add prometheus display format

[andrewsmedina]

Prometheus (prometheus.io) is a very powerfull open-source monitoring solution. Prometheus get the data scraping metrics from the applications.

To use prometheus with nginx-module-vts is a good way to monitoring and aggregate metrics from more than one nginx servers.

A way to do it is the nginx-module-vts support prometheus as a display format.

[vozlt]

Hi, Thanks for your suggestion.
I'll get it checked soon.

[vozlt]

Hi, the prometheus exporter for nginx-module-vts has been released by hnlq715 in issue #62.
Try to nginx-vts-exporter.

The screenshot nginx_server_connections as follows:

[gianrubio]

@vozlt I'm already using nginx-vts-exporter to collect prometheus metrics. It's really easy to use.
My question is why not merge the nginx-vts-exporter with this project and have prometheus metrics by default?

[vozlt]

@gianrubio Thanks for your good suggestion.
I have never used prometheus but I installed it because of this issue.
So I don't know detail about it yet.
But I think that it is better to make it a 3rd-party module like nginx-vts-exporter outside nginx-module-vts to do not affect nginx performance.

[gianrubio]

@vozlt prometheus could be an another format for the stats. It'll be a new endpoint like status/format/prometheus. The kubernetes project are using your module for the nginx ingress controller. I'm really interested to make the prometheus format easy to use for everyone.

It'll be similar to this screenshot.

kubernetes/ingress-nginx#72

@hnlq715 what do you think?

[sysulq]

The requested feature may break the internal design in vozlt/nginx-module-vts. In my oppinion, it's much wiser to start a new project for nginx-prometheus-metrics, or use hnlq715/nginx-vts-exporter instead.

[gianrubio]

@hnlq715, could you explain why this will break the design?

[pavelnemirovsky]

Guys did u merge this PR to master?

[trnl]

Guys, I also agree that it will not break any designs in nginx-module-vts. Prometheus is just another output similar to existing json, jsonp and html and rendering of it will be a no brainer.

[sysulq]

@gianrubio sorry to reply your question so lately...

Prometheus in vts should use histogram instead of gauge type for time metrics, not like hnlq715/nginx-vts-exporter, which means a huge change for vts.

Another choice here would be hnlq715/nginx-prometheus-metrics, which use lua embedded in nginx.

[trnl]

@hnlq715, I agree that this will be a change for vts, however, I believe histograms are not mandatory for now and simple nginx-vts-exporter style gauge output will be fine. I believe that was an original suggestion by @gianrubio.

[sysulq]

@trnl @vozlt @gianrubio
I agree, without supporting histogram type, this change would be acceptable.

But it would be great to support histogram for time metrics, and we could achieve this with lua or c module. After all, we could have p90, p95 or p99 request/response time metrics... or even apdex score

[sysulq]

Hope this could help you guys, simple lua code generates nginx's prometheus metrics.
https://github.com/hnlq715/nginx-prometheus-metrics

[trnl]

@hnlq715, actually yes. We rely on that now. However it will be nice to have the same stuff in nginx-ingress-controller for Kubernetes.

[ghost]

#114

[vozlt]

That feature has added.
All mertric names are started with nginx_vts_(*).
I can not guarantee that it will work well because I know little about prometheus.
Please test it.
Thanks.

Latest commit: 2f9434f

[vozlt]

The metric type of histogram in format/prometheus has added.
See the directive vhost_traffic_status_histogram_buckets for details.
I quickly added the feature and did not test it much.
So it is necessary to verify that I have implemented the function correctly according to prometheus histogram metric type.
Please test it.
Thanks.

Latest commit: 2fd3ae4

[hadret]

Overall format/prometheus appears to work very well. I haven't test the histogram buckets yet, but I noticed some other things missing when it comes to the upstreams part:

• State of the backend is not provided
• Weight of the backend is not provided
• MaxFails of the backend is not provided
• FailTimeout of the backend is not provided

All of the above are very useful to have and are provided by other formats.