List of enhancements

[KalenWessel]

Info

Grafana version: v8.4.3
Dark: v0.9.2

First of all, I want to say this is an awesome tool! Out of everything I've looked at DARK seems to work the best for supporting the idea of Dashboards as Code in an easy-to-read YAML format.

Grafana is a very feature-rich product so it comes with no surprise that DARK and Grabana don't currently handle every option. Here is a list of settings and issues I have come across while trying to port some pre-existing dashboards over to DARK.

The converter tool does not handle converting most of my pre-existing dashboards. I'm guessing my dashboards contain settings which DARK is unable to convert.

Example:

❯ docker run --rm -it -u $(id -u):$(id -g) -v $(pwd):/workspace kphoen/dark-converter:latest convert-yaml \
        -i dashboard.json \
        -o converted-dashboard.yaml

WARNING: The requested image's platform (linux/amd64) does not match the detected host platform (linux/arm64/v8) and no specific platform was requested
2022-04-12T22:32:34Z	ERROR	converter/json.go:113	could not unmarshall dashboard	{"error": "json: cannot unmarshal object into Go struct field Board.panels of type string"}
2022-04-12T22:32:34Z	ERROR	converter/json.go:55	could parse input	{"error": "json: cannot unmarshal object into Go struct field Board.panels of type string"}
2022-04-12T22:32:34Z	FATAL	cmd/toyaml.go:30	Could not convert dashboard	{"error": "json: cannot unmarshal object into Go struct field Board.panels of type string"}

When trying to manually port dashboards from grafana over to DARK I have run into some missing features:

Missing Features

I tried to look through all the docs and code. It's possible I simply missed the correct fields.

Alerts

Example:

apiVersion: k8s.kevingomez.fr/v1
kind: GrafanaDashboard
metadata:
  name: sample-dashboard
folder: Sample
spec:
  title: Sample Dashboard
  editable: true
  shared_crosshair: true
  tags: [dark, generated]
  auto_refresh: 5s
  time: [now-6h, now]
  timezone: utc

  variables:
    - datasource:
        name: Source
        label: Prometheus
        type: prometheus
  rows:
    - name: SLIs
      panels:
        - timeseries:
            title: Uptime
            description: Synthetic probe success rate for ArgoCD
            datasource: $Source
            span: 4
            height: 350px
            targets:
              - prometheus:
                  query: probe_success{instance="https://argocd.localhost", environment="mgmt"}
                  legend: Uptime
                  ref: A
            visualization:
              gradient_mode: opacity
              fill_opacity: 10
              tooltip:
                "all_series"
            axis:
              unit: percentunit
              max: 1
            alert:
              title: "WARNING: ArgoCD Uptime alert"
              message: "ArgoCD uptime has dropped below 99%"
              evaluate_every: 1m
              for: 5m
              notify: 1
              tags:
                severity: warning
              on_no_data: no_data
              on_execution_error: alerting
              if:
                - operand: and
                  value: {func: last, ref: A, from: 5m, to: now}
                  threshold: {below: 0.99}

• Alerts do not appear under the main alert/list page.
• When creating an alert for say a time series panel, the Alert count will go up but no alert is linked to the panel.
  
• Unable to set labels

Time Series

Unable to set the following:

• line interpolation
• line width
• Connect Null Values
• Thresholds

Single Stats Panel

Unable to set the following:

• Stat styles:
  • Orientation
  • Text mode
  • Text alignment

Misc

• Values Map (ex null -> 0 with Color orange)
• Single threshold value. Currently requires at least two threshold values. Should this not support the ability to use the "base" threshold and then one extra threshold value?

[K-Phoen]

Hey!

Thanks for this very good feedback :)

I'm going to answer it in order:

• alerts
  • there has been a lot of work on alerting support in grabana lately. Until recently, the new "unified alerting" system that Grafana exposes wasn't supported. It now is, which should solve your problem
  • alert labels: they're called tags in dark/grabana → https://github.com/K-Phoen/grabana/blob/092b4d3027b029893ffdba61d2933fd42845185d/decoder/alert.go#L19
• timeseries
  • line interpolation: supported via https://github.com/K-Phoen/grabana/blob/092b4d3027b029893ffdba61d2933fd42845185d/decoder/timeseries.go#L182
  • line width: supported via https://github.com/K-Phoen/grabana/blob/092b4d3027b029893ffdba61d2933fd42845185d/decoder/timeseries.go#L183
  • "Connect Null Values": not supported, there is an open ticket
  • thresholds: not supported in YAML, only in the Go builder. [Open ticket](https://github.com/K-Phoen/grabana/issues/
• single stat panel: I would recommend using the non-deprecated stat panel instead, which supports the options you mentioned

[KalenWessel]

Amazing, thanks for working on these! I'll spend some time building some more dashboards and report back any more findings. Cheers!