Anomaly Detection and Alerting Setup


vmanomaly typical setup diagram

Note: Configurations used throughout this guide can be found here

Note: Starting from v1.13.0 node-exporter observability preset is available for vmanomaly. Please find the guide here.

1. What is vmanomaly?#

VictoriaMetrics Anomaly Detection (vmanomaly) is a service that continuously scans time series stored in VictoriaMetrics and detects unexpected changes within data patterns in real-time. It does so by utilizing user-configurable machine learning models.

All the service parameters are defined in a config file.

Note: Starting from 1.10.0, each vmanomaly configuration file can support more that one model type. To utilize different models on your data, it is no longer necessary to run multiple instances of the vmanomaly process. Please refer to model config section for more details.

Note: Starting from 1.11.0, each vmanomaly configuration file can support more that one model type, each attached to one (or more) schedulers. To utilize different models with different schedulers on your data, it is no longer necessary to run multiple instances of the vmanomaly process. Please refer to model and scheduler config sections for more details.

vmanomaly does the following:

  • periodically queries user-specified metrics
  • computes an anomaly score for them
  • pushes back the computed anomaly score to VictoriaMetrics.

What is anomaly score?#

Anomaly score is a calculated non-negative (in interval [0, +inf)) numeric value. It takes into account how well data fit a predicted distribution, periodical patterns, trends, seasonality, etc.

The value is designed to:

  • fall between 0 and 1 if model consider that datapoint is following usual pattern
  • exceed 1 if the datapoint is abnormal

Then, users can enable alerting rules based on the anomaly score with vmalert.

2. What is vmalert?#

vmalert is an alerting tool for VictoriaMetrics. It executes a list of the given alerting or recording rules against configured -datasource.url.

Alerting rules allow you to define conditions that, when met, will notify the user. The alerting condition is defined in a form of a query expression via MetricsQL query language. For example, in our case, the expression anomaly_score > 1.0 will notify a user when the calculated anomaly score exceeds a threshold of 1.0.

3. How does vmanomaly works with vmalert?#

Compared to classical alerting rules, anomaly detection is more “hands-off” and data-aware. Instead of thinking of critical conditions to define, user can rely on catching anomalies that were not expected to happen. In other words, by setting up alerting rules, a user must know what to look for, ahead of time, while anomaly detection looks for any deviations from past behavior.

Practical use case is to put anomaly score generated by vmanomaly into alerting rules with some threshold.

In this tutorial we are going to:

4. Data to analyze#

Let’s talk about data used for anomaly detection in this tutorial. We are going to collect our own CPU usage data with Node Exporter into the VictoriaMetrics database.

On a Node Exporter’s metrics page, part of the output looks like this:

# HELP node_cpu_seconds_total Seconds the CPUs spent in each mode.
# TYPE node_cpu_seconds_total counter
node_cpu_seconds_total{cpu="0",mode="idle"} 94965.14
node_cpu_seconds_total{cpu="0",mode="iowait"} 51.25
node_cpu_seconds_total{cpu="0",mode="irq"} 0
node_cpu_seconds_total{cpu="0",mode="nice"} 0
node_cpu_seconds_total{cpu="0",mode="softirq"} 1682.18
node_cpu_seconds_total{cpu="0",mode="steal"} 0
node_cpu_seconds_total{cpu="0",mode="system"} 995.37
node_cpu_seconds_total{cpu="0",mode="user"} 12378.05
node_cpu_seconds_total{cpu="1",mode="idle"} 94386.53
node_cpu_seconds_total{cpu="1",mode="iowait"} 51.22

In this context, the metric node_cpu_seconds_total provides a comprehensive breakdown of the time each CPU core has spent in various operational modes. These modes include: user, system, iowait, idle, irq&softirq, guest, and steal. Each of these eight modes is mutually exclusive, offering distinct insights into CPU activity. For instance, a predominant iowait suggests disk or network bottlenecks, while elevated levels in user or system indicate significant CPU utilization.

The node_cpu_seconds_total metric is classified as a counter type. To analyze the duration each CPU core spends in these modes, it is necessary to compute the rate of change per second using the rate function: rate(node_cpu_seconds_total). For a more refined and smoother aggregation of data by mode, we apply the sum function. The resulting query is formulated as follows: sum(rate(node_cpu_seconds_total[5m])) by (mode, instance, job).

Below is an illustrative example of how this query might be visualized in Grafana: node_cpu_rate_graph

This query will yield a total of eight time series, each corresponding to a CPU mode. The number of series is unaffected by the number of CPU cores, due to the by aggregation applied. These series serve as the input for vmanomaly, where the service independently fits a separate instance of the configured model type to each of time series.

5. vmanomaly configuration and parameter description#

Parameter description: The configuration file for vmanomaly comprises 4 essential sections:

  1. scheduler - This section determines the frequency of model inferences and training, including the time range for model training. Starting from v1.11.0, multiple individual schedulers are supported for each model type in a single config.

  2. models - Here, you define specific parameters and configurations for the models being used for anomaly detection. Starting from v1.10.0, multiple model configurations are supported in a single config.

  3. reader - This section outlines the methodology for data reading, including the data source location.

  4. writer - Specifies the destination and method for writing the generated output.

  5. monitoring (optional) - Describes how to monitor and expose health check metrics of vmanomaly.

Detailed parameters in each section:

  • schedulers (PeriodicScheduler is used here)

    • infer_every - Specifies the frequency at which the trained models perform inferences on new data, essentially determining how often new anomaly score data points are generated. Format examples: 30s, 4m, 2h, 1d (time units: ’s’ for seconds, ’m’ for minutes, ‘h’ for hours, ’d’ for days). This parameter essentially asks, at regular intervals (e.g., every 1 minute), whether the latest data points appear abnormal based on historical data.
    • fit_every - Sets the frequency for retraining the models. A higher frequency ensures more updated models but requires more CPU resources. If omitted, models are retrained in each infer_every cycle. Format is similar to infer_every.
    • fit_window - Defines the data interval for training the models. Longer intervals allow for capturing extensive historical behavior and better seasonal pattern detection but may slow down the model’s response to permanent metric changes and increase resource consumption. A minimum of two full seasonal cycles is recommended. Example format: 3h for three hours of data.
  • models

    • class - Specifies the model to be used. Options include custom models (guide here) or a selection from built-in models, such as the Facebook Prophet (model.prophet.ProphetModel).
    • args - Model-specific parameters, formatted as a YAML dictionary in the key: value structure. Parameters available in FB Prophet can be used as an example.
  • reader

    • datasource_url - The URL for the data source, typically an HTTP endpoint serving /api/v1/query_range.
    • queries - MetricsQL (an extension of PromQL) expressions for anomaly detection. Multiple queries can be formatted as <QUERY_ALIAS>: "QUERY", where QUERY_ALIAS is used as a label in the generated metrics and anomaly scores.
  • writer

    • datasource_url - Specifies the endpoint for output data, typically an HTTP endpoint serving /api/v1/import.

Below is an illustrative example of a vmanomaly_config.yml configuration file.

    # class: 'periodic'  # or "scheduler.periodic.PeriodicScheduler" until v1.13.0
    infer_every: "1m"
    fit_every: "2m"
    fit_window: "3h"

    class: "prophet"  # or "model.prophet.ProphetModel" until v1.13.0
      interval_width: 0.98

  datasource_url: "http://victoriametrics:8428/"
  sampling_period: "60s" 
    node_cpu_rate: "sum(rate(node_cpu_seconds_total[5m])) by (mode, instance, job)"

  datasource_url: "http://victoriametrics:8428/"

  pull: # Enable /metrics endpoint.
    addr: ""
    port: 8490

6. vmanomaly output#

As the result of running vmanomaly, it produces the following metrics:

  • anomaly_score - the main one. Ideally, if it is between 0.0 and 1.0 it is considered to be a non-anomalous value. If it is greater than 1.0, it is considered an anomaly (but you can reconfigure that in alerting config, of course),
  • yhat - predicted expected value,
  • yhat_lower - predicted lower boundary,
  • yhat_upper - predicted upper boundary,
  • y - initial query result value.

Here is an example of how output metric will be written into VictoriaMetrics: anomaly_score{for="node_cpu_rate", instance="node-exporter:9100", job="node-exporter", mode="idle"} 0.85

7. vmalert configuration#

Here we provide an example of the config for vmalert vmalert_config.yml.

- name: AnomalyExample
  - alert: HighAnomalyScore
    expr: 'anomaly_score > 1.0'
      severity: warning
      summary: Anomaly Score exceeded 1.0. `sum(rate(node_cpu_seconds_total))` is showing abnormal behavior. 

In the query expression expr, it’s crucial to establish a criterion based on the generated anomaly scores. Typically, an anomaly score ranging from 0.0 to 1.0 indicates that the analyzed value falls within normal behavior. Scores exceeding 1.0 signal increasing confidence from our model that the observed value is anomalous.

Selecting an appropriate threshold for the anomaly score depends on your specific requirements and the context of the data. One effective method for determining this threshold is through visual analysis. By plotting the anomaly_score metric in conjunction with the predicted ’expected’ range, delineated by yhat_lower and yhat_upper, you can make a more informed decision. Later in this tutorial, we will demonstrate this process with a practical example.

8. Docker Compose configuration#

You can find the docker-compose.yml and all configs in this folder

Now we are going to configure the docker-compose.yml file to run all needed services. Here are all services we are going to run:

  • vmanomaly - VictoriaMetrics Anomaly Detection service.
  • victoriametrics - VictoriaMetrics Time Series Database
  • vmagent - is an agent which helps you collect metrics from various sources, relabel and filter the collected metrics and store them in VictoriaMetrics or any other storage systems via Prometheus remote_write protocol.
  • grafana - visualization tool.
  • node-exporter - Prometheus Node Exporter exposes a wide variety of hardware- and kernel-related metrics.
  • vmalert - VictoriaMetrics Alerting service.
  • alertmanager - Notification services that handles alerts from vmalert.

Grafana setup#

Create a data source manifest#

In the provisioning/datasources/ directory, create a file called datasource.yml with the following content:

The default username/password pair is admin:admin

apiVersion: 1

    - name: VictoriaMetrics
      type: prometheus
      access: proxy
      url: http://victoriametrics:8428
      isDefault: true
        prometheusType: Prometheus
        prometheusVersion: 2.24.0

Define a dashboard provider#

In the provisioning/dashboards/ directory, create a file called dashboard.yml with the following content:

apiVersion: 1

- name: Prometheus
  orgId: 1
  folder: ''
  type: file
    path: /var/lib/grafana/dashboards

Scrape config#

Let’s create prometheus.yml file for vmagent configuration.

  scrape_interval: 10s

  - job_name: 'vmagent'
      - targets: ['vmagent:8429']
  - job_name: 'vmalert'
      - targets: ['vmalert:8880']
  - job_name: 'victoriametrics'
      - targets: ['victoriametrics:8428']
  - job_name: 'node-exporter'
      - targets: ['node-exporter:9100']
  - job_name: 'vmanomaly'
      - targets: [ 'vmanomaly:8490' ]

vmanomaly licensing#

We will utilize the license key stored locally in the file vmanomaly_license.

For additional licensing options, please refer to the VictoriaMetrics Anomaly Detection documentation on licensing.

Alertmanager setup#

Let’s create alertmanager.yml file for alertmanager configuration.

  receiver: blackhole

  - name: blackhole


Let’s wrap it all up together into the docker-compose.yml file.

    container_name: vmagent
    image: victoriametrics/vmagent:v1.100.1
      - "victoriametrics"
      - 8429:8429
      - vmagentdata-guide-vmanomaly-vmalert:/vmagentdata
      - ./prometheus.yml:/etc/prometheus/prometheus.yml
      - "--promscrape.config=/etc/prometheus/prometheus.yml"
      - "--remoteWrite.url=http://victoriametrics:8428/api/v1/write"
      - vm_net
    restart: always

    container_name: victoriametrics
    image: victoriametrics/victoria-metrics:v1.100.1
      - 8428:8428
      - vmdata-guide-vmanomaly-vmalert:/storage
      - "--storageDataPath=/storage"
      - "--httpListenAddr=:8428"
      - "--vmalert.proxyURL=http://vmalert:8880"
      - "-search.disableCache=1" # for guide only, do not use in production
      - vm_net
    restart: always

    container_name: grafana
    image: grafana/grafana-oss:10.2.1
      - "victoriametrics"
      - 3000:3000
      - grafanadata-guide-vmanomaly-vmalert:/var/lib/grafana
      - ./provisioning/datasources:/etc/grafana/provisioning/datasources
      - ./provisioning/dashboards:/etc/grafana/provisioning/dashboards
      - ./vmanomaly_guide_dashboard.json:/var/lib/grafana/dashboards/vmanomaly_guide_dashboard.json
      - vm_net
    restart: always

    container_name: vmalert
    image: victoriametrics/vmalert:v1.100.1
      - "victoriametrics"
      - 8880:8880
      - ./vmalert_config.yml:/etc/alerts/alerts.yml
      - "--datasource.url=http://victoriametrics:8428/"
      - "--remoteRead.url=http://victoriametrics:8428/"
      - "--remoteWrite.url=http://victoriametrics:8428/"
      - "--notifier.url=http://alertmanager:9093/"
      - "--rule=/etc/alerts/*.yml"
      # display source of alerts in grafana
      - "--external.url=" #grafana outside container
      # when copypaste the line be aware of '$$' for escaping in '$expr'
      - '--external.alert.source=explore?orgId=1&left=["now-1h","now","VictoriaMetrics",{"expr": },{"mode":"Metrics"},{"ui":[true,true,true,"none"]}]'
      - vm_net
    restart: always
    container_name: vmanomaly
    image: victoriametrics/vmanomaly:v1.11.0
      - "victoriametrics"
      - "8490:8490"
      - vm_net
    restart: always
      - ./vmanomaly_config.yml:/config.yaml
      - ./vmanomaly_license:/license
    platform: "linux/amd64"
      - "/config.yaml"
      - "--license-file=/license"
    container_name: alertmanager
    image: prom/alertmanager:v0.25.0
      - ./alertmanager.yml:/config/alertmanager.yml
      - "--config.file=/config/alertmanager.yml"
      - 9093:9093
      - vm_net
    restart: always

    container_name: node-exporter
      - 9100:9100
    pid: host
    restart: unless-stopped
      - vm_net

  vmagentdata-guide-vmanomaly-vmalert: {}
  vmdata-guide-vmanomaly-vmalert: {}
  grafanadata-guide-vmanomaly-vmalert: {}

Before running our docker-compose make sure that your directory contains all required files:

all files

This docker-compose file will pull docker images, set up each service and run them all together with the command:

docker-compose up -d

To check if vmanomaly is up and running you can check docker logs:

docker logs vmanomaly -f

9. Model results#

To look at model results we need to go to grafana on the localhost:3000. Data vmanomaly need some time to generate more data to visualize. Let’s investigate model output visualization in Grafana. On the Grafana Dashboard Vmanomaly Guide for each mode of CPU you can investigate:

  • initial query result - sum(rate(node_cpu_seconds_total[5m])) by (mode, instance, job)
  • anomaly_score
  • yhat - Predicted value
  • yhat_lower - Predicted lower boundary
  • yhat_upper - Predicted upper boundary

Each of these metrics will contain same labels our query sum(rate(node_cpu_seconds_total[5m])) by (mode, instance, job) returns.

Anomaly scores for each metric with its according labels.#

Query: anomaly_score Anomaly score graph

Check out if the anomaly score is high for datapoints you think are anomalies. If not, you can try other parameters in the config file or try other model type.

As you may notice a lot of data shows anomaly score greater than 1. It is expected as we just started to scrape and store data and there are not enough datapoints to train on. Just wait for some more time for gathering more data to see how well this particular model can find anomalies. In our configs we put 2 weeks of data needed to fit the model properly.

Lower and upper boundaries and predicted values.#

Queries: yhat_lower, yhat_upper and yhat

yhat lower and yhat upper

Boundaries of ’normal’ metric values according to model inference.


On the page http://localhost:8880/vmalert/groups you can find our configured Alerting rule:

alert rule

According to the rule configured for vmalert we will see Alert when anomaly score exceed 1. You will see an alert on Alert tab. http://localhost:8880/vmalert/alerts: alerts firing

10. Conclusion#

We’ve explored the integration and practical application of VictoriaMetrics Anomaly Detection (vmanomaly) in conjunction with vmalert. This tutorial has taken you through the necessary prerequisites, setup, and configurations required for anomaly detection in time series data.

Key takeaways include:

  1. Understanding vmanomaly and vmalert: We’ve discussed the functionalities of vmanomaly and vmalert, highlighting how they work individually and in tandem to detect anomalies in time series data.

  2. Practical Configuration and Setup: By walking through the setup of a docker-compose environment, we’ve demonstrated how to configure and run VictoriaMetrics along with its associated services, including vmanomaly and vmalert.

  3. Data Analysis and Monitoring: The guide provided insights on how to collect, analyze, and visualize data using Grafana, interpreting the anomaly scores and other metrics generated by vmanomaly.

  4. Alert Configuration: We’ve shown how to set up and customize alerting rules in vmalert based on produced anomaly scores, enabling proactive monitoring and timely response to potential issues.

As you continue to use VictoriaMetrics Anomaly Detection and vmalert, remember that the effectiveness of anomaly detection largely depends on the appropriateness of the model chosen, the accuracy of configurations and the data patterns observed. This guide serves as a starting point, and we encourage you to experiment with different configurations and models to best suit your specific data needs and use cases. In case you need a helping hand - contact us.

Note: Starting from v1.13.0 node-exporter observability preset is available for vmanomaly. Please find the guide here.