LinuxCzar

Engineering Software, Linux, and Observability. The website of Jack Neely.    

Generate Test Prometheus TSDB Data

This tool generates Prometheus 2.x TSDB blocks containing test data. If you have ever wanted to test Prometheus or Thanos behavior on a specific time series foot print, say a year’s worth of data, then this tool is for you.

Find it on GitHub:

https://github.com/jjneely/stuff/tree/master/cmd/historygen

Use it:

$ ./historygen -h
Usage of ./historygen:
  -b duration
        TSDB block length (default 2h0m0s)
  -c int
        Number of time series to generate (default 1)
  -d duration
        Time duration of historical data to generate (default 720h0m0s)
  -i duration
        Duration between samples (default 15s)
  -o string
        Output directory to generate TSDB blocks in (default "data/")

Historygen will create the directory named by the -o option and populate it with TSDB blocks containing time series data. Each block produced will have the duration specified by -b (block length). The total amount of history to generate is controlled by -d for duration. So, for 720 hours of history (about a month) the tool would generate 360 TSDB blocks spanning the time range from the time the command was run until 720 hours ago.

All time series generated have the name test, a job="testdata" label, and an instance="test-metric-XXX" where XXX is a zero based integer that uniquely defines each time series as requested by the -c option. Using -c 500 would produce time series that look like:

test{instance="test-metric-000", job="testdata"}
test{instance="test-metric-001", job="testdata"}
...
test{instance="test-metric-499", job="testdata"}

The -i option defines the simulated scrape interval of the data. No jitter is added at this writing and the data points are spaced evenly apart at this duration. The default matches Prometheus’s default of a 15 second scrape interval. The data points stored in each time series is a nonosecond resolution timestamp of when the data was created. This simulates Prometheus’s Counters.

Why is this Stuff Called Stuff?

I’m out of names, and I needed a quick place to store some random bits of Go code that I found useful.

Where Did This Originally Come From?

The Thanos authors wrote some interesting bench mark tools that this idea and the tsdb package was originally lifted from. Its been modified to create the time series naming/labeling scheme as documented above, and to use nanosecond timestamps as the values for each stored data point.

How Do You Get and Build This?

With Go, of course. Golang 1.11 or better is required most likely.

$ go get -u github.com/jjneely/stuff/cmd/historygen
$ cd ~/go/src/github.com/jjneely/stuff/cmd/historygen
$ go build

Ok, I Generated a Pile of Data, What Next?

Once the data is generated, insert it into Prometheus. Ideally, the block length specified matches the uncompacted block lengths that the Prometheus instance is creating or is configured to create. Shut down Prometheus, copy the TSDB blocks into the data directory, start Prometheus. Note that TSDB blocks cannot overlap time ranges with other blocks (although Prometheus 2.8 does introduce some support for this). The safe thing to do is wipe Prometheus’s TSDB directory, copy in the generated data, then start Prometheus. If the Prometheus WAL is left intact, Prometheus will later cut a new TSDB block from the WAL that will overlap with the generated test data. Perhaps a Thanos Sidecar component is configured here to upload this data to GCS or S3 buckets.

You can also use gsutil (assuming Google Cloud Platform) and copy this data directly into a GCS bucket. Point a Thanos Store and Compact component at it and the test data will become available to your Thanos Query component. I’ve used this to test Thanos for long data histories.

 Previous  Up  Next


comments powered by Disqus