Prometheus metrics hardcoded to start at port 8080

[mmonaco]

The metrics added in #310 listen separately at each mount, with a starting port of 8080. We should be able to customize the start port, but I think it would need to come from the sidecar annotation (gke-gcsfuse/prometheus-start-port: "9090" for example).

With this, prometheus-port doesn't need to be a disallowed flag. Instead it would skip incrementing from the start port.

(Are there options to disable the metrics if users don't want them at all?)

[hime]

Hi @mmonaco, to disable metrics, could you add the following volumeAttribute to your spec? See details here

volumeAttributes:
 ...
 disableMetrics: "true"

[mmonaco]

Sure, that's a workaround, but taking 8080 by default is quite abrasive.

[jcxplorer]

Just want to chime in and say that this change would have caused a production outage for us, which we fortunately caught in one of our staging clusters. We had one deployment with this plugin enabled that had a container already using port 8080, which started failing after a routine GKE upgrade.

Given the potential impact of this change, should metrics collection be opt-in instead?

[sirohia]

I agree. The metrics collection port should be customizable. Hard coding to 8080 is very strict.

[hime]

Thank you all for your feedback. We have recently disabled metrics exporting by default in #375 to prevent these types of frictions when using the driver. Regarding the port, I agree that setting this port number as default is rather strict. Marking this issue as an enhancement to eventually increase flexibility with users that would like to use metrics exporting.

[avkudryashov]

We have recently disabled metrics exporting by default in #375

Thank you for the rapid fix. Can you also provide some information to the Kubernetes Engine release notes? Since this is a breaking change, it would be good to know which GKE version includes the fixed driver version.