Skip to content

README.md

Latest commit

 

History

History
181 lines (124 loc) · 4.51 KB

README.md

File metadata and controls

181 lines (124 loc) · 4.51 KB
Raw

Local deployment startup guide

This directory contains the reproducible deployment files for the local Milvus Cluster and its monitoring stack.

deployments/
|-- milvus-cluster/   Milvus v2.6.21 distributed deployment
\-- monitoring/       Prometheus, Grafana, and cAdvisor

What should be running

The complete local environment contains:

Group Containers
Milvus Cluster milvus-cluster-etcd, milvus-cluster-minio, milvus-cluster-mixcoord, milvus-cluster-proxy, milvus-cluster-streamingnode, milvus-cluster-datanode, milvus-cluster-querynode
Monitoring milvus-prometheus, milvus-grafana, milvus-cadvisor
Management UI attu

The old milvus-standalone container is retained only as a backup of the previous environment. Keep it stopped while the Cluster is running because both deployments use ports 19530 and 9091.

Start after restarting Windows

First, start Docker Desktop and wait until its Docker Engine is ready. Then open PowerShell and run:

Set-Location D:\IdeaProjects\milvus-lab

.\deployments\milvus-cluster\start.ps1

docker start attu

docker compose -f .\deployments\monitoring\docker-compose.yml up -d

This order is recommended because:

  1. the Cluster startup script creates the shared milvus-lab Docker network;
  2. it starts etcd and MinIO before starting the Milvus components;
  3. Attu connects to the Cluster through host.docker.internal:19530;
  4. Prometheus starts after the Cluster and scrapes all Milvus components.

These commands are idempotent. It is safe to run them when the containers are already running.

Automatic restart behavior

The Milvus Cluster and monitoring containers use the unless-stopped restart policy. If Docker Desktop starts automatically with Windows, they will normally recover automatically.

The existing attu container does not have an automatic restart policy. Start it manually after a reboot:

docker start attu

Running the complete startup sequence is still recommended because it verifies the expected Compose configuration and waits for every Milvus Cluster container to become healthy.

Verify the environment

Check the Cluster and its resource usage:

.\deployments\milvus-cluster\status.ps1

All seven Cluster containers should show healthy.

Check the monitoring stack:

docker compose -f .\deployments\monitoring\docker-compose.yml ps

Expected monitoring containers:

milvus-prometheus
milvus-grafana
milvus-cadvisor

Check all currently running containers:

docker ps

Check the Milvus Proxy health endpoint:

Invoke-WebRequest -UseBasicParsing http://localhost:9091/healthz

An HTTP 200 response means the Proxy is healthy.

Local addresses

Service Address
Milvus SDK http://localhost:19530
Attu http://localhost:3000
Milvus WebUI http://localhost:9091/webui/
Jaeger Trace UI http://localhost:16686
MinIO Console http://localhost:9001
Prometheus http://localhost:9090
Prometheus targets http://localhost:9090/targets
Grafana http://localhost:3001
Milvus Cluster dashboard http://localhost:3001/d/milvus-cluster-local/milvus-cluster-local
cAdvisor http://localhost:8080

Milvus authentication is disabled in this local environment.

Stop the environment

Stop the Cluster while retaining all data:

.\deployments\milvus-cluster\stop.ps1

Stop monitoring while retaining Prometheus and Grafana data:

docker compose -f .\deployments\monitoring\docker-compose.yml down

Stop Attu:

docker stop attu

Do not add --volumes to either Compose down command unless the stored data should be permanently deleted.

Common startup problems

Port 19530 or 9091 is already in use

Confirm that the old Standalone container is stopped:

docker stop milvus-standalone
.\deployments\milvus-cluster\start.ps1

Attu is unavailable

docker start attu
docker logs attu --tail 50

Attu should connect to host.docker.internal:19530.

Prometheus shows Milvus targets as DOWN

Start the Cluster first, then verify its health:

.\deployments\milvus-cluster\start.ps1
.\deployments\milvus-cluster\status.ps1

Prometheus retries failed targets automatically; no metrics data needs to be deleted.

Detailed documentation