Re: [ruote:4033] How to handle exceptions the right way

[John Mettraux]

On Sun, Mar 24, 2013 at 08:59:42AM -0700, Adrien Kohlbecker wrote:
>
> I'm working with ludovic_ on the ruote-resque implementation, and I was
> wondering what's the best way to handle errors raised inside remote
> processes
> There seems to be different implementations in ruote-amqp and
> ruote-nats/ruote-beanstalk

Hello Adrien,

welcome to ruote's mailing list.

> At the moment, I'm sending back a workitem with an 'error' key and the
> receiver calls @context.error_handler.action_handle('error', fei, error)
> But in ruote-amqp, you use the flunk method, which does not seem to work in
> my tests (it does not display an error in ruote-kit)

Ruote-amqp should be followed (it's the most recent of the three iirc).

I added an assertion to the error propagation specs in ruote-amqp:

https://github.com/kennethkalmer/ruote-amqp/commit/6d2e36c8ebda418b438da5f64fbdc525a0734fcd

The error makes it in. Ruote-kit should display it. Maybe it's something
else. If you can provide a way to reproduce the issue, that'd be great.


> How would you go about implementing this ?
> The work in progress is hosted
> at https://github.com/adrienkohlbecker/ruote-resque

Very nice :-) Looking forward to see it completed.


Cheers,

--
John Mettraux - http://lambda.io/jmettraux

[Adrien Kohlbecker]

Hello John,

Just a follow up regarding this issue with errors not bubbling up Ruote when using flunk : 

It seems that when I switch to FsStorage, it works. (I was using ruote-postgres which would be our backend of choice when we deploy ruote)

I put this specific issue aside for the time being, but now I'm running into another issue with storage, this time with HashStorage. 

When I run the specs of ruote-resque using FsStorage, they pass. When I change the storage to HashStorage, they fail.

I don't know if the two issues share the same root cause, but they seem related.

What do you think ?

If you would like to reproduce the issue, the spec is at /spec/lib/ruote/resque/receiver_spec.rb

Regards,
Adrien

[John Mettraux]

On Fri, Mar 29, 2013 at 09:09:20AM -0700, Adrien Kohlbecker wrote:
>
> Just a follow up regarding this issue with errors not bubbling up Ruote
> when using flunk :
> It seems that when I switch to FsStorage, it works. (I was using
> ruote-postgres which would be our backend of choice when we deploy ruote)

Hello Adrien,

ruote-postgres is very young, looking at the commit messages, it doesn't seem
ready for now. Maybe you'd better use ruote-sequel until the ruote-postgres
team announces their library as ready.

> I put this specific issue aside for the time being, but now I'm running
> into another issue with storage, this time with HashStorage.
> When I run the specs of ruote-resque using FsStorage, they pass. When I
> change the storage to HashStorage, they fail.
> I don't know if the two issues share the same root cause, but they seem
> related.
>
> What do you think ?

FsStorage "naturally" makes deep copy of the documents it stores. Maybe the
HashStorage issue is related to mutability (storing something that's modified
later on).

> If you would like to reproduce the issue, the spec is at
> /spec/lib/ruote/resque/receiver_spec.rb

OK, I'll try it with the HashStorage. Give me a few days though.


Best regards,

[John Mettraux]

On Sat, Mar 30, 2013 at 10:20:12AM +0900, John Mettraux wrote:
>
> On Fri, Mar 29, 2013 at 09:09:20AM -0700, Adrien Kohlbecker wrote:
> >
> > I put this specific issue aside for the time being, but now I'm running
> > into another issue with storage, this time with HashStorage.
> > When I run the specs of ruote-resque using FsStorage, they pass. When I
> > change the storage to HashStorage, they fail.
> > I don't know if the two issues share the same root cause, but they seem
> > related.
> >

> > (...)

> >
> > If you would like to reproduce the issue, the spec is at
> > /spec/lib/ruote/resque/receiver_spec.rb

Salut Adrien,

I gave it a try and and it broke for me too, but it's nothing serious.

The first thing I noticed: in line 52 of the spec,

system('rm -rf spec/tmp/ruote_work')

is FsStorage specific. The better way to do it (works with all well-behaved
storage implementations):

@board.storage.purge!

The second thing: the receiver in the spec is instantiated and bound to
Resque but it's never shut down. When running the specs one by one they are
successful, when running them all, previous receivers fetch the work of the
current spec receiver... With FsStorage it looks alright, since the storage
is filesystem backed whatever ruote worker + fs storage pair picks up, it
ends up in the fs... With the HashStorage, the old receiver feeds the flunk
message to its old HashStorage whose Worker was carefully shut down by your
after(:each) block. The flunk msg just disappears...

You have to shut down the old receiver before the new spec interprets.


I hope it helps, meilleures salutations,

[Adrien Kohlbecker]

Thanks John, it does work indeed if I kill the listener thread :)