Alerting
4 minute read
The United Manufacturing Hub utilizes a TimescaleDB database, which is based on PostgreSQL. Therefore, you can use the PostgreSQL plugin in Grafana to implement and configure alerts and notifications.
Why should I use it?
Alerts based on real-time data enable proactive problem detection. For example, you will receive a notification if the temperature of machine oil or an electrical component of a production line exceeds limitations. By utilizing such alerts, you can schedule maintenance, enhance efficiency, and reduce downtime in your factories.
What can I do with it?
Grafana alerts help you keep an eye on your production and manufacturing processes. By setting up alerts, you can quickly identify problems, ensuring smooth operations and high-quality products. An example of using alerts is the tracking of the temperature of an industrial oven. If the temperature goes too high or too low, you will get an alert, and the responsible team can take action before any damage occurs. Alerts can be configured in many different ways, for example, to set off an alarm if a maximum is reached once or if it exceeds a limit when averaged over a time period. It is also possible to include several values to create an alert, for example if a temperature surpasses a limit and/or the concentration of a component is too low. Notifications can be sent simultaneously across many services like Discord, Mail, Slack, Webhook, Telegram, or Microsoft Teams. It is also possible to forward the alert with SMS over a personal Webhook. A complete list can be found on the Grafana page about alerting.
How can I use it?
For a detailed tutorial on how to set up an alert, please visit our learn page with the detailed step-by-step tutorial. Here you can find an overview of the process.
Install the PostgreSQL plugin in Grafana: Before you can formulate alerts, you need to install the PostgreSQL plugin, which is already integrated into Grafana.
Alert Rule: When creating an alert, you first have to set the alert rule in Grafana. Here you set a name, specify which values are used for the rule, and when the rule is fired. Additionally, you can add labels for your rules, to link them to the correct contact points. You have to use SQL to select the desired values.
Contact Point: In a contact point you create a collection of addresses and services that should be notified in case of an alert. This could be a Discord channel or Slack for example. When a linked alert is triggered, everyone within the contact point receives a message. The messages can be preconfigured and are specific to every service or contact.
Notification Policies: In a notification policy, you establish the connection of a contact point with the desired alerts. This is done by adding the labels of the desired alerts and the contact point to the policy.
Mute Timing: In case you do not want to receive messages during a recurring time period, you can add a mute timing to Grafana. If added to the notification policy, no notifications will be sent out by the contact point. This could be times without shifts, like weekends or during regular maintenance.
Silence: You can also add silences for a specific time frame and labels, in case you only want to mute alerts once.
An alert is only sent out once after being triggered. For the next alert, it has to return to the normal state, so the data no longer violates the rule.
What are the limitations?
It can be complicated to select and manipulate the desired values to create the correct function for your application. Grafana cannot differentiate between data points of the same source. For example, you want to make a temperature threshold based on a single sensor. If your query selects the last three values and two of them are above the threshold, Grafana will fire two alerts which it cannot tell apart. This results in errors. You have to configure the rule to reduce the selected values to only one per source to avoid this. It can be complicated to create such a specific rule with this limitation, and it requires some testing.
Another thing to keep in mind is that the alerts can only work with data from the database. It also does not work with the machine status; these values only exist in a raw, unprocessed form in TimescaleDB and are not processed through an API like process values.