The default value for archiving is:
- for Baseline_* runs for data older than 90 days,
- for all other runs for data older than 14 days.

Add a UI option for loading archived data.
diff --git a/devops/scripts/benchmarks/history.py b/devops/scripts/benchmarks/history.py
@@ -9,7 +9,7 @@
 import socket
 from utils.result import Result, BenchmarkRun
 from options import Compare, options
-from datetime import datetime, timezone
+from datetime import datetime, timezone, timedelta
 from utils.utils import run
 from utils.validate import Validate
 
@@ -223,3 +223,27 @@ def get_compare(self, name: str) -> BenchmarkRun:
             return self.compute_average(data)
 
         raise Exception("invalid compare type")
+
+    def partition_runs_by_age(self) -> tuple[list[BenchmarkRun], list[BenchmarkRun]]:
+        """
+        Partition runs into current and archived based on their age.
+        Returns:
+            tuple: (current_runs, archived_runs)
+        """
+        current_runs = []
+        archived_runs = []
+
+        for run in self.runs:
+            archive_after = (
+                options.archive_baseline_days
+                if run.name.startswith("Baseline_")
+                else options.archive_pr_days
+            )
+            cutoff_date = datetime.now(timezone.utc) - timedelta(days=archive_after)
+
+            if run.date > cutoff_date:
+                current_runs.append(run)
+            else:
+                archived_runs.append(run)
+
+        return current_runs, archived_runs
diff --git a/devops/scripts/benchmarks/html/index.html b/devops/scripts/benchmarks/html/index.html
@@ -53,6 +53,10 @@ <h3>Display Options</h3>
                             <input type="checkbox" id="custom-range">
                             Adjust Y-axis for comparisons
                         </label>
+                        <label title="Load older benchmark results that have been archived.">
+                            <input type="checkbox" id="show-archived-data">
+                            Include archived runs
+                        </label>
                     </div>
                 </div>
 
diff --git a/devops/scripts/benchmarks/html/scripts.js b/devops/scripts/benchmarks/html/scripts.js
@@ -7,21 +7,22 @@
 let activeRuns = new Set(defaultCompareNames);
 let chartInstances = new Map();
 let suiteNames = new Set();
-let timeseriesData, barChartsData, allRunNames;
 let activeTags = new Set();
+let timeseriesData, barChartsData, allRunNames;
 let layerComparisonsData;
 let latestRunsLookup = new Map();
 let pendingCharts = new Map(); // Store chart data for lazy loading
 let chartObserver; // Intersection observer for lazy loading charts
 let annotationsOptions = new Map(); // Global options map for annotations
+let archivedDataLoaded = false;
 
 // DOM Elements
 let runSelect, selectedRunsDiv, suiteFiltersContainer, tagFiltersContainer;
 
 // Observer for lazy loading charts
 function initChartObserver() {
     if (chartObserver) return;
-    
+
     chartObserver = new IntersectionObserver((entries) => {
         entries.forEach(entry => {
             if (entry.isIntersecting) {
@@ -196,7 +197,7 @@ function createChart(data, containerId, type) {
                 maxTicksLimit: 10
             }
         };
-        
+
         // Add dependencies version change annotations
         if (Object.keys(data.runs).length > 0) {
             ChartAnnotations.addVersionChangeAnnotations(data, options);
@@ -210,8 +211,8 @@ function createChart(data, containerId, type) {
                 ...runData,
                 // For timeseries (historical results charts) use runName,
                 // otherwise use displayLabel (for layer comparison charts)
-                label: containerId.startsWith('timeseries') ? 
-                    runData.runName : 
+                label: containerId.startsWith('timeseries') ?
+                    runData.runName :
                     (runData.displayLabel || runData.label)
             }))
         } : {
@@ -223,12 +224,12 @@ function createChart(data, containerId, type) {
 
     const chart = new Chart(ctx, chartConfig);
     chartInstances.set(containerId, chart);
-    
+
     // Add annotation interaction handlers for time-series charts
     if (type === 'time') {
         ChartAnnotations.setupAnnotationListeners(chart, ctx, options);
     }
-    
+
     return chart;
 }
 
@@ -263,7 +264,7 @@ function drawCharts(filteredTimeseriesData, filteredBarChartsData, filteredLayer
     chartInstances.forEach(chart => chart.destroy());
     chartInstances.clear();
     pendingCharts.clear();
-    
+
     initChartObserver(); // For lazy loading charts
 
     // Create timeseries charts
@@ -394,7 +395,7 @@ function metadataForLabel(label, type) {
     if (benchmarkMetadata[label]?.type === type) {
         return benchmarkMetadata[label];
     }
-    
+
     // Then fall back to prefix match for backward compatibility
     for (const [key, metadata] of Object.entries(benchmarkMetadata)) {
         if (metadata.type === type && label.startsWith(key)) {
@@ -578,6 +579,12 @@ function updateURL() {
         url.searchParams.set('customRange', 'true');
     }
 
+    if (!isArchivedDataEnabled()) {
+        url.searchParams.delete('archived');
+    } else {
+        url.searchParams.set('archived', 'true');
+    }
+
     history.replaceState(null, '', url);
 }
 
@@ -835,6 +842,9 @@ function setupRunSelector() {
     runSelect = document.getElementById('run-select');
     selectedRunsDiv = document.getElementById('selected-runs');
 
+    // Clear existing options first to prevent duplicates when reloading with archived data
+    runSelect.innerHTML = '';
+
     allRunNames.forEach(name => {
         const option = document.createElement('option');
         option.value = name;
@@ -848,6 +858,9 @@ function setupRunSelector() {
 function setupSuiteFilters() {
     suiteFiltersContainer = document.getElementById('suite-filters');
 
+    // Clear existing suite filters before adding new ones
+    suiteFiltersContainer.innerHTML = '';
+
     benchmarkRuns.forEach(run => {
         run.results.forEach(result => {
             suiteNames.add(result.suite);
@@ -883,10 +896,16 @@ function isCustomRangesEnabled() {
     return rangesToggle.checked;
 }
 
+function isArchivedDataEnabled() {
+    const archivedDataToggle = document.getElementById('show-archived-data');
+    return archivedDataToggle.checked;
+}
+
 function setupToggles() {
     const notesToggle = document.getElementById('show-notes');
     const unstableToggle = document.getElementById('show-unstable');
     const customRangeToggle = document.getElementById('custom-range');
+    const archivedDataToggle = document.getElementById('show-archived-data');
 
     notesToggle.addEventListener('change', function () {
         // Update all note elements visibility
@@ -909,9 +928,25 @@ function setupToggles() {
         updateCharts();
     });
 
+    // Add event listener for archived data toggle
+    if (archivedDataToggle) {
+        archivedDataToggle.addEventListener('change', function() {
+            if (archivedDataToggle.checked) {
+                loadArchivedData();
+            } else {
+                if (archivedDataLoaded) {
+                    // Reload the page to reset
+                    location.reload();
+                }
+            }
+            updateURL();
+        });
+    }
+
     // Initialize from URL params if present
     const notesParam = getQueryParam('notes');
     const unstableParam = getQueryParam('unstable');
+    const archivedParam = getQueryParam('archived');
 
     if (notesParam !== null) {
         let showNotes = notesParam === 'true';
@@ -927,11 +962,22 @@ function setupToggles() {
     if (customRangesParam !== null) {
         customRangeToggle.checked = customRangesParam === 'true';
     }
+
+    if (archivedDataToggle && archivedParam !== null) {
+        archivedDataToggle.checked = archivedParam === 'true';
+
+        if (archivedDataToggle.checked) {
+            loadArchivedData();
+        }
+    }
 }
 
 function setupTagFilters() {
     tagFiltersContainer = document.getElementById('tag-filters');
 
+    // Clear existing tag filters before adding new ones
+    tagFiltersContainer.innerHTML = '';
+
     const allTags = [];
 
     if (benchmarkTags) {
@@ -1087,38 +1133,124 @@ window.addSelectedRun = addSelectedRun;
 window.removeRun = removeRun;
 window.toggleAllTags = toggleAllTags;
 
+// Helper function to fetch and process benchmark data
+function fetchAndProcessData(url, isArchived = false) {
+    const loadingIndicator = document.getElementById('loading-indicator');
+
+    return fetch(url)
+        .then(response => {
+            if (!response.ok) { throw new Error(`Got response status ${response.status}.`) }
+            return response.json();
+        })
+        .then(data => {
+            const newRuns = data.runs || data;
+
+            if (isArchived) {
+                // Merge with existing data for archived data
+                benchmarkRuns = benchmarkRuns.concat(newRuns);
+
+                // Merge metadata and tags if available
+                if (data.metadata) {
+                    benchmarkMetadata = { ...benchmarkMetadata, ...data.metadata };
+                }
+
+                if (data.tags) {
+                    benchmarkTags = { ...benchmarkTags, ...data.tags };
+                }
+
+                archivedDataLoaded = true;
+            } else {
+                // Replace existing data for primary data
+                benchmarkRuns = newRuns;
+                benchmarkMetadata = data.metadata || benchmarkMetadata || {};
+                benchmarkTags = data.tags || benchmarkTags || {};
+            }
+
+            initializeCharts();
+        })
+        .catch(error => {
+            console.error(`Error fetching ${isArchived ? 'archived' : 'remote'} data:`, error);
+            loadingIndicator.textContent = 'Fetching remote data failed.';
+        })
+        .finally(() => {
+            loadingIndicator.style.display = 'none';
+        });
+}
+
 // Load data based on configuration
 function loadData() {
     const loadingIndicator = document.getElementById('loading-indicator');
     loadingIndicator.style.display = 'block'; // Show loading indicator
 
     if (typeof remoteDataUrl !== 'undefined' && remoteDataUrl !== '') {
         // Fetch data from remote URL
-        fetch(remoteDataUrl)
-            .then(response => {
-                if (!response.ok) { throw new Error(`Got response status ${response.status}.`) }
-                return response.json();
-            })
-            .then(data => {
-                benchmarkRuns = data.runs || data;
-                benchmarkMetadata = data.metadata || benchmarkMetadata || {};
-                benchmarkTags = data.tags || benchmarkTags || {};
-                initializeCharts();
-            })
-            .catch(error => {
-                console.error('Error fetching remote data:', error);
-                loadingIndicator.textContent = 'Fetching remote data failed.';
-            })
-            .finally(() => {
-                loadingIndicator.style.display = 'none'; // Hide loading indicator
-            });
+        fetchAndProcessData(remoteDataUrl);
     } else {
         // Use local data (benchmarkRuns and benchmarkMetadata should be defined in data.js)
         initializeCharts();
         loadingIndicator.style.display = 'none'; // Hide loading indicator
     }
 }
 
+// Function to load archived data and merge with current data
+// Archived data consists of older benchmark results that have been separated from
+// the primary dataset but are still available for historical analysis.
+function loadArchivedData() {
+    const loadingIndicator = document.getElementById('loading-indicator');
+    loadingIndicator.style.display = 'block';
+
+    if (archivedDataLoaded) {
+        updateCharts();
+        loadingIndicator.style.display = 'none';
+        return;
+    }
+
+    if (typeof remoteDataUrl !== 'undefined' && remoteDataUrl !== '') {
+        // For remote data, construct the archive URL by adding _archive before the extension
+        const archiveUrl = remoteDataUrl.replace(/(\.[^.]+)$/, '_archive$1');
+
+        // Check if we're using local JSON files
+        if (remoteDataUrl.startsWith('./') && remoteDataUrl.endsWith('.json')) {
+            fetchAndProcessData(archiveUrl, true);
+        } else {
+            fetchAndProcessData(archiveUrl, true);
+        }
+    } else {
+        // For local data using a static js file
+        const script = document.createElement('script');
+        script.src = 'data_archive.js';
+        script.onload = () => {
+            if (typeof archivedBenchmarkRuns !== 'undefined') {
+                // Merge the archived runs with current runs
+                benchmarkRuns = benchmarkRuns.concat(archivedBenchmarkRuns);
+
+                // Merge metadata and tags if available
+                if (typeof archivedBenchmarkMetadata !== 'undefined') {
+                    benchmarkMetadata = { ...benchmarkMetadata, ...archivedBenchmarkMetadata };
+                }
+
+                if (typeof archivedBenchmarkTags !== 'undefined') {
+                    benchmarkTags = { ...benchmarkTags, ...archivedBenchmarkTags };
+                }
+
+                archivedDataLoaded = true;
+
+                initializeCharts();
+            } else {
+                console.error('Archived runs data not found in data_archive.js');
+            }
+            loadingIndicator.style.display = 'none';
+        };
+
+        script.onerror = () => {
+            console.error('Failed to load data_archive.js');
+            loadingIndicator.style.display = 'none';
+        };
+
+        document.head.appendChild(script);
+    }
+}
+
 // Initialize when DOM is ready
 document.addEventListener('DOMContentLoaded', () => {
     loadData();
diff --git a/devops/scripts/benchmarks/main.py b/devops/scripts/benchmarks/main.py
@@ -316,7 +316,7 @@ def main(directory, additional_env_vars, save_name, compare_names, filter):
         if options.output_directory is None:
             html_path = os.path.join(os.path.dirname(__file__), "html")
 
-        generate_html(history.runs, compare_names, html_path, metadata)
+        generate_html(history, compare_names, html_path, metadata)
 
 
 def validate_and_parse_env_args(env_args):
@@ -558,6 +558,22 @@ def validate_and_parse_env_args(env_args):
         help="Location of detect_version.cpp used to query e.g. DPC++, L0",
         default=None,
     )
+    parser.add_argument(
+        "--archive-baseline-after",
+        type=int,
+        help="Archive baseline results (runs starting with 'Baseline_') older than this many days. "
+        "Archived results are stored separately and can be viewed in the HTML UI by enabling "
+        "'Include archived runs'. This helps manage the size of the primary dataset.",
+        default=options.archive_baseline_days,
+    )
+    parser.add_argument(
+        "--archive-pr-after",
+        type=int,
+        help="Archive PR and other non-baseline results older than this many days. "
+        "Archived results are stored separately and can be viewed in the HTML UI by enabling "
+        "'Include archived runs'. PR runs typically have a shorter retention period than baselines.",
+        default=options.archive_pr_days,
+    )
 
     args = parser.parse_args()
     additional_env_vars = validate_and_parse_env_args(args.env)
diff --git a/devops/scripts/benchmarks/options.py b/devops/scripts/benchmarks/options.py
diff --git a/devops/scripts/benchmarks/output_html.py b/devops/scripts/benchmarks/output_html.py
