James,
This would be a good solution for people using Camel with ActiveMQ (and
possibly also for ServiceMix), but isn't Camel supposed to be more
transport/technology-agnostic? If Camel is also intended to be used as
a mediation/routing engine for CXF, MINA, ..., I don't think there is a
single, non-sucky default error handler configuration for all those.
Would it be an option to have no default error handler inside Camel, but
let the underlying system add something non-sucky (with a big WARN
message) implicitly whenever a good default for that system exists and
no error handler was set in the RouteBuilder itself?
Gert
James Strachan wrote:
BTW had a thought for changing the default to something non-sucky.
How about if we kept the retry-6-times (maybe with exponential backoff
being the default) - but rather than using the log as the dead letter
queue (which sucks - my bad! :) - we defaulted the dead letter queue
to being something like "activemq:Camel.DeadLetter")
Then if the caller doesn't have ActiveMQ configured, its gonna fail -
hence the whole thing will fail? If folks have ActiveMQ installed then
the default is something reasonable - use retry then a real,
persistent dead letter queue?
I like the idea of convention over configuration and doing the right
thing out of the box; though for things like retry/dead letter its
maybe worth users specifying that stuff; however I wonder if using a
DLQ with ActiveMQ is a sensible default - as its either reliable &
persistent if ActiveMQ is available -or it fails (and so acts like no
error handler :)
Thoughts?
2008/6/20 Claus Ibsen <[EMAIL PROTECTED]>:
Hi Gert
Great insight. And sorry about the problems with the API breakings. We should target all
the "fix the tangle etc." tickets to Camel 2.0, so there wont be another API
breaking.
+1 for that one
Do you have a list at hand what the API breaking were between 1.3 and 1.4? We
should at least document this in the release note that there can be API
breakings.
+1 of course for the dead letter channel as well. We cant have them being
retried 6 times within 1 sec. Interval and then just stoved away in a log ;)
Maybe somekind of INFO/WARN level when the route context is created/started
that the DLC is NoErrorHandler so end users will notice that they might haven't
thought/set a error handling strategy fitting to their needs.
Med venlig hilsen
Claus Ibsen
......................................
Silverbullet
Skovsgårdsvænget 21
8362 Hørning
Tlf. +45 2962 7576
Web: www.silverbullet.dk
-----Original Message-----
From: Gert Vanthienen [mailto:[EMAIL PROTECTED]
Sent: 19. juni 2008 08:40
To: [EMAIL PROTECTED]
Subject: Deprecating old methods and changing the default error handler
L.S.,
While I was doing a project with a customer last week, I ran into a few
problems with using Camel for them.
First problem were changes in the API (e.g. Endpoint.getContext()
becomes Endpoint.getCamelContext()). I needed to rebuild
servicemix-camel to get some fixes there, but along came a new version
of Camel and that was breaking a lot of the existing code. We might
want to consider deprecating the old methods instead of removing them.
Not entirely sure about this though: it will probably slow us down a
bit, but on the other hand, when a user updates from 1.3 to 1.4 he
probably doesn't expect API breakage of this type. They would expect
this to happen from a major release though (Camel 2.0), so I would
suggest cleaning up all deprecated code at that point. Wdyt?
A second thing that kind surprises people is our implicit use of a dead
letter channel in every RouteBuilder. Let me give an example:
public class MyRouteBuilder extends RouteBuilder {
public void configure() {
from("jbi:endpoint:urn:ns:service:endpoint1").to("jbi:endpoint:urn:ns:service:endpoint2");
}
}
If you look at the code above, it appears to be routing from one JBI
endpoint to the next one, without doing anything else. Now, in reality,
if something goes wrong at the second endpoint (e.g. FTP server goes
down), the default DeadLetterChannel kicks in and after 5 retries, it
will "log away" the message. The first JBI component will never know
something went wrong, because the JBI Exchange going back will just say
DONE and the only trace of something going wrong is in the logs -- the
message itself was lost in the process. We should at the very least
warn people to explicitly set an errorHandler() when using Camel inside
Servicemix (usually just errorHandler(noErrorHandler()), but actually I
don't think we should make this behavior the default. It might be good
for testing/demos, but in real life you probably want to specify a more
advanced error handling routine or let the underlying system (e.g.
ServiceMix) take care of it. In my mind, we should seriously consider
making no error handler the default for Camel 2.0. Once again: wdyt?
Regards,
Gert
