Issue collecting duplicate entries from /proc/mounts

[filippog]

Hi,
on node_exporter 0.13.0 rc2 I've noticed scrape failures on duplicate entries in /proc/mounts.

* [from Gatherer #1] collected metric node_filesystem_size label:<name:"device" value:"/dev/md0" > label:<name:"fstype" value:"ext4" > label:<name:"mountpoint" value:"/exp/backup-others" > gauge:<value:1.1903515017216e+13 >  was collected before with the same name and label values

e.g. with

/dev/md0 /exp/backup-others ext4 rw,relatime,errors=remount-ro,data=ordered 0 0
/dev/md0 /exp/backup-others ext4 rw,relatime,errors=remount-ro,data=ordered 0 0

Admittedly this is a machine misconfiguration though AFAICT duplicate entries are harmless on (recent?) Linux. I think duplicate mounts shouldn't make the whole collector fail but e.g. be logged and continue instead.

[discordianfish]

@beorn7 Looks like in such case we just send the same set of ConstMetric down the Collect channel.
Is that something the client library should handle or should we prevent this here? Or both?

[beorn7]

Ideally, your exporter code makes sure to never create duplicate metrics.

But you can also leverage the client library. With v0.8 of client_golang, you can configure the behavior in case of errors. You could log the error and not expose the offending metric (first metric wins) instead of letting the whole scrape fail (which is the default).

You should never expose inconsistent metrics (which the client library tries to avoid as far as possible).

[discordianfish]

@filippog This is fixed in master and will be included in the 0.14.0 release.

[filippog]

sweet, thanks @discordianfish !

[aparcar]

I tried the master version and have a similar problem with node_hwmon_temp_celsius

An error has occurred during metrics collection:

collected metric node_hwmon_temp_celsius label:<name:"chip" value:"dell_smm" > label:<name:"sensor" value:"ambient" > gauge:<value:30 >  was collected before with the same name and label values

Should I open a new issue?

[discordianfish]

@aparcar OH.. yes please, a new issue would make sense.

[Vincent-Le]

Hi @discordianfish, can you please share details on how this was fixed? We're seeing the same issue with collectd exporter.
Thanks!

[pgier]

@Vincent-Le if you take a look at #399, this was fixed by using a map to keep track of existing metrics and prevent duplicates.