Put the fleet view in the dashboard, so nothing else is needed to see it
CI / test (push) Successful in 1m32s
CI / test (push) Successful in 1m32s
The metrics were exported and the only way to look at them was to stand up Prometheus beside the binary. For a project whose whole shape is one static binary with no dependencies, that is the wrong default: the common case is somebody wanting to know how full their cluster is, and that should not cost a monitoring stack. /admin now shows capacity per GPU class with a usage bar, free whole nodes per fabric — the number that decides whether a replica spanning nodes can land, since free cards on the wrong fabric are not capacity for a model that needs this one — and which replicas are holding hardware nobody is asking for. Nodes gain their memory and how many model files they can see. Taken from the same Summary that ctl status uses rather than derived in the page, because two implementations of the same arithmetic drift apart and the one on the screen is the one people would believe. A faulty card counts as neither used nor free, matching the scheduler. Counting it free is how a cluster looks like it has room it does not, and a page that disagreed with placement about that would be worse than no page. Prometheus stays supported and is now the answer to a narrower question: history, alerting, several clusters at once. The dashboard shows what is true now and keeps nothing, which is the honest division between them. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
1 parent
1ea067d08e
commit
c232f41a66
7 files changed
+158
-5
No files matched your search
+11
-2
@@ -564,8 +564,17 @@ serving — requests, latency, time to first token, tokens, admission, per
|
||||
endpoint — and what it is made of, per node and per GPU, including memory,
|
||||
temperature, ECC faults and utilisation.
|
||||
|
||||
Metrics get their own listener, because the admin one also carries pprof and
|
||||
pprof can dump the process heap, where API keys sit in plaintext:
|
||||
**The dashboard is the answer for most clusters.** `/admin` shows capacity per
|
||||
GPU class, free whole nodes per fabric, which replicas are holding hardware
|
||||
nothing is asking for, and the same node and model tables `ctl` prints — with
|
||||
no monitoring stack to stand up beside it. One binary, no dependencies, which
|
||||
is the point.
|
||||
|
||||
Prometheus is worth adding when you want history, alerts, or one view across
|
||||
several clusters. The dashboard shows what is true now; it keeps nothing.
|
||||
|
||||
Metrics then get their own listener, because the admin one also carries pprof
|
||||
and pprof can dump the process heap, where API keys sit in plaintext:
|
||||
|
||||
```yaml
|
||||
admin_listen: "127.0.0.1:9090" # dashboard, management API, pprof — keep local
|
||||
|
||||
@@ -18,6 +18,7 @@ type DashboardData struct {
|
||||
Replicas []adminapi.ReplicaView
|
||||
Unplaced []adminapi.ExplainView
|
||||
Usage []store.KeyUsage
|
||||
Summary adminapi.Summary
|
||||
|
||||
// Charts are pre-rendered rather than assembled in the template: building
|
||||
// SVG out of template fragments is how a chart ends up unescaped.
|
||||
@@ -53,6 +54,14 @@ type Source interface {
|
||||
Models() []adminapi.ModelView
|
||||
Replicas(model string) []adminapi.ReplicaView
|
||||
Explain(model string) adminapi.ExplainView
|
||||
// Summary is the fleet in aggregate: capacity per GPU class, free whole
|
||||
// nodes per fabric, what is idle.
|
||||
//
|
||||
// The tables below say what exists; this says how much of it is in use,
|
||||
// which is a different question and the one asked first. Deriving it in the
|
||||
// page would mean two implementations of the same arithmetic drifting
|
||||
// apart, so it comes from where ctl status gets it.
|
||||
Summary() adminapi.Summary
|
||||
}
|
||||
|
||||
// UsageSource is the accounting half, absent when authentication is off.
|
||||
@@ -103,6 +112,7 @@ func Dashboard(src Source, usage UsageSource, version string, refresh time.Durat
|
||||
}
|
||||
}
|
||||
d.Unplaced = unplaced(src, d.Models)
|
||||
d.Summary = src.Summary()
|
||||
}
|
||||
if usage != nil {
|
||||
rows, err := usage.UsageSince(r.Context(), time.Now().Add(-24*time.Hour))
|
||||
|
||||
@@ -110,3 +110,13 @@ select { font: inherit; padding: .35rem .4rem; border: 1px solid var(--line);
|
||||
.note.truncated { color: var(--warn, #a15c00); margin: .2rem 0 .8rem; }
|
||||
/* Offered under a reply that contains files worth keeping. */
|
||||
.link.download { display: inline-block; margin: .3rem 0 .8rem; }
|
||||
|
||||
/* A usage bar inside a table cell. Drawn with two elements rather than a
|
||||
gradient so the number stays selectable and readable by a screen reader. */
|
||||
.meter { position: relative; min-width: 9rem; }
|
||||
.meter > span {
|
||||
display: block; height: .75rem; border-radius: 2px;
|
||||
background: var(--accent); opacity: .55;
|
||||
}
|
||||
.meter > em { font-style: normal; font-size: .8rem; color: var(--muted); }
|
||||
.bad { color: var(--bad); }
|
||||
@@ -20,19 +20,57 @@
|
||||
</nav>
|
||||
</header>
|
||||
|
||||
<h2>Capacity</h2>
|
||||
{{if .Summary.GPUs}}
|
||||
<table><tr><th>Class</th><th class="num">In use</th><th class="num">Free</th><th class="num">Faulty</th><th>Usage</th></tr>
|
||||
{{range .Summary.GPUs}}<tr>
|
||||
<td><code>{{.Class}}</code></td>
|
||||
<td class="num">{{.Used}}/{{.Total}}</td>
|
||||
<td class="num">{{.Free}}</td>
|
||||
<td class="num">{{if .Unhealthy}}<span class="bad">{{.Unhealthy}}</span>{{else}}—{{end}}</td>
|
||||
<td class="meter"><span style="width:{{pct .Used .Total}}%"></span><em>{{pct .Used .Total}}%</em></td>
|
||||
</tr>{{end}}</table>
|
||||
{{else}}<p class="empty">No GPUs have been reported.</p>{{end}}
|
||||
|
||||
{{if .Summary.Fabrics}}
|
||||
<h2>Fabrics</h2>
|
||||
<table><tr><th>Fabric</th><th class="num">Nodes</th><th class="num">Fully free</th><th>Weights</th></tr>
|
||||
{{range .Summary.Fabrics}}<tr>
|
||||
<td><code>{{.Fabric}}</code></td>
|
||||
<td class="num">{{.Nodes}}</td>
|
||||
<td class="num">{{.LargestFreeBlock}}</td>
|
||||
<td>{{if .NodesWithWeights}}{{.Models}} model(s) on {{.NodesWithWeights}}/{{.Nodes}} nodes{{else}}<span class="note">not reported</span>{{end}}</td>
|
||||
</tr>{{end}}</table>
|
||||
<p class="note">Fully free counts whole unused nodes, which is what decides whether a
|
||||
replica spanning nodes can land: free cards on the wrong fabric are not capacity
|
||||
for a model that needs this one.</p>
|
||||
{{end}}
|
||||
|
||||
{{if .Summary.IdleReplicas}}
|
||||
<h2>Holding hardware nothing is asking for</h2>
|
||||
<table><tr><th>Replica</th><th>Model</th><th class="num">GPUs</th><th class="num">Unused for</th></tr>
|
||||
{{range .Summary.IdleReplicas}}<tr>
|
||||
<td><code>{{.ID}}</code></td>
|
||||
<td>{{.Model}}</td>
|
||||
<td class="num">{{.GPUs}}</td>
|
||||
<td class="num">{{duration .IdleSec}}</td>
|
||||
</tr>{{end}}</table>
|
||||
{{end}}
|
||||
|
||||
<h2>Nodes</h2>
|
||||
{{if .Nodes}}
|
||||
<table><tr><th>Node</th><th>State</th><th class="num">GPUs</th><th>Classes</th><th>Fabrics</th><th class="num">Replicas</th><th>Last seen</th><th>Agent</th></tr>
|
||||
<table><tr><th>Node</th><th>State</th><th class="num">GPUs</th><th class="num">RAM</th><th>Classes</th><th>Fabrics</th><th class="num">Models</th><th class="num">Replicas</th><th>Last seen</th></tr>
|
||||
{{range .Nodes}}<tr>
|
||||
<td><code>{{.ID}}</code>{{if .MgmtAddr}} <span class="note">{{.MgmtAddr}}</span>{{end}}</td>
|
||||
<td class="state state-{{.State}}">{{.State}}</td>
|
||||
<td class="num">{{.GPUFree}}/{{.GPUTotal}}</td>
|
||||
<td class="num">{{if .MemoryGB}}{{.MemoryGB}}G{{else}}—{{end}}</td>
|
||||
<td>{{join .Classes}}</td>
|
||||
<td>{{join .Fabrics}}</td>
|
||||
<td class="num">{{if .Models}}{{len .Models}}{{else}}—{{end}}</td>
|
||||
<td class="num">{{.Replicas}}</td>
|
||||
<td>{{or .LastSeenAgo "—"}}</td>
|
||||
<td><span class="note">{{or .AgentVersion "—"}}</span></td>
|
||||
</tr>{{if .Unhealthy}}<tr><td colspan="8" class="note">{{join .Unhealthy}}</td></tr>{{end}}
|
||||
</tr>{{if .Unhealthy}}<tr><td colspan="9" class="note bad">{{join .Unhealthy}}</td></tr>{{end}}
|
||||
{{end}}</table>
|
||||
{{else}}<p class="empty">No agents have registered.</p>{{end}}
|
||||
|
||||
|
||||
@@ -45,6 +45,22 @@ var funcs = template.FuncMap{
|
||||
"duration": func(sec int64) string {
|
||||
return time.Duration(sec * int64(time.Second)).Round(time.Second).String()
|
||||
},
|
||||
// pct is for the usage meters, and is clamped because the two counts it
|
||||
// divides can disagree for an instant — a GPU freed between one read and
|
||||
// the next — and a bar wider than its column wrecks the table it is in.
|
||||
"pct": func(used, total int) int {
|
||||
if total <= 0 {
|
||||
return 0
|
||||
}
|
||||
p := used * 100 / total
|
||||
if p < 0 {
|
||||
return 0
|
||||
}
|
||||
if p > 100 {
|
||||
return 100
|
||||
}
|
||||
return p
|
||||
},
|
||||
"ago": func(t time.Time) string {
|
||||
if t.IsZero() {
|
||||
return "never"
|
||||
|
||||
@@ -107,6 +107,21 @@ func (fakeSource) Nodes() []adminapi.NodeView {
|
||||
}}
|
||||
}
|
||||
|
||||
func (fakeSource) Summary() adminapi.Summary {
|
||||
return adminapi.Summary{
|
||||
Nodes: adminapi.NodeTotals{Total: 1, Ready: 1},
|
||||
GPUs: []adminapi.ClassUse{
|
||||
{Class: "h100-80g", Total: 8, Used: 3, Unhealthy: 1},
|
||||
},
|
||||
Fabrics: []adminapi.FabricUse{
|
||||
{Fabric: "ib-a", Nodes: 1, LargestFreeBlock: 0, NodesWithWeights: 1, Models: 2},
|
||||
},
|
||||
IdleReplicas: []adminapi.IdleReplica{
|
||||
{ID: "served-abc", Model: "served", IdleSec: 3600, GPUs: 4},
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
func (fakeSource) Models() []adminapi.ModelView {
|
||||
return []adminapi.ModelView{
|
||||
{Name: "served", Ready: 1, Min: 1, Max: 2, Profiles: []string{"tp4"}},
|
||||
@@ -225,6 +240,7 @@ func (emptySource) Nodes() []adminapi.NodeView { return nil }
|
||||
func (emptySource) Models() []adminapi.ModelView { return nil }
|
||||
func (emptySource) Replicas(string) []adminapi.ReplicaView { return nil }
|
||||
func (emptySource) Explain(string) adminapi.ExplainView { return adminapi.ExplainView{} }
|
||||
func (emptySource) Summary() adminapi.Summary { return adminapi.Summary{} }
|
||||
|
||||
func TestStaticServesAssetsAndNothingElse(t *testing.T) {
|
||||
// The assets live in their own embed. A file server over the tree that also
|
||||
@@ -409,3 +425,56 @@ func TestTheDashboardOffersOnlyLinksThatExistWhereItIsServed(t *testing.T) {
|
||||
t.Error("the public page does not say where metrics and pprof live")
|
||||
}
|
||||
}
|
||||
|
||||
func TestTheDashboardShowsCapacityWithoutPrometheus(t *testing.T) {
|
||||
// The whole point of putting this on the page: knowing how full a cluster
|
||||
// is should not require standing up a monitoring stack beside it. The
|
||||
// tables say what exists; this says how much of it is in use, which is the
|
||||
// question asked first.
|
||||
h := Dashboard(fakeSource{}, fakeUsage{}, "v1", time.Second, true, nil)
|
||||
code, page := get(t, h, "http://127.0.0.1:9090/", nil)
|
||||
if code != http.StatusOK {
|
||||
t.Fatalf("status %d", code)
|
||||
}
|
||||
for _, want := range []string{
|
||||
"Capacity", "h100-80g", "3/8", // used of total
|
||||
"Fabrics", "ib-a",
|
||||
"Holding hardware nothing is asking for", "served-abc",
|
||||
} {
|
||||
if !strings.Contains(page, want) {
|
||||
t.Errorf("the dashboard does not show %q", want)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestAFaultyCardIsShownAsNeitherUsedNorFree(t *testing.T) {
|
||||
// Counting a faulty card as free is how a cluster looks like it has room it
|
||||
// does not, and the page must not disagree with the scheduler about that.
|
||||
h := Dashboard(fakeSource{}, fakeUsage{}, "v1", time.Second, true, nil)
|
||||
code, page := get(t, h, "http://127.0.0.1:9090/", nil)
|
||||
if code != http.StatusOK {
|
||||
t.Fatalf("status %d", code)
|
||||
}
|
||||
if !strings.Contains(page, "Faulty") {
|
||||
t.Error("faulty cards are not shown")
|
||||
}
|
||||
// 8 total, 3 used, 1 faulty, so 4 free.
|
||||
if !strings.Contains(page, ">4<") {
|
||||
t.Errorf("free count is wrong or missing:\n%s", page)
|
||||
}
|
||||
}
|
||||
|
||||
func TestAUsageMeterCannotOverrunItsColumn(t *testing.T) {
|
||||
// The two counts it divides can disagree for an instant — a GPU freed
|
||||
// between one read and the next — and a bar wider than its cell wrecks the
|
||||
// table.
|
||||
pct, ok := funcs["pct"].(func(int, int) int)
|
||||
if !ok {
|
||||
t.Fatal("the pct helper is missing")
|
||||
}
|
||||
for _, tc := range [][2]int{{0, 0}, {9, 8}, {-1, 8}, {100, 8}} {
|
||||
if got := pct(tc[0], tc[1]); got < 0 || got > 100 {
|
||||
t.Errorf("pct(%d,%d) = %d", tc[0], tc[1], got)
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -471,6 +471,7 @@ func (fakeCluster) Nodes() []adminapi.NodeView {
|
||||
}
|
||||
func (fakeCluster) Models() []adminapi.ModelView { return nil }
|
||||
func (fakeCluster) Replicas(string) []adminapi.ReplicaView { return nil }
|
||||
func (fakeCluster) Summary() adminapi.Summary { return adminapi.Summary{} }
|
||||
func (fakeCluster) Explain(string) adminapi.ExplainView { return adminapi.ExplainView{} }
|
||||
|
||||
func TestTheChatPageTalksToTheModelWithoutTheBrowserHoldingAKey(t *testing.T) {
|
||||
|
||||
Reference in new issue
Block a user