Option to disable disk metrics from collectd/signalfx-metadata about signalfx-agent HOT 6 CLOSED

@jalaziz do you mean the disk.utilization metric that collectd/signalfx-metadata emits? If you turn off the collectd/df plugin and replace it with the filesystems plugin, the collectd/signalfx-metadata monitor should no longer emit disk.utilization, but it would be emitted by the filesystems monitor only. One potentially confusing thing is that the disk.utilization metric in 4.x of the agent has a dimension plugin: signalfx-metadata, but that will be removed in the 5.0 release (pending release in the next few weeks) so that piece of cruft will go away then. See if that is what you are seeing, but if not, let me know.

from signalfx-agent.

@keitwb Sorry! Didn't get notified for some reason.

Indeed, I meant disk.utilization and disk.summary_utilization.

Thanks for the tip, didn't realize disabling collectd/df would also disable those metrics!

We don't use the plugin dimension, so that should be fine unless it is used to trigger something on the SignalFx UI?

I will test it out soon and post the results.

from signalfx-agent.

Ok cool. Anything in the UI that is used as a trigger will be updated before release. Let me know if you run into any issues.

from signalfx-agent.

@keitwb Replacing collectd/df with filesystems worked fine, however there were a couple "breaking" differences.

1. With collectd/df, mount paths are normalized. For example, a /var/lib mount would have plugin_instance:var-lib. With filesystems, we have plugin_instance:/var/lib. Arguably, filesystems is more correct, but it does break existing detectors.

2. With collectd/df, the root mount path has plugin_instance:root. With filesystems, we have plugin_instance: and mountpoint: (empty strings). Shouldn't they be /?

from signalfx-agent.

For 1. Yes, the filesystem monitor the change from - to the conventional / separator is going to be a breaking change. Also in the 5.0 release, filesystems will no longer emit a plugin_instance dimension at all, it will just have mountpoint, device, and fs_type dimensions.

For 2. this is a bug due to overly simplistic stripping of the hostfs mountpoint prefix that I have confirmed (I'm assuming you are running the agent in some kind of containerized environment and not straight on the host). It should be fixed in #1190 which will go out in 5.0 (might backport it to 4.x but not high priority since we are about to release 5.0).

I also added a guide for migrating in #1190. The easiest thing to do to transition detectors seamlessly without any false alerts or downtime is to run both monitors simultaneously, switch your detectors, then turn off the collectd/df monitor in your configs. You will still be able to run collectd/df (and collectd/signalfx-metadata) in 5.0 so you won't be forced to change them right away. You should not (need to confirm this) be charged for any extra usage when both monitors are active due to how the accounting works since the metric names are the same and the hostname would be the same.

from signalfx-agent.

Going to close this unless you have any other issues/concerns about this.

from signalfx-agent.