Improving an Azure Function that writes IoT Hub data to TimescaleDB

In an earlier post, I used an Azure Function to write data from IoT Hub to a TimescaleDB hypertable on PostgreSQL. Although that function works for demo purposes, there are several issues. Two of those issues will be addressed in this post:

  1. the INSERT INTO statement used the NOW() function instead of the enqueuedTimeUtc field; that field is provided by IoT Hub and represents the time the message was enqueued
  2. the INSERT INTO query does not use upsert functionality; if for some reason you need to process the IoT Hub data again, you will end up with duplicate data; you code should be idempotent

Using enqueuedTimeUtc

Using the time the event was enqueued means we need to retrieve that field from the message that our Azure Function receives. The Azure Function receives outside information via two parameters: context and eventHubMessage. The enqueuedTimeUtc field is retrieved via the context variable: context.bindingData.enqueuedTimeUtc.

In the INSERT INTO statement, we need to use TIMESTAMP ‘UCT time’. In JavaScript, that results in the following:

'insert into conditions(time, device, temperature, humidity) values(TIMESTAMP \'' + context.bindingData.enqueuedTimeUtc + '\',\'' + eventHubMessage.device + '\' ...

Using upsert functionality

Before adding upsert functionality, add a unique constraint to the hypertable like so (via pgAdmin):

CREATE UNIQUE INDEX on conditions (time, device); 

It needs to be on time and device because the time field on its own is not guaranteed to be unique. Now modify the INSERT INTO statement like so:

'insert into conditions(time, device, temperature, humidity) values(TIMESTAMP \'' + context.bindingData.enqueuedTimeUtc + '\',\'' + eventHubMessage.device + '\',' + eventHubMessage.temperature + ',' + eventHubMessage.humidity + ') ON CONFLICT DO NOTHING'; 

Notice the ON CONFLICT clause? When any constraint is violated, we do nothing. We do not add or modify data, we leave it all as it was.

The full Azure Function code is below:

Azure Function code with IoT Hub enqueuedTimeUtc and upsert

Conclusion

The above code is a little bit better already. We are not quite there yet but the two changes make sure that the date of the event is correct and independent from when the actual processing is done. By adding the constraint and upsert functionality, we make sure we do not end up with duplicate data when we reprocess data from IoT Hub.

Dashboard your TimescaleDB data with Grafana

In an earlier post, I looked at storing time-series data with TimescaleDB on Azure Database for PostgreSQL. To visualize your data, there are many options as listed here. Because TimescaleDB is built on PostgreSQL, you can use any tool that supports PostgreSQL such as Power BI or Tableau.

Grafana is a bit of a special case because TimescaleDB engineers actually built the data source, which is designed to take advantage of the time-series capabilities. For a detailed overview of the capabilities of the data source, see the Grafana documentation.

Let’s take a look at a simple example to get started. I have a hypertable called conditions with four columns: time, device, temperature, humidity. An IoT Simulator is constantly writing data for five devices: pg-1 to pg-5.

On a multi-tier deployment of Grafana, I added the PostgreSQL data source:

PostgreSQL data source in Grafana

One setting in the data source is particularly noteworthy:

TimescaleDB support in the PostgreSQL datasource

Grafana has the concept of macro’s such as $_timeGroup or $_interval, as noted in the preceding image. The macro is translated to what the underlying data source supports. In this case, with TimescaleDB enabled, the macro results in the use of time_bucket, which is specific for TimescaleDB.

Creating a dashboard

Create a dashboard from the main page:

Creating a new dashboard

You will get a new dashboard with an empty panel:

Click Add Query. You will notice Grafana proposes a query. In this case it is very close because we only have one data source and table:

Grafana proposes the following query

Let’s modify this a bit. In the top right corner, I switched the time interval to last 30 minutes. Because the default query uses WHERE Macro: $_timeFilter, only the last 30 minutes will be shown. That’s another example of a macro. I would like to show the average temperature over 10 second intervals. That is easy to do with a GROUP BY and $_interval. In GROUP BY, click the + and type or select time to use the time field. You will notice the following:

GROUP BY with $_interval

Just click $_interval and select 10s. Now add the humidity column to the SELECT statement:

Adding humidity

When you click the Generated SQL link, you will see the query built by the query builder:

Generated SQL

Notice that the query uses time_bucket. The GROUP BY 1 and ORDER BY 1 just means group and order on the first field which is the time_bucket. If the query builder is not sufficient, you can click Edit SQL and specify your query directly. When you switch back to query builder, your custom SQL statement might be overwritten if the builder does not support it.

When you save your dashboard, you should see something like:

Pretty boring temperature and humidity graphWi

Now, let’s add a few gauges. In the top right row of icons, the first one should be Add panel. Choose the Gauge visualization and set your query:

Temperature Gauge

In Visualization, set Stat to Current:

Stat field on current

When the panel is finished, navigate back to the dashboard and duplicate the gauge. Modify the duplicated gauge to show humidity. Also change the titles. The dashboard now looks like:

Conditions dashboard

Grafana can be configured to auto refresh the dashboard. In the image below, refresh was set to every 5 seconds:

Setting auto refresh

Your dashboard will now update every 5 seconds for a more dynamic experience.

Joins

You can join hypertables with regular tables quite easily. This is one of the advantages of using a relational database such as PostgreSQL for your time-series data. The screenshot below shows a graph of the temperature per device location. The device location is stored in a regular table.

Join between hypertable and regular table: they are all just tables in the end

Here is the full dashboard:

Conclusion

Grafana, in combination with PostgreSQL and TimescaleDB, is a flexible solution for dashboarding your IoT time-series data. We have only scratched the surface here but it’s clear you can be up and running fast! Give it a go and tell me what you think in the comments or via @geertbaeke!