Split NVIDIA tasks by homogeneous GPU groups

This commit is contained in:
Mikhail Chusavitin
2026-04-06 11:58:13 +03:00
parent fc5c100a29
commit 981315e6fd
5 changed files with 569 additions and 75 deletions
+98 -36
View File
@@ -1656,6 +1656,12 @@ func renderBenchmark(opts HandlerOptions) string {
<script>
let benchmarkES = null;
function benchmarkTaskIDs(payload) {
if (payload && Array.isArray(payload.task_ids) && payload.task_ids.length) return payload.task_ids;
if (payload && payload.task_id) return [payload.task_id];
return [];
}
function benchmarkSelectedGPUIndices() {
return Array.from(document.querySelectorAll('.benchmark-gpu-checkbox'))
.filter(function(el) { return el.checked && !el.disabled; })
@@ -1755,17 +1761,37 @@ function runNvidiaBenchmark() {
return payload;
});
}).then(function(d) {
status.textContent = 'Task ' + d.task_id + ' queued.';
term.textContent += 'Task ' + d.task_id + ' queued. Streaming log...\n';
benchmarkES = new EventSource('/api/tasks/' + d.task_id + '/stream');
benchmarkES.onmessage = function(e) { term.textContent += e.data + '\n'; term.scrollTop = term.scrollHeight; };
benchmarkES.addEventListener('done', function(e) {
benchmarkES.close();
benchmarkES = null;
term.textContent += (e.data ? '\nERROR: ' + e.data : '\nCompleted.') + '\n';
term.scrollTop = term.scrollHeight;
status.textContent = e.data ? 'Failed.' : 'Completed.';
});
const taskIds = benchmarkTaskIDs(d);
if (!taskIds.length) throw new Error('No benchmark task was queued.');
status.textContent = taskIds.length === 1 ? ('Task ' + taskIds[0] + ' queued.') : ('Queued ' + taskIds.length + ' tasks.');
const streamNext = function(idx, failures) {
if (idx >= taskIds.length) {
status.textContent = failures ? 'Completed with failures.' : 'Completed.';
return;
}
const taskId = taskIds[idx];
term.textContent += '\n[' + (idx + 1) + '/' + taskIds.length + '] Task ' + taskId + ' queued. Streaming log...\n';
benchmarkES = new EventSource('/api/tasks/' + taskId + '/stream');
benchmarkES.onmessage = function(e) { term.textContent += e.data + '\n'; term.scrollTop = term.scrollHeight; };
benchmarkES.addEventListener('done', function(e) {
benchmarkES.close();
benchmarkES = null;
if (e.data) failures += 1;
term.textContent += (e.data ? '\nERROR: ' + e.data : '\nCompleted.') + '\n';
term.scrollTop = term.scrollHeight;
streamNext(idx + 1, failures);
});
benchmarkES.onerror = function() {
if (benchmarkES) {
benchmarkES.close();
benchmarkES = null;
}
term.textContent += '\nERROR: stream disconnected.\n';
term.scrollTop = term.scrollHeight;
streamNext(idx + 1, failures + 1);
};
};
streamNext(0, 0);
}).catch(function(err) {
status.textContent = 'Error.';
term.textContent += 'ERROR: ' + err.message + '\n';
@@ -1779,13 +1805,24 @@ benchmarkLoadGPUs();
func renderBenchmarkResultsCard(exportDir string) string {
columns, runs := loadBenchmarkHistory(exportDir)
if len(runs) == 0 {
return `<div class="card"><div class="card-head">Benchmark Results</div><div class="card-body"><p style="color:var(--muted);font-size:13px">No saved benchmark runs yet.</p></div></div>`
}
return renderBenchmarkResultsCardFromRuns(
"Benchmark Results",
"Composite score by saved benchmark run and GPU.",
"No saved benchmark runs yet.",
columns,
runs,
)
}
func renderBenchmarkResultsCardFromRuns(title, description, emptyMessage string, columns []benchmarkHistoryColumn, runs []benchmarkHistoryRun) string {
if len(runs) == 0 {
return `<div class="card"><div class="card-head">` + html.EscapeString(title) + `</div><div class="card-body"><p style="color:var(--muted);font-size:13px">` + html.EscapeString(emptyMessage) + `</p></div></div>`
}
var b strings.Builder
b.WriteString(`<div class="card"><div class="card-head">Benchmark Results</div><div class="card-body">`)
b.WriteString(`<p style="color:var(--muted);font-size:13px;margin-bottom:12px">Composite score by saved benchmark run and GPU.</p>`)
b.WriteString(`<div class="card"><div class="card-head">` + html.EscapeString(title) + `</div><div class="card-body">`)
if strings.TrimSpace(description) != "" {
b.WriteString(`<p style="color:var(--muted);font-size:13px;margin-bottom:12px">` + html.EscapeString(description) + `</p>`)
}
b.WriteString(`<div style="overflow-x:auto">`)
b.WriteString(`<table><thead><tr><th>Test</th><th>Time</th>`)
for _, col := range columns {
@@ -1820,7 +1857,10 @@ func loadBenchmarkHistory(exportDir string) ([]benchmarkHistoryColumn, []benchma
return nil, nil
}
sort.Strings(paths)
return loadBenchmarkHistoryFromPaths(paths)
}
func loadBenchmarkHistoryFromPaths(paths []string) ([]benchmarkHistoryColumn, []benchmarkHistoryRun) {
columnByKey := make(map[string]benchmarkHistoryColumn)
runs := make([]benchmarkHistoryRun, 0, len(paths))
for _, path := range paths {
@@ -2005,6 +2045,12 @@ func renderBurn() string {
<script>
let biES = null;
function burnTaskIDs(payload) {
if (payload && Array.isArray(payload.task_ids) && payload.task_ids.length) return payload.task_ids;
if (payload && payload.task_id) return [payload.task_id];
return [];
}
function burnProfile() {
const selected = document.querySelector('input[name="burn-profile"]:checked');
return selected ? selected.value : 'smoke';
@@ -2106,6 +2152,9 @@ function streamTask(taskId, label) {
});
}
function streamBurnTask(taskId, label, resetTerminal) {
return streamBurnTaskSet([taskId], label, resetTerminal);
}
function streamBurnTaskSet(taskIds, label, resetTerminal) {
if (biES) { biES.close(); biES = null; }
document.getElementById('bi-output').style.display = 'block';
document.getElementById('bi-title').textContent = '— ' + label + ' [' + burnProfile() + ']';
@@ -2113,27 +2162,40 @@ function streamBurnTask(taskId, label, resetTerminal) {
if (resetTerminal) {
term.textContent = '';
}
term.textContent += 'Task ' + taskId + ' queued. Streaming...\n';
return new Promise(function(resolve) {
biES = new EventSource('/api/tasks/' + taskId + '/stream');
biES.onmessage = function(e) { term.textContent += e.data + '\n'; term.scrollTop = term.scrollHeight; };
biES.addEventListener('done', function(e) {
biES.close();
biES = null;
term.textContent += (e.data ? '\nERROR: ' + e.data : '\nCompleted.') + '\n';
term.scrollTop = term.scrollHeight;
resolve({ok: !e.data, error: e.data || ''});
});
biES.onerror = function() {
if (biES) {
if (!Array.isArray(taskIds) || !taskIds.length) {
term.textContent += 'ERROR: no tasks queued.\n';
return Promise.resolve({ok:false, error:'no tasks queued'});
}
const streamNext = function(idx, failures) {
if (idx >= taskIds.length) {
return Promise.resolve({ok: failures === 0, error: failures ? (failures + ' task(s) failed') : ''});
}
const taskId = taskIds[idx];
term.textContent += '[' + (idx + 1) + '/' + taskIds.length + '] Task ' + taskId + ' queued. Streaming...\n';
return new Promise(function(resolve) {
biES = new EventSource('/api/tasks/' + taskId + '/stream');
biES.onmessage = function(e) { term.textContent += e.data + '\n'; term.scrollTop = term.scrollHeight; };
biES.addEventListener('done', function(e) {
biES.close();
biES = null;
}
term.textContent += '\nERROR: stream disconnected.\n';
term.scrollTop = term.scrollHeight;
resolve({ok: false, error: 'stream disconnected'});
};
});
term.textContent += (e.data ? '\nERROR: ' + e.data : '\nCompleted.') + '\n';
term.scrollTop = term.scrollHeight;
resolve(failures + (e.data ? 1 : 0));
});
biES.onerror = function() {
if (biES) {
biES.close();
biES = null;
}
term.textContent += '\nERROR: stream disconnected.\n';
term.scrollTop = term.scrollHeight;
resolve(failures + 1);
};
}).then(function(nextFailures) {
return streamNext(idx + 1, nextFailures);
});
};
return streamNext(0, 0);
}
function runBurnTaskSet(tasks, statusElId) {
@@ -2161,7 +2223,7 @@ function runBurnTaskSet(tasks, statusElId) {
if (status) status.textContent = 'Running ' + (idx + 1) + '/' + enabled.length + '...';
return enqueueBurnTask(t.target, t.label, t.extra, !!t.nvidia)
.then(function(d) {
return streamBurnTask(d.task_id, t.label, false);
return streamBurnTaskSet(burnTaskIDs(d), t.label, false);
})
.then(function() {
return runNext(idx + 1);