Observability Engineering by Charity Majors

Observability Engineering by Charity Majors

Author:Charity Majors [Majors, Charity]
Language: eng
Format: epub
ISBN: 9781492076445
Publisher: O'Reilly
Published: 2022-05-13T00:00:00+00:00


Context-aware burn alerts

When using context-aware, or historical, burn alerts, you keep a rolling total of the number of good and bad events that have happened over the entire window of the SLO, rather than just the baseline window. This section unpacks the various calculations you need to make for effective burn alerts. But you should tread carefully when practicing these techniques. The computational expense of calculating these values at each evaluation interval can quickly become financially expensive as well. At Honeycomb, we found this out the hard way when small SLO data sets suddenly started to rack up over $5,000 of AWS Lambda costs per day. :-)

To see how the considerations are different for context-aware burn alerts, let’s work an example. Say you have a service with an SLO target indicating 99% of units will succeed over a moving 30-day window. In a typical month, the service sees 43,800 units. In the previous 26 days, you have already failed 285 units out of 37,960. In the past 24 hours, the service has seen 1,460 units, 130 of which failed. To achieve the 99% target, only 1% of units are allowed to fail per month. Based on typical traffic volume (43,800 units), only 438 units are allowed to fail (your error budget). You want to know if, at this rate, you will burn your error budget in the next four days.

In this example, you want to project forward on a scale of days. Using the maximum practical extrapolation factor of 4, as noted previously, you set a baseline window that examines the last one day’s worth of data to extrapolate forward four days from now.

You must also consider the impact that your chosen scale has on your sliding window. If your SLO is a sliding 30-day window, your adjusted lookback window would be 26 days: 26 lookback days + 4 extrapolated days = your 30-day sliding window, as shown in Figure 13-6.



Download



Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.