Crash Handler

Home | Getting Started | API | Elements | Actions | Validators | Handlers | Configuration Options | Plugins | Mobile | Troubleshooting | About

Table of Contents

1 Crash Handler

The crash handler provides a simple mechanism for dealing with page crashes to provide more user-friendly notifications, as well as giving the programmer the ability to hook whatever kind of logging they want into errors.

The previous functionality in Nitrogen, when a page simply crashed, was to return "Internal Server Error" and send a 500 error to the client. This, unfortunately, results in a big scary white page with a useless error message for the connected client, or in the case of AJAX, silently failing (buttons that do nothing, and the user thinks "Hey, maybe it's just running slowly", when in fact, it's crashed.

The crash handler allows you to present a template to the user, wire events, and overall present something more useful (or at the very least, more friendly) to your customer (for example, presenting a formatted page saying "Sorry, something has gone awry, we've been notified and are looking into the problem").

Behavior Functions

init(Config, State)

Initialize the handler

  • Return Value - {ok, NewState}
finish(Config, State)

Clean up the handler

  • Return Value - {ok, NewState}
first_request(ErrorType, Error, Stacktrace, Config, State)

If the client's request was a "first request" (not a postback, comet, or websocket request), this will be called with the Error information.

  • ErrorType - This is merely the error type retrieved from the class portion of a catch statement, typically this will simply be error, or it could be throw if it's a programmer-thrown error, or exit if the process called exit/1
  • Error - This will be the more descriptive nature of the error, for example badarg or {badmatch, V}. See Errors and Error Handling Section 10.4 for more descriptions of the error.
  • Stacktrace - This will be the result of calling erlang:stack_trace/0 at the time of the error.
  • Config - Any configuration associated with the crash handler.
  • State - Any state information associated with the crash handler (likely unused).
  • Return Value - The body to be sent to the browser.

postback_request(ErrorType, Error, Stacktrace, Config, State)

If the client's request was a an AJAX, Comet, Websocket, or other non-initial request, this function will be called.

It takes all the same arguments, but the return value is discarded, hence the recommendation to simply return ok.

Instead, anything you wire to the browser will be sent. This could allow you to send a simple, useful little alert box to the user saying "We're sorry, there was an issue processing this request, please try again or reload the page". At least then the user will get some kind of feedback and not just silently failing.

Example

Here is the complete text of the default crash handler, which is simple, and for production use, is probably too simple, but it demonstrates the most basic form of crash handler.

-module(default_crash_handler).
-behaviour(crash_handler).
-include("wf.hrl").
-export([
    init/2,
    finish/2,
    first_request/5,
    postback_request/5
]).

init(_Config,State) ->
    {ok, State}.

finish(_Config, State) ->
    {ok, State}.

first_request(Type, Error, Stacktrace, _Config, _State) ->
    %% Print the error message to the Erlang console
    ?LOG("~p~n", [{error, Type, Error, Stacktrace}]),

    %% Set the response status code to 500 (internal server error)
    wf:status_code(500),

    %% Send just the text "Internal Server Error" with no formatting or layout
    "Internal Server Error".

postback_request(Type, Error, Stacktrace, _Config, _State) ->
    %% Print the error message to the Erlang console
    ?LOG("~p~n", [{error, Type, Error, Stacktrace}]),

    %% Set the status code to 500 (internal server error)
    wf:status_code(500),

    %% Do nothing else
    ok.

A more useful example would be a crash handler that uses a template and pops a message to the user when a postback is used.

-module(simple_crash_handler).
-behaviour(crash_handler).
-include_lib("nitrogen_core/include/wf.hrl").
-export([
    init/2,
    finish/2,
    first_request/5,
    postback_request/5,
    body/1
]).

init(_Config,State) ->
    {ok, State}.

finish(_Config, State) ->
    {ok, State}.

first_request(Type, Error, Stacktrace, _Config, _State) ->
    %% Print the error message to the Erlang console
    ?LOG("~p~n", [{error, Type, Error, Stacktrace}]),

    %% Set the response status code to 500 (internal server error)
    wf:status_code(500),

    #template{
      file="./site/templates/error.html",
      bindings=[{'Stacktrace', Stacktrace}],
      module_aliases=[{page,?MODULE}]
    }.

%% This is assuming that the error.html template includes a call to [[[page:body(Stacktrace)]]]
body(Stacktrace) ->
    [
      #h1{text="UH OH! Something went wrong!"},
      #panel{text=wf:f("Here's the contents of the error: ~p",[Stacktrace])}
    ].


postback_request(Type, Error, Stacktrace, _Config, _State) ->
    %% Print the error message to the Erlang console
    ?LOG("~p~n", [{error, Type, Error, Stacktrace}]),

    %% Note, we don't set the status code to 500. If we did, the browser will
    %% simply discard any javascript. So we keep a 200 status code and print
    %% a friendlier error message.

    Msg = wf:f("I'm sorry, but there was an error. Here's the stack trace: ~p",[Stacktrace]),
    wf:wire(#alert{text=Msg}).

Recommendations for first requests

It's very likely that for handling first requests, you'll want to display a template to the user, rather than simple raw text or hand rolled HTML.

In that event, it's probably very likely you'll want to redirect requests to the standard page module to the crash handler module (or a custom crash module). This is done with the module_aliases attribute on the #template element. This is especially useful if you want to use the same template for your errors that you might use with the rest of the page.

If you plan on sending any of the arguments passed to first_request/5 to a template, make sure you take advantage of the bindings attribute.

Recommendations for postback requests

If you want to gracefully handle crashes related to postbacks, comets, etc, then you need to make sure you do not set the statuscode to a failure code (like 500). You might as well keep it as the default 200, and then wire whatever commands you wish.

At the very least, we recommend giving the user some kind of feedback, even if it's a simple javascript alert (See Alert Action below).

What if your handler crashes?

If your custom crash handler crashes, then Nitrogen will fall back to its default behavior of throwing a 500 error and returning "Internal Server Error".

It's recommended that, much like Erlang supervisors, that your crash handler modules be very simple, with minimal moving parts, so as to prevent this from happening. For example: It's probably not a great idea to hinge your crash handler to database availability, unless you verify with the ErrorType and Error arguments that the error is not related to database availability.

Using those arguments, you can custom tailor the error messages to certain classes and show different templates or content depending on the message.

But the simplest error template should just be a mostly static HTML template that probably doesn't even make an page calls. That will ensure that your crash handler doesn't crash, effectively negating its existence.

Date: 2013-10-03 13:41:15 CDT

Author: Jesse Gumm (@jessegumm)

Org version 7.8.02 with Emacs version 23

Validate XHTML 1.0

Comments

Note:To specify code blocks, just use the generic code block syntax:
<pre><code>your code here</code></pre>


comments powered by Disqus