id:Songmu, the director of Mackerel’s development team, took to the stage at JTF2016, held this past Sunday, July 24th. He presented under the title "Mackerel technology. Where we’ve come from and where we’re going."
A big thanks to all of those who attended.
Documents from the speech have been published and can be found at the link below (unfortunately, the document is only in Japanese).
Here is this week’s update.
Monitoring can now be done based on future predictions
Linear regression has been added to the list of functions that can be used with expression monitoring.
With this, it is now possible to configure monitoring based on future predictions.
As an example, let’s try monitoring the number of days remaining until the file system’s free space is used up.
timeLeftForecast() to obtain the number of seconds before the linear regression value reaches the threshold.
Below is an example expression for monitoring.
scale(timeLeftForecast(host('host_id', 'filesystem.drive.used'), '3mo', 2000000000000), 1/86400)
The expression above corresponds to "the value calculated with linear regression using three months of the
filesystem.drive.used metric, to predict the days remaining until filesystem reaches 2TB."
Let’s take the expression apart and explain it one by one.
First of all,
host('host_id', 'filesystem.drive.used') is used to obtain the
filesystem.drive.used metric value for the specified host.
timeLeftForecast() to this. The parameters of
timeLeftForecast are "metrics", "duration", and "threshold", so the example above corresponds to the number of seconds until the value reaches 2000000000000byte(2TB), using linear regression with three months worth of data of the specified host's metric value.
Lastly, the number of seconds until the obtained file system becomes empty is converted into days using the
As shown above, expression monitoring lets users conduct monitoring in different ways using the obtained metrics.
For more information, refer to the help page below.
Expression monitoring is offered as an experimental feature.
To find out how to use experimental features, refer to the help page below.
Expressions can now be edited while viewing the graph
An expression editor has been added to two places: the expression monitoring configuration screen and the customized graph permalink.
The input expression is reflected in the graph at real time, making trial and error easier when writing an expression.
For more information on expression monitoring, please visit the help page mentioned above. For more on customized graphs, follow the link below.
Operations on graph ranges are now reflected in the URL
When adjusting a graph’s display range with the slider and such, the display range is now reflected in the URL.
This makes it easier to share graphs whose display range has been adjusted from the screen.
View-only users can now be specified during member invitation
This feature was released last week, but until now it was only possible to switch existing members.
With this update, you can now specify whether a member is view-only or not before sending the invitation.