belfastelixir/dms

Dead Man's Switch Elixir OTP Application

DMS (Dead Man's Switch)

A simple OTP Application (and eventually an HTTP API) which enables short circuiting when a service becomes unavailable.

Used as a learning project for members of Belfast Elixir.

• Build: mix do deps.get, compile
• Test: mix test
• Documentation: mix docs
• REPL: iex -S mix
• Observe: :observer.start

API Examples

# Ping will register service with that id as alive
id = "my-svc-id"
DMS.ping(id) # => :pong
DMS.alive?(id) # => true ~ Service with id responds as alive
# After @timeout period and no further pings
DMS.alive?(id) # => false ~ Service with id responds as dead

# An id which has never pinged will also respond as dead
false = DMS.alive?("another-svc-id") #=> false

Checklist

• (#13) Model a Service as a Process (DMS.Service)
• (#1) Register Service against an id (DMS.Servce.Registry)
• (#2) Add a supervisor for Service Process (DMS.Service.Supervisor)
• (#3) Add ping(id) function which will init Service if doesn't exist
• (#4) Add alive?(id) function which returns true if service alive otherwise false
• (#5) Add timer to service process which will terminate the process if it hasn't been pinged within @timeout
• (#12) Add Documentation with ExDoc
• (#6) Add Credo & Dialyzer for linting and static analysis.
• (#7) Add API to set service as down (before timeout).
• (#8) Enhance DMS.id to contain service id and account token + hash it before use as key in DMS.Servise.Registry.
• (#9) Add HTTP API.
• (#10) Add Pub/Sub push based API to notify when service is down.
• (#11) Publish on HEX package manager.

dms's Issues

Enhance DMS.id to contain service id and account token + hash it before use as key in DMS.Servise.Registry.

Make kill switch timing testable without sleeps

Refactor kill_switch function into own module to allow time-based unit testing without Proces.sleep call.

Register Service against an id (DMS.Servce.Registry)

Add Pub/Sub push based API to notify when service is down.

Signal when service process times out!

Add alive?(id) function which returns true if service alive otherwise false

Fix race condition on Service initialisation

DMS/lib/dms/service.ex

Lines 25 to 35 in 1efe918

	if exists?(id) do
	GenServer.cast(via_registry(id), :ping)
	:pong
	else
	child_spec = %{
	id: id,
	restart: :temporary,
	start: {__MODULE__, :start_link, [id]}
	}
	{:ok, _} = DynamicSupervisor.start_child(@supervisor, child_spec)

This would result in a race condition when 2 attempts to start the same service happen at once.

This should be resolved by leveraging the return value of DynamicSupervisor.start_child/2 which is the following:

  {:ok, pid()}
  | :ignore
  | {:error, {:already_started, pid()} | {:shutdown, term()} | term()}

In particular the :already_started value sounds like a good option to resolve this: "If a process with the specified server name already exists, this function returns {:error, {:already_started, pid}} with the PID of that process."

However the processes are not named, and the Registry is being used, so we may need to explore the Registry functionality in order to find the correct solution.

Add HTTP API

Add Documentation with [ExDoc](https://github.com/elixir-lang/ex_doc)

Model a Service as a Process (`DMS.Service`)

Add a supervisor for Service Process (DMS.Service.Supervisor)

Once #1 has been completed we should implement this.

Add in CI

We want this repo to have a CI flow. The CI should be executed when a new branch is pushed or a push to master occurs.

Don't forget to add the relevant buttons to the README.

Add Credo & Dialyzer for linting and static analysis.

Compiler Warnings Introduced

warning: unused import DMS.Service.Registry.whereis_name/1
  lib/dms/service.ex:8

Generated dms app
warning: unused import Task
  test/dms_test.exs:13

Slow Test on Kill Switch

The timer/kill switch signal is causing the test which verifies it to be too slow.

This will need to be extracted, mocked or configured in such a way that it can run much faster for the sake of not incurring long wait times within the unit tests.

Add timer to service process which will terminate the process if it hasn't been pinged within @timeout

Improve definition and validation of "Service.id()" parameter

During the last couple of meet-ups the topic of valid Service ids has came up.

The current constraints are as follows:

• Any non-empty binary < 256 bytes.

The current thinking is that this will be improved to be:

• Alpha-numeric.
• Supports - and _.
• Does not allow print characters or special characters.

Remove redundant ping call when service process created

This line is not required.

DMS/lib/dms/service.ex

Line 31 in c2661fa

GenServer.cast(via_registry(id), :ping)

Add ping(id) function which will init Service if doesn't exist

Handle ping/1 function failure condition

DMS.ping/1 has a contract of returning :ping on the successful acknowledgement that service will return true when asked if alive for timeout period.

It also states it will return :pang when this is not the case when the DMS will not guarantee to signal alive to be true for the service. This condition is not implemented.

Publish on HEX package manager

Add API to set service as down (before timeout).