11. Failure

In this lesson we'll cover the last agent arm we haven't touched on yet: on-fail. We'll also touch on one last concept, which is the helper core.

Failures

When crashes or errors occur in certain cases, Gall passes them to an agent's on-fail arm for handling. This arm is very seldom used, almost all agents leave it for default-agent to handle, which just prints the error message to the terminal. While you're unlikely to use this arm, we'll briefly go over its behavior for completeness.

on-fail takes a term error message and a tang, typically containing a stack trace, and often with additional messages about the error. If it weren't delegated to on-fail:def, it would begin with:

++ on-fail
|= [=term =tang]
^- (quip card _this)
....

Gall calls on-fail in four cases:

  • When there's a crash in the on-arvo arm.
  • When there's a crash in the on-agent arm.
  • When there's a crash in the on-leave arm.
  • When an agent produces a %watch card but the wire, ship, agent and path specified are the same as an existing subscription.

For an on-arvo failure, the term will always be %arvo-response, and the tang will contain a stack trace.

For on-agent, the term will be the head of the sign (%poke-ack, %fact, etc). The tang will contain a stack trace and a message of "closing subscription".

For an on-leave failure, the term will always be %leave, and the tang will contain a stack trace.

For a %watch failure, the term will be %watch-not-unique. The tang will include a message of "subscribe wire not unique", as well as the agent name, the wire, the target ship and the target agent.

How you might handle these cases (if you wanted to manually handle them) depends on the purpose of your particular agent.

Helper core

Back in the lesson on lustar virtual arms, we briefly mentioned a common pattern is to define a deferred expression for a helper core named hc like:

+* this .
def ~(. (default-agent this %.n) bowl)
hc ~(. +> bowl)

The name do is also used frequently besides hc.

A helper core is a separate core composed into the subject of the agent core, containing useful functions for use by the agent arms. Such a helper core would typically contain functions that would only ever be used internally by the agent - more general functions would usually be included in a separate /lib library and imported with a faslus (/+) rune. Additionally, you might recall that the example agent of the subscriptions lesson used a barket (|^) rune to create a door in the on-poke arm with a separate handle-poke arm. That approach is typically used when functions will only be used in that one arm. The helper core, on the other hand, is useful when functions will be used by multiple agent arms.

The conventional pattern is to have the helper core below the agent core, so the structure of the agent file is like:

[imports]
[state types core]
[agent core]
[helper core]

Recall that the build system will implicitly compose any discrete expressions. If we simply added the helper core below the agent core, the agent core would be composed into the subject of the helper core, which is the opposite of what we want. Instead, we must inversely compose the two cores with a tisgal (=<) rune. We add the tisgal rune directly above the agent core like:

.....
=<
|_ =bowl:gall
+* this .
def ~(. (default-agent this %.n) bowl)
hc ~(. +> bowl)
++ on-init
.....

We can then add the helper core below the agent core. The helper core is most typically a door like the agent core, also with the bowl as its sample. This is just so any functions you define in it have ready access to the bowl. It would look like:

|_ =bowl:gall
++ some-function ...
++ another ....
++ etc ...
--

Back in the lustar virtual arm of the agent core, we give it a deferred expression name of hc and call it like so:

hc ~(. +> bowl)

To get to the helper core we composed from within the door, we use a censig expression to call +> of the subject (.) with the bowl as its sample. After that, any agent arms can make use of helper core functions by calling them like (some-function:hc ....).

Summary

  • on-fail is called in certain cases of crashes or failures.
  • Crashes in the on-agent, on-arvo, or on-watch arms will trigger a call to on-fail.
  • A non-unique %watch card will also trigger a call to on-fail.
  • on-fail is seldom used - most agents just leave it to %default-agent to handle, which just prints the error to the terminal.
  • A helper core is an extra core of useful functions, composed into the subject of the agent core.
  • Helper cores are typically placed below the agent core, and composed with a tisgal (=<) rune.
  • The helper core is typically a door with the bowl as a sample.
  • The helper core is typically given a name of hc or do in the lustar virtual arm of the agent core.