Add VictoriaMetrics for historical metrics (Mar 13+)

- Single-node VM deployment with 200Gi NVMe, 2y retention
- Traefik IngressRoute at vm.vultrlabs.dev (TLS + basic auth)
- Backfill script: pulls vLLM/DCGM metrics from Mimir, writes to VM
- Retain StorageClass so historical data survives PVC deletion
- README with deployment + Grafana mixed-datasource instructions
This commit is contained in:
2026-04-09 19:29:18 +00:00
parent 7ade5ecac8
commit bf6d62b9a8
10 changed files with 690 additions and 0 deletions

View File

@@ -0,0 +1,10 @@
##############################################################################
# Namespace for VictoriaMetrics (historical metrics store)
##############################################################################
apiVersion: v1
kind: Namespace
metadata:
name: victoriametrics
labels:
app.kubernetes.io/part-of: victoriametrics

View File

@@ -0,0 +1,16 @@
##############################################################################
# StorageClass — Vultr Block Storage CSI (for VictoriaMetrics)
# Separate StorageClass with Retain policy so historical data isn't lost
##############################################################################
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
name: vultr-block-storage-vm
provisioner: block.csi.vultr.com
parameters:
disk_type: "nvme"
storage_type: "block"
reclaimPolicy: Retain # Keep the volume even if PVC is deleted
allowVolumeExpansion: true
volumeBindingMode: WaitForFirstConsumer

View File

@@ -0,0 +1,105 @@
##############################################################################
# VictoriaMetrics Single-Node Deployment
# Stores historical metrics from Mimir (Mar 13present) for Grafana queries
##############################################################################
apiVersion: apps/v1
kind: Deployment
metadata:
name: victoriametrics
namespace: victoriametrics
labels:
app.kubernetes.io/name: victoriametrics
spec:
replicas: 1
selector:
matchLabels:
app.kubernetes.io/name: victoriametrics
template:
metadata:
labels:
app.kubernetes.io/name: victoriametrics
annotations:
prometheus.io/scrape: "true"
prometheus.io/port: "8428"
spec:
securityContext:
fsGroup: 65534
containers:
- name: victoriametrics
image: victoriametrics/victoria-metrics:v1.115.0
args:
- "-storageDataPath=/data"
- "-retentionPeriod=2y" # Keep historical data for 2 years
- "-httpListenAddr=:8428"
- "-search.maxQueryDuration=120s" # Long-running queries OK for historical
- "-search.maxSamplesPerQuery=100000000" # High limit for wide historical queries
- "-memory.allowedBytes=4GB" # Memory budget
- "-search.maxUniqueTimeseries=5000000" # Allow high cardinality
ports:
- name: http
containerPort: 8428
volumeMounts:
- name: data
mountPath: /data
resources:
requests:
cpu: "2"
memory: 4Gi
limits:
cpu: "4"
memory: 8Gi
livenessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 30
periodSeconds: 15
readinessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 10
periodSeconds: 5
volumes:
- name: data
persistentVolumeClaim:
claimName: victoriametrics-data
---
##############################################################################
# PVC — Vultr Block Storage for VictoriaMetrics data
##############################################################################
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: victoriametrics-data
namespace: victoriametrics
spec:
storageClassName: vultr-block-storage-vm
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 200Gi
---
##############################################################################
# Service — ClusterIP (Traefik handles external access)
##############################################################################
apiVersion: v1
kind: Service
metadata:
name: victoriametrics
namespace: victoriametrics
labels:
app.kubernetes.io/name: victoriametrics
spec:
selector:
app.kubernetes.io/name: victoriametrics
ports:
- name: http
port: 8428
targetPort: http

View File

@@ -0,0 +1,58 @@
##############################################################################
# VictoriaMetrics Traefik IngressRoute
# External: https://vm.vultrlabs.dev → Traefik → victoriametrics:8428
##############################################################################
---
# HTTP redirect to HTTPS
apiVersion: traefik.io/v1alpha1
kind: IngressRoute
metadata:
name: victoriametrics-redirect
namespace: victoriametrics
spec:
entryPoints:
- web
routes:
- match: Host(`vm.vultrlabs.dev`)
kind: Rule
middlewares:
- name: redirect-https
namespace: victoriametrics
services:
- name: victoriametrics
port: 8428
---
# HTTPS with basic auth
apiVersion: traefik.io/v1alpha1
kind: IngressRoute
metadata:
name: victoriametrics
namespace: victoriametrics
spec:
entryPoints:
- websecure
routes:
- match: Host(`vm.vultrlabs.dev`)
kind: Rule
middlewares:
- name: basic-auth
namespace: victoriametrics
services:
- name: victoriametrics
port: 8428
tls:
certResolver: letsencrypt
---
# HTTPS redirect middleware
apiVersion: traefik.io/v1alpha1
kind: Middleware
metadata:
name: redirect-https
namespace: victoriametrics
spec:
redirectScheme:
scheme: https
permanent: true

View File

@@ -0,0 +1,33 @@
##############################################################################
# Basic Auth Middleware for VictoriaMetrics Traefik IngressRoute
# CHANGE THE PASSWORD BEFORE PRODUCTION USE!
#
# To generate a new htpasswd entry:
# htpasswd -nb <username> <password>
# Then base64 encode it:
# echo -n '<htpasswd-output>' | base64
# Update the secret below with the new value.
##############################################################################
---
apiVersion: v1
kind: Secret
metadata:
name: basic-auth-secret
namespace: victoriametrics
type: Opaque
# Generate with: htpasswd -nb vultr_vm <password> | base64
# See .env for credentials
stringData:
users: |-
vultr_vm:$apr1$ZtK5B1K4$SCWPgREqKwfcrCr4FA6En1
---
apiVersion: traefik.io/v1alpha1
kind: Middleware
metadata:
name: basic-auth
namespace: victoriametrics
spec:
basicAuth:
secret: basic-auth-secret

View File

@@ -0,0 +1,17 @@
##############################################################################
# Secrets for backfill (Mimir credentials)
# IMPORTANT: Update the password before running!
#
# To create the secret:
# kubectl create secret generic backfill-credentials \
# --from-literal=mimir-password='YOUR_PASSWORD' -n victoriametrics
##############################################################################
apiVersion: v1
kind: Secret
metadata:
name: backfill-credentials
namespace: victoriametrics
type: Opaque
stringData:
mimir-password: "REPLACE_WITH_MIMIR_PASSWORD"

185
victoriametrics/README.md Normal file
View File

@@ -0,0 +1,185 @@
# VictoriaMetrics — Historical Metrics Store
VictoriaMetrics instance for querying historical vLLM + DCGM metrics (March 13, 2026 onward) that couldn't be backfilled into M3DB.
## Why VictoriaMetrics Instead of M3DB?
M3DB doesn't support backfill. Period. See the [main README](../README.md#why-backfill-doesnt-work) for the full story.
VictoriaMetrics has a first-class `/api/v1/import` endpoint that accepts data with any timestamp — no `bufferPast` gates, no block size hacks, no special namespaces. You just send the data and it works.
## Architecture
```
┌─────────────────────────────────────────────────┐
│ Vultr VKE Cluster │
│ │
Mimir ──import──▶ VictoriaMetrics (1 pod, 200Gi NVMe) │
│ ↓ PromQL queries │
│ Traefik (TLS + basic auth) │
│ ↓ │
│ vm.vultrlabs.dev │
└─────────────────────────────────────────────────┘
Grafana queries both:
- M3DB (m3db.vultrlabs.dev) → real-time data (1h blocks, going forward)
- VictoriaMetrics (vm.vultrlabs.dev) → historical data (Mar 13present)
```
## Quick Start
### 1. Deploy VictoriaMetrics
```bash
# Apply manifests
kubectl apply -k .
# Wait for pod to be running
kubectl -n victoriametrics get pods -w
# Verify it's healthy
kubectl -n victoriametrics port-forward svc/victoriametrics 8428:8428 &
curl http://localhost:8428/health
```
### 2. Configure DNS
Get the Traefik LoadBalancer IP and point `vm.vultrlabs.dev` at it:
```bash
kubectl -n traefik get svc traefik
```
### 3. Set Up Basic Auth
Generate htpasswd and update the secret in `04-basic-auth-middleware.yaml`:
```bash
htpasswd -nb vultr_vm <your-password>
# Copy output, base64 encode it:
echo -n '<htpasswd-output>' | base64
# Update the secret and apply
kubectl apply -f 04-basic-auth-middleware.yaml
```
### 4. Run Backfill
```bash
# Create the secret with Mimir credentials
kubectl create secret generic backfill-credentials \
--from-literal=mimir-password='YOUR_MIMIR_PASSWORD' -n victoriametrics
# Upload the backfill script as a configmap
kubectl create configmap backfill-script \
--from-file=backfill.py=backfill.py -n victoriametrics
# Run the backfill pod
kubectl apply -f backfill-pod.yaml
# Watch progress
kubectl logs -f backfill -n victoriametrics
# Cleanup when done
kubectl delete pod backfill -n victoriametrics
kubectl delete configmap backfill-script -n victoriametrics
kubectl delete secret backfill-credentials -n victoriametrics
```
### 5. Verify
```bash
# In-cluster
kubectl -n victoriametrics exec deploy/victoriametrics -- \
curl -s 'http://localhost:8428/api/v1/query?query=vllm:prompt_tokens_total' | python3 -m json.tool
# External (with auth)
curl -u vultr_vm:<password> "https://vm.vultrlabs.dev/api/v1/query?query=up"
```
## Grafana Configuration
Add VictoriaMetrics as a **Prometheus** datasource:
- **URL:** `https://vm.vultrlabs.dev` (with basic auth)
- **In-cluster URL:** `http://victoriametrics.victoriametrics.svc.cluster.local:8428`
### Mixed Queries (M3DB + VictoriaMetrics)
Use a **Mixed** datasource in Grafana to query both:
1. Create two Prometheus datasources:
- `M3DB``https://m3db.vultrlabs.dev`
- `VictoriaMetrics``https://vm.vultrlabs.dev`
2. Create a **Mixed** datasource that includes both
3. In dashboards, use the mixed datasource — Grafana sends the query to both backends and merges results
Alternatively, use dashboard variables to let users toggle between datasources for different time ranges.
## Metrics Stored
| Metric | Description |
|--------|-------------|
| `vllm:prompt_tokens_total` | vLLM prompt token count |
| `vllm:generation_tokens_total` | vLLM generation token count |
| `DCGM_FI_DEV_GPU_UTIL` | GPU utilization (DCGM) |
All metrics are tagged with `tenant=serverless-inference-cluster` and `cluster=serverless-inference-cluster`.
## VictoriaMetrics API Reference
| Endpoint | Purpose |
|----------|---------|
| `/api/v1/import` | Import data (Prometheus format) |
| `/api/v1/export` | Export data |
| `/api/v1/query` | PromQL instant query |
| `/api/v1/query_range` | PromQL range query |
| /health | Health check |
| /metrics | Internal metrics |
## Storage
- **Size:** 200Gi NVMe (Vultr Block Storage)
- **StorageClass:** `vultr-block-storage-vm` (Retain policy — data survives PVC deletion)
- **Retention:** 2 years
- **Volume expansion:** `kubectl edit pvc victoriametrics-data -n victoriametrics`
## Useful Commands
```bash
# Check VM health
kubectl -n victoriametrics exec deploy/victoriametrics -- curl -s http://localhost:8428/health
# Check storage stats
kubectl -n victoriametrics exec deploy/victoriametrics -- \
curl -s 'http://localhost:8428/api/v1/query?query=vm_rows' | python3 -m json.tool
# Query historical data
curl -u vultr_vm:<password> \
"https://vm.vultrlabs.dev/api/v1/query_range?query=vllm:prompt_tokens_total&start=1773360000&end=1742000000&step=60"
# Restart VM (if needed)
kubectl rollout restart deployment/victoriametrics -n victoriametrics
# Scale to 0 (preserve data, stop the pod)
kubectl scale deployment/victoriametrics --replicas=0 -n victoriametrics
```
## Re-running Backfill
If you need to import additional time ranges or new metrics:
1. Edit `backfill.py` — update `START_TS`, `END_TS`, or `METRICS`
2. Recreate the configmap and pod (see step 4 above)
3. VictoriaMetrics is idempotent for imports — duplicate data points are merged, not duplicated
To convert timestamps:
```bash
# Date → Unix timestamp
date -u -d '2026-03-13 00:00:00' +%s # 1773360000
# Unix timestamp → date
date -u -d @1773360000
```

View File

@@ -0,0 +1,46 @@
##############################################################################
# Backfill Pod — One-shot job to import historical metrics from Mimir
#
# Usage:
# kubectl create configmap backfill-script \
# --from-file=backfill.py=backfill.py -n victoriametrics
# kubectl apply -f backfill-pod.yaml
# kubectl logs -f backfill -n victoriametrics
#
# Cleanup:
# kubectl delete pod backfill -n victoriametrics
# kubectl delete configmap backfill-script -n victoriametrics
##############################################################################
apiVersion: v1
kind: Pod
metadata:
name: backfill
namespace: victoriametrics
spec:
restartPolicy: Never
containers:
- name: backfill
image: python:3.12-slim
command: ["python3", "/scripts/backfill.py"]
env:
- name: MIMIR_USERNAME
value: "vultr_sea_inference"
- name: MIMIR_PASSWORD
valueFrom:
secretKeyRef:
name: backfill-credentials
key: mimir-password
- name: VM_URL
value: "http://victoriametrics.victoriametrics.svc.cluster.local:8428"
- name: START_TS
value: "1773360000" # 2026-03-13T00:00:00Z
- name: CHUNK_HOURS
value: "6"
volumeMounts:
- name: script
mountPath: /scripts
volumes:
- name: script
configMap:
name: backfill-script

211
victoriametrics/backfill.py Normal file
View File

@@ -0,0 +1,211 @@
#!/usr/bin/env python3
"""
Backfill historical metrics from Mimir to VictoriaMetrics.
Uses VictoriaMetrics /api/v1/import endpoint which happily accepts
data with any timestamp — no bufferPast gates, no block size hacks.
Usage:
# Run in-cluster (as a pod, see backfill-pod.yaml)
python3 backfill.py
# Or locally with port-forward
kubectl port-forward -n victoriametrics svc/victoriametrics 8428:8428
VM_URL=http://localhost:8428 python3 backfill.py
"""
import urllib.request
import urllib.error
import urllib.parse
import json
import ssl
import os
import time
import base64
import sys
# ── Configuration ──────────────────────────────────────────────────
MIMIR_URL = os.environ.get("MIMIR_URL", "https://metrics.vultrlabs.com/prometheus")
MIMIR_USER = os.environ.get("MIMIR_USERNAME", "REPLACE_WITH_MIMIR_USERNAME")
MIMIR_PASS = os.environ.get("MIMIR_PASSWORD", "REPLACE_WITH_MIMIR_PASSWORD")
VM_URL = os.environ.get("VM_URL", "http://victoriametrics.victoriametrics.svc.cluster.local:8428")
# Time range: March 13, 2026 00:00:00 UTC → now
START_TS = int(os.environ.get("START_TS", "1773360000")) # 2026-03-13T00:00:00Z
END_TS = int(os.environ.get("END_TS", str(int(time.time()))))
STEP = os.environ.get("STEP", "10s")
CHUNK_HOURS = int(os.environ.get("CHUNK_HOURS", "6"))
# Metrics to backfill
METRICS = [
"vllm:prompt_tokens_total",
"vllm:generation_tokens_total",
"DCGM_FI_DEV_GPU_UTIL",
]
# Extra labels to add to all imported data (e.g. tenant/cluster context)
EXTRA_LABELS = {
"tenant": "serverless-inference-cluster",
"cluster": "serverless-inference-cluster",
}
# ── Helpers ────────────────────────────────────────────────────────
def ssl_ctx():
ctx = ssl.create_default_context()
ctx.check_hostname = False
ctx.verify_mode = ssl.CERT_NONE
return ctx
def mimir_query(path):
"""Query Mimir API with basic auth."""
auth = base64.b64encode(f"{MIMIR_USER}:{MIMIR_PASS}".encode()).decode()
req = urllib.request.Request(f"{MIMIR_URL}{path}")
req.add_header("Authorization", f"Basic {auth}")
resp = urllib.request.urlopen(req, context=ssl_ctx(), timeout=300)
return json.loads(resp.read().decode())
def vm_import(lines):
"""Push data to VictoriaMetrics /api/v1/import."""
data = "\n".join(lines).encode("utf-8")
req = urllib.request.Request(
f"{VM_URL}/api/v1/import",
data=data,
method="POST",
)
req.add_header("Content-Type", "application/octet-stream")
try:
resp = urllib.request.urlopen(req, timeout=300)
return True
except urllib.error.HTTPError as e:
body = e.read().decode()[:200]
print(f" VM import ERROR {e.code}: {body}", flush=True)
return False
def format_prom_metric_name(raw_name):
"""Convert Mimir metric name to valid Prometheus metric name for VM.
VictoriaMetrics import format uses: metric_name{label1="val1",...} timestamp value
Colons in metric names are valid in Prometheus but we keep them as-is since
VM handles them fine.
"""
return raw_name
# ── Main ───────────────────────────────────────────────────────────
print(f"VictoriaMetrics Backfill", flush=True)
print(f"========================", flush=True)
print(f"Source: {MIMIR_URL}", flush=True)
print(f"Target: {VM_URL}", flush=True)
print(f"Range: {START_TS}{END_TS} ({CHUNK_HOURS}h chunks)", flush=True)
print(f"Metrics: {', '.join(METRICS)}", flush=True)
print(f"Extra labels: {EXTRA_LABELS}", flush=True)
print(flush=True)
total_samples = 0
total_errors = 0
for metric in METRICS:
print(f"\n{'='*60}", flush=True)
print(f"Metric: {metric}", flush=True)
print(f"{'='*60}", flush=True)
metric_samples = 0
chunk_start = START_TS
while chunk_start < END_TS:
chunk_end = min(chunk_start + CHUNK_HOURS * 3600, END_TS)
chunk_label = f"[{time.strftime('%Y-%m-%d %H:%M', time.gmtime(chunk_start))}{time.strftime('%Y-%m-%d %H:%M', time.gmtime(chunk_end))}]"
print(f" {chunk_label} ...", end="", flush=True)
try:
path = (
f"/api/v1/query_range?"
f"query={urllib.parse.quote(metric)}"
f"&start={chunk_start}&end={chunk_end}&step={STEP}"
)
data = mimir_query(path)
if data.get("status") != "success":
print(f" Mimir returned status={data.get('status')}", flush=True)
chunk_start = chunk_end
continue
series_list = data["data"]["result"]
if not series_list:
print(f" no data", flush=True)
chunk_start = chunk_end
continue
# Build import lines in VictoriaMetrics native format
# Format: metric_name{label1="val1",label2="val2"} timestamp value
import_lines = []
chunk_count = 0
for series in series_list:
labels = dict(series["metric"])
# Remove __name__ from labels (it's the metric name)
metric_name = labels.pop("__name__", metric)
# Add extra labels
labels.update(EXTRA_LABELS)
# Build label string
label_parts = [f'{k}="{v}"' for k, v in sorted(labels.items())]
label_str = ",".join(label_parts)
# Build import lines: one per sample
for ts_str, val_str in series["values"]:
# Convert timestamp (seconds) to ms for VM
ts_ms = int(float(ts_str) * 1000)
try:
val = float(val_str)
except (ValueError, TypeError):
# Handle +Inf, -Inf, NaN
if val_str == "+Inf":
val = float("inf")
elif val_str == "-Inf":
val = float("-inf")
else:
continue
import_lines.append(f'{metric_name}{{{label_str}}} {ts_ms} {val_str}')
chunk_count += 1
if import_lines:
ok = vm_import(import_lines)
if ok:
print(f" {chunk_count} samples imported", flush=True)
metric_samples += chunk_count
else:
print(f" IMPORT FAILED ({chunk_count} samples lost)", flush=True)
total_errors += chunk_count
else:
print(f" 0 samples", flush=True)
except Exception as e:
print(f" ERROR: {e}", flush=True)
total_errors += 1
chunk_start = chunk_end
print(f" Total for {metric}: {metric_samples} samples", flush=True)
total_samples += metric_samples
print(f"\n{'='*60}", flush=True)
print(f"BACKFILL COMPLETE", flush=True)
print(f"Total samples imported: {total_samples}", flush=True)
print(f"Total errors: {total_errors}", flush=True)
print(f"{'='*60}", flush=True)
# Verify by querying VM
print(f"\nVerifying import...", flush=True)
try:
verify_path = f"/api/v1/query?query={urllib.parse.quote('count(up)')}"
req = urllib.request.Request(f"{VM_URL}{verify_path}")
resp = urllib.request.urlopen(req, timeout=30)
print(f"VM is responding to queries ✓", flush=True)
except Exception as e:
print(f"VM query check failed: {e}", flush=True)

View File

@@ -0,0 +1,9 @@
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
resources:
- 00-namespace.yaml
- 01-storageclass.yaml
- 02-deployment.yaml
- 03-ingressroute.yaml
- 04-basic-auth-middleware.yaml