Create endpoint for site metrics

[jwfuller]

Prometheus uses a scrape model rather than a push model for gathering metrics. We need each site to have an endpoint for scraping these metrics. We should secure this information transfer because it will contain user lists. This will replace atlas_statistics

[AlanBCole]

First try to explain work involved

Main Questions:

• What data do we want to gather for useful metrics? (Look into atlas_statistics)
• What form must the data be returned? JSON response?

Some more background for this issue:

• Prometheus is the tool we will be using to gather performance data from our websites/services.
• The 'scrape' or 'pull' model for gathering data means that Prometheus initiates http requests to services, assets, websites etc to gather desired data in contrast to services themselves initiating communication and sending data to Prometheus. This is why we need an endpoint to receive and respond to Prometheus' requests
• How does atlas_statistics work presently?
  • There are a number of parts in this puzzle, all of them seem to be somewhat dependent on each other:
    • The main action is in atlas_statistics module -> defines hook_atlas_statistics. This is what other modules call to gather their data.
      • cron job is set up to run every 4 hours that calls this hook and sends data to Atlas API via atlas_api_request()
    • atlas module (not our repo... but a drupal module) - defines atlas_api_request() which is used to send data to the Atlas API. This function is called in more places than just this and atlas_statistics modules.
    • cu_atlas module -> One of (the main one outside of atlas_statistics?) the places data is collected
    • atlas repo/API -> seems to be the place metrics data is sent from express websites/atlas module.
      • What does it do with the data?

Other Questions

• What exactly does 'secure' mean here?
  • simply sending all data over https?
  • removing sensitive details before sending http response? If so, what is too sensitive to include in this response?
• What are the shortcomings of atlas_statistics that we need to remedy with a new endpoint? -> Metrics gathering works on push model. Prometheus uses pull model.
• Will the Atlas API still need to be involved?

Working plan for this issue:

• create similar functionality as hook_atlas_statistics
• create endpoint something like that found at atlas_statistics.module LL 61-72, but one which sends a json response instead of just displaying it.
• no need to interface with Atlas API.

[jwfuller]

What exactly does 'secure' mean here?

Anonymous users should not be able to access the data, I would look at token based authentication as a start

What are the shortcomings of atlas_statistics that we need to remedy with a new endpoint?

I don't think the Web Express side has any clear failures, Atlas is not the best place to store this data and that is the main thing we are fixing

Will the Atlas API still need to be involved?

Not with metrics

[AlanBCole]

Issue

• new module -> maybe named express_site_metrics that creates an endpoint returning metric data for Prometheus.
• deprecate atlas_statistics

specific work items

• install method setting auth token to use when receiving Prometheus' requests
• Drupal authentication so that anonymous users cannot access new endpoint
• Token based authentication on top of ^ that
• JSON response
• new hook for modules to implement (as with hook_atlas_statistics) so modules can be included in site metrics as needed.
• implementations of this ^ hook where hook_atlas_statistics is implemented to be sure present metrics data is duplicated.