BackOff Supervisor

[richardimaoka]

Break down to multiple issues / articles

[richardimaoka]

---

Summary:

Write these at the end? (as it will be giving too much info to beginners)

Note: implementation details can change so don't go in too much

BackoffSupervisor - companion obj with different factory methods

• def props(options: BackoffOptions): Props
  • typically used as:
  • BackoffSupervisor.props(Backoff.onFailure(...)), or
  • BackoffSupervisor.props(Backoff.onStop(...))
• def props(childProps, childName, ..., randomFactor, replyWhileStopped): Props
  • //hmmm? was it necessary if we pass default for strategy in the below signature?
• def propsWithSupervisorStrategy(childProps, childName, ..., randomFactory, strategy, replyWhileStopped): Props

Note that these factory methods accepts childProps i.e. child actor definition as a parameter.

Also this companion object defines...:

• GetCurrentChild <- send this to BackoffSupervisor then you get BackoffSupervisor's child (ask (!) pattern)
• GetRestartCount <- send this to BackoffSupervisor then you get the restart count of the child child (ask (!) pattern)

HandleBackoff - trait having message-sending implementation with Backoff algorithm

---

BackoffOptions - a companion to create Backoff "Options" to be passed to BackoffSupervisor factory metods (i.e. def props)

• onFailure
• onStop

Note that these factory methods accepts childProps i.e. child actor definition as a parameter.

---

BackoffSupervisor actually has two concrete classes - it's confusing but in two separate files:

• final class BackoffSupervisor in BAckoffSupervisor.scala
  • differentiate it from its companion BackoffSupervisor object
• private class BackoffOnRestartSupervisor in BackoffOnRestartSupervisor.scala

These two classes are instantiated as follows, when creating backoff options:

• BackoffOption's factory methods to create BackoffOption
  • onFailure()
  • onStop()
• Or BackoffSupervisor's factory methods props() to create BackoffOption inside

Then, BackoffOptionsImpl -> def props = { backoffType match { ... } } (this is too much implementation detail, so you might not want to go into this in the article)

[richardimaoka]

What is...

Pic to illustrate restart with back off Link to actor lifecycle Pic to illustrate exponential interval

why

Normal situation, ask and reply -> vine to illustrate

When target fails, DB down, web server down

Retry ... sometimes it works ok (slow target), sometimes not

Typical configuration, or hard coding ... pseudo code to specify reply

Problem ... slowness could be resolved by quicker retries However, 1000s of clients doing quick retries... target slowed down by retries

Conclusion, retry interval to change over time