Skip to content

Latest commit

 

History

History
88 lines (64 loc) · 4.75 KB

README.md

File metadata and controls

88 lines (64 loc) · 4.75 KB
Raw

Background

This is an out of the box solution for ingesting periodic health pings from services that cannot expose a public HTTP endpoint - such as serverless webhooks, cron jobs, etc. This ping service receives pings with a service name, a frequency and optionally a timestamp. When this service receives a ping, it's expecting to receive another ping within the timeframe inferred from the frequency param - if it doesn't hear within that frequency, an alert will be sent to a downstream alerting service (which can be configured).

Running Locally

I use asdf as my local package manager - to get up and running you'll need to run the following from the root of the directory:

# on mac
brew install asdf
asdf install

Additionally, you'll need to set the following environment variable PING_ALERT_URI - this is the Alerts Service (AS) callback URI:

echo 'export PING_ALERT_URI="some.alert.service"' >> ~/.zshrc # or ~/.bashrc if you use bash

After that's set up run the following from the root of the project directory:

mix deps.get && mix deps.compile

Now you'll be able to start up the server locally or run the tests using either of the following commands:

iex -S mix
mix test

About the Ping Server

GET /ping

The /ping endpoint accepts 3 parameters, 2 are necessary (name, frequency) and 1 is optional (timestamp), in the following example both are valid requests:

curl "127.0.0.1:4000/ping?name=test_service&frequency=10m"
curl "127.0.0.1:4000/ping?name=test_service&frequency=10m&timestamp=1664743905"

If a timestamp parameter is not given, a timestamp will be generated on insertion into the backend server.

The following are accepted as parameters:

  • The name parameter is any valid string.
  • The frequency parameter lets the service know how often it should be expecting a health ping. Valid formats are #{integer}#{time_unit}, with the following as valid time units - [W, D, h, m, s], representing [Weekly, Daily, Hourly, Minutely, Secondly] frequencies respectively. You can change frequencies on the fly - i.e. if I receive a ping at 10:00:00AM with a frequency of 1m, and then at 10:00:30AM I receive another ping from the same service with a frequency of 10m the service will expect the next ping to come at 10:10:30AM instead of 10:01:00AM.

For example, if we have a job called daily_digest. This job will call the ping service every time it is run as follows:

GET /ping?name=daily_digest&frequency=1D

Once the ping service receives this request, it should expect the subsequent request within the next day, failing to hear back within that timeframe an alert endpoint will be called. For local development you can set up your own webhook on webhook.site as the alert endpoint.

DELETE /ping/:name

In the case that an upstream service becomes deprecated, there is an endpoint for removing that service from the internal monitoring system.

This endpoint accepts 1 parameter:

  • name — the name of the service

For example, if we have a job called daily_digest and we deprecate this job (maybe in favor of a weekly_digest job) we no longer have a need to monitor this job because we're no longer expecting pings. To remove this job we could call:

curl -X "DELETE" "127.0.0.1:4000/ping/daily_digest"

Performance Rewrite

The HTTP server for this service was rewritten so that Phoenix and it's dependencies were dropped and replaced with a plain old plug cowboy http server. The original Phoenix project was spun up using the following flags: --no-assets --no-ecto --no-html --no-gettext --no-live --no-dashboard --no-mailer so that this service would have as lean of an API as possible, but using Phoenix still came with a very large performance overhead cost.

By ripping out Phoenix and its dependencies, I saw about 90% faster API response times.

The original Phoenix version is in the branch phoenix-version if you'd like to test it out yourself.

Endpoint HTTP Server Avg Response Time
GET /ping Plug Cowboy 104μs
Phoenix 1005μs
DELETE /ping Plug Cowboy 63μs
Phoenix 702μs

Further Work

  • Add /ping/remove endpoint which accepts a name parameter.
  • Performance tuning - speed up API by ditching Phoenix.
  • Replace :tesla with :req
  • Add guard for last_ping_timestamp - don't allow incoming pings to overwrite previous pings
  • Guard against negative intervals
  • Move state off of this service - maybe try out SQLite -> Fly's LiteFS. Could potentially have an adapter for stateful / stateless (current)
  • Add some automated CI/CD for testing/formatting/linting.