Open codefaux opened 1 year ago
discordianfish
commented
1 year ago hum.. your mdstat looks entirely different from what we expect: https://github.com/prometheus/procfs/blob/b4a1860af088340784210567cf7d5be61ff0d12b/testdata/fixtures.ttar#L2255
Is this really a snippet of /proc/mdstat?
codefaux
commented
1 year ago Really, it actually is. I'm as surprised as you.
codefaux
commented
1 year ago Or at least I was... I've been informed that Unraid does not use the standard md module. I apologize; I only learned that literally just now.
There are many Unraid users, and many of them use this Prometheus Node Exporter.
I would LOVE to see support for Unraid's md module in the exporter so I can get proper metrics, but this is clearly not a "mainline use scenario" so I understand if the decision is made not to support it.
As of right now, the suggestion has been raised to simply disable the md collector via --no-collector.mdadm -- this is obviously a valid workaround since the exporter functions in its current state otherwise.
Just so I can get a concrete answer; is there any intent to (eventually, no target date asked) support Unraid's md module, or should we consider the workaround permanent?
discordianfish
commented
1 year ago Interesting! I mean I'm not opposed to support unraid if someone submits clear PR(s) for this:
codefaux
commented
1 year ago Roger. I'll consider tackling it myself, but my work isn't typically up to scale wirth what's going on here. We'll see how it goes, thanks for your time.
Close or leave open?
SuperQ
commented
1 year ago We can leave this open for tracking.
dswarbrick
commented
1 year ago Are the contents of /sys/block/md*/md closer to that of the traditional md module? There was an github issue floating around a few years ago about migrating the mdraid parser from /proc/mdstat to sysfs.
codefaux
commented
1 year ago I'm guessing no, since /sys/block/md*/md does not exist - unless I misunderstand.
dswarbrick
commented
1 year ago I guess it stands to reason that if they're not using the standard Linux mdraid module, the associated sysfs entries won't be there either.
It looks like you're not the first person to encounter this: https://forums.unraid.net/bug-reports/stable-releases/692-node-exporter-cant-parse-mdstat-to-get-disk-information-for-prometheus-r1638/
Host operating system:
FWIW it's Unraid 6.11.x series
node_exporter version:
node_exporter command line flags
None
node_exporter log output
Snipping all except this relevant line -- pretty sure the rest isn't significant given it seems to be a parsing-related issue with proc output of my system. Please correct me if I'm wrong and I'll paste the entire output.
Are you running node_exporter in Docker?
Nope, direct from terminal as foreground process for troubleshooting purposes
What did you do that exposed an error?
Installed exporter via OS plugin (Prometheus Node Exporter) and noticed errors in log output, ran executable directly to narrow replication steps.
What did you expect to see?
Fewer errors, probably zero
What did you see instead?
An entire error, every time Prometheus polled it
I'm trying to be humorous with the last two, ideally it doesn't just tick someone off. Anyway, since it'll come up; I want to share mdstat but it's full to the brim with in-warranty serial numbers and I've been cautioned against exposing that to the public. I've removed the lines with valid disk data. I expect the one causing trouble is number 29,
rdevStatus.29=DISK_NP_DSBL-- I did not edit any lines in 29's block. I don't know what that disk is, my OS (Unraid) manages md and related systems for me.