search

matusnovak / prometheus-zfs

Prometheus exporter for (some) ZFS statistics via zpool iostatus and zfs get space
The Unlicense
18 stars 4 forks source link

Pool size #6

Open martijndierckx opened 1 year ago

martijndierckx commented 1 year ago

My goal is to follow up on the available space of my pools in prometheus. I have available: zfsprom_alloc, zfsprom_size and zfsprom_free.However, these values are on physical pool level as reported by zpool list. And thus not very reliable in day-to-day usage.

I would like to follow up on the available and used (+calculated total) values as reported by zfs list -o space. Is this info available? If not, can it be made available?

matusnovak commented 1 year ago

Hi @martijndierckx

Do you mean listing space usage for datasets, not just for the pools?

Funny thing, I was trying to solve this at my work recently. As far I could have searched, there was not an easy way to get ZFS dataset space usage programmatically. And, none of the following applications support it: NetData, InfluxDB Telegraf ZFS plugin, CollectD with ZFS plufin, node_exporter, zpool_influxdb CLI.

What I ended up doing is using InfluxDB Telegraf with exec input type and calling a custom bash script that takes an output from zfs get -p -H all, transforms each line a little bit (removing whitespace, adding {}, etc.), and echoes such metrics in a Prometheus format.

Unfortunately this is not possible with the libzfs Python library used in this project. However, I could write something that calls the zfs list -o space via a shell from the Python code and exposes the metrics that way.

This could be added as zfsprom_list_xyz metrics with tag dataset=..., while the original ones such as zfsprom_size would stay as they are. I would prefer to keep the original metrics unchanged so it won't cause any issues for the existing users.

Right now I am busy with other things. I can come back to this in ~2 weeks and implement it. However, if you need this sooner then feel free to submit a PR :)

martijndierckx commented 1 year ago

Sounds like a plan! Other approach would be to ask to implement this on libzfs? I do find it weird however that this information is nowhere available. It can't possibly be that we're the first to think of this ...

matusnovak commented 11 months ago

Sorry for a late update. My life got busy.

I have been trying to include the output of zfs list -o space, but I am no longer able to successfully built the Docker image. It seems that the recent version of some package (either the newer py-libzfs from TrueNAS or some Python/Cython package) broke the build.

I will not implement this feature.

I suggest that you should look into the zpool_influxdb CLI which should be located here: /usr/lib/zfs-linux/zpool_influxdb if you are using the latest ZFS debian package on the Ubuntu.

You can use the InfluxDB Telegraf with the exec plugin to read the metrics. Essentially, the plugin expects that the binary will print the metrics in InfluxDB format to the stdout, which is exactly what the zpool_influxdb does. It works out of the box. You can then use any of the output plugins to send the metrics forward, or even expose them as Prometheus metrics on some HTTP port.

The zpool_influxdb is far more useful and powerful than this prometheus-zfs project.