---
title: Vector
metaTitle: Collect Host Metrics with Vector and Prometheus Remote Write
description: Collect CPU, memory, filesystem, and network metrics with Vector and send them to OpenObserve using Prometheus remote write and Snappy compression.
---

# Collect host metrics with Vector

Vector's [`host_metrics` source](https://vector.dev/docs/reference/configuration/sources/host_metrics/)
collects CPU, memory, filesystem and network metrics. Send these through the
[`prometheus_remote_write` sink](https://vector.dev/docs/reference/configuration/sinks/prometheus_remote_write/)
to OpenObserve's metrics endpoint. Sending them to the logs `/_json` endpoint
stores them as logs instead. For log collection, see the [Vector logs guide](../logs/vector.md).

## Set up credentials

Install [Vector](https://vector.dev/docs/setup/installation/) on the host you
want to monitor, then set the following environment variables in Bash. Use the ingestion
URL, organization ID, user and passcode from **Data sources > Custom > Metrics** in your
OpenObserve instance. `OPENOBSERVE_URL` is the base URL without a trailing slash,
for example `http://localhost:5080` for a local instance. Use HTTPS for a remote
instance, and keep credentials out of configuration files and source control.

```bash
export OPENOBSERVE_URL="http://localhost:5080"
export OPENOBSERVE_ORG="default"
export OPENOBSERVE_USER="your-email@example.com"
read -rs -p "Ingestion passcode: " OPENOBSERVE_PASSCODE
export OPENOBSERVE_PASSCODE
export VECTOR_DANGEROUSLY_ALLOW_ENV_VAR_INTERPOLATION=true
```

Vector 0.57 and newer disable [environment variable interpolation](https://vector.dev/docs/reference/environment_variables/)
by default. The last line enables it for this example. Use only environment
values you control, and consider Vector's [secrets backend](https://vector.dev/docs/reference/configuration/secrets/)
for credentials on shared hosts.

## Configure and run Vector

Save the following as `host-metrics.yaml`:

```yaml
sources:
  host_metrics:
    type: host_metrics
    scrape_interval_secs: 60

sinks:
  openobserve_metrics:
    type: prometheus_remote_write
    inputs: [host_metrics]
    endpoint: "${OPENOBSERVE_URL}/api/${OPENOBSERVE_ORG}/prometheus/api/v1/write"
    auth:
      strategy: basic
      user: "${OPENOBSERVE_USER}"
      password: "${OPENOBSERVE_PASSCODE}"
    compression: snappy
    healthcheck:
      enabled: false
```

From the directory where you saved the file, validate and run it:

```bash
vector validate --no-environment host-metrics.yaml
vector --config host-metrics.yaml
```

## Verify ingestion

The example collects host metrics every 60 seconds. After the first collection,
open **Metrics** in the same organization and query `host_cpu_seconds_total`.
The available metrics depend on the host operating system and enabled collectors.
When Vector runs in a container, its host metrics describe the resources visible
inside that container. Run it on the host to monitor that host directly.

## Troubleshooting

The configuration uses Snappy compression, as required by the
[Prometheus remote-write protocol](https://prometheus.io/docs/specs/prw/remote_write_spec/).
Do not copy `compression: gzip` or the `http` sink from a Vector logs example.
The metrics endpoint is `/api/<organization>/prometheus/api/v1/write`, and the
sink type must be `prometheus_remote_write`.

The sink's startup health check is disabled, and `--no-environment` skips
component checks, so validation alone does not prove that metrics were ingested. If Vector reports HTTP 400, check OpenObserve's server
logs for the rejected payload. For HTTP 401 or 403, check the selected
organization, ingestion user, passcode and permissions.