Downsample data with InfluxDB

One of the most common use cases for InfluxDB tasks is downsampling data to reduce the overall disk usage as data collects over time. In previous versions of InfluxDB, continuous queries filled this role.

This article walks through creating a continuous-query-like task that downsamples data by aggregating data within windows of time, then storing the aggregate value in a new bucket.

Requirements

To perform a downsampling task, you need to the following:

A “source” bucket

The bucket from which data is queried.

A “destination” bucket

A separate bucket where aggregated, downsampled data is stored.

Some type of aggregation

To downsample data, it must be aggregated in some way. What specific method of aggregation you use depends on your specific use case, but examples include mean, median, top, bottom, etc. View Flux’s aggregate functions for more information and ideas.

Example downsampling task script

The example task script below is a very basic form of data downsampling that does the following:

  1. Defines a task named “cq-mem-data-1w” that runs once a week.
  2. Defines a data variable that represents all data from the last 2 weeks in the mem measurement of the system-data bucket.
  3. Uses the aggregateWindow() function to window the data into 1 hour intervals and calculate the average of each interval.
  4. Stores the aggregated data in the system-data-downsampled bucket under the my-org organization.

    1. // Task Options
    2. option task = {
    3. name: "cq-mem-data-1w",
    4. every: 1w,
    5. }
    6. // Defines a data source
    7. data = from(bucket: "system-data")
    8. |> range(start: -duration(v: int(v: task.every) * 2))
    9. |> filter(fn: (r) => r._measurement == "mem")
    10. data
    11. // Windows and aggregates the data in to 1h averages
    12. |> aggregateWindow(fn: mean, every: 1h)
    13. // Stores the aggregated data in a new bucket
    14. |> to(bucket: "system-data-downsampled", org: "my-org")

Again, this is a very basic example, but it should provide you with a foundation to build more complex downsampling tasks.

Add your task

Once your task is ready, see Create a task for information about adding it to InfluxDB.

Things to consider

  • If there is a chance that data may arrive late, specify an offset in your task options long enough to account for late-data.
  • If running a task against a bucket with a finite retention policy, do not schedule tasks to run too closely to the end of the retention policy. Always provide a “cushion” for downsampling tasks to complete before the data is dropped by the retention policy.

tasks