Scott,
Understood. Let me see if I can trap this one error in a safe way on the
server. Then we can make the best decision.
Randy
Scott Wilson wrote:
On 19 Aug 2010, at 14:40, Randy Watler wrote:
Scott,
Can you comment on the JIRA issue?
I am not sure if you think this issue is best fixed in the widget code or on
the server. If we want to fix it on the server, (as you appeared to be leaning
in previous comments), we'll need some ability to retry a request. This is not
easy to accomplish in all cases since the HTTP responses my be committed when
the exception is caught wrapping the transaction. I am not sure if it is
committed or not in this case, (something to look I suppose). As I noted, I
believe there is a race condition in the widget code that is the real issue
here. We could try to eliminate that or handle the error condition and retry
from the widget.
Thoughts? I'd like to take a crack at fixing this whichever way you prefer ASAP.
I don't think we can rely on widgets being coded "correctly" - so we shouldn't
generate a 500 error, but instead handle things more gracefully even if the widget is
written so as to make all kinds of odd simultaneous requests.
As you say it wouldn't be trivial to trap and retry as the requests are coming
in simultaneously on different server threads and wrapped in separate
transactions.
I wonder if we could check in the exception case to see if the object was
created anyway, and then return 200 instead of 500 and just log the issue?
Randy
Scott Wilson wrote:
Hi everyone,
I've just updated the Pluggable Persistence Branch so it is ready to merge into
the main trunk.
Currently there is one outstanding issue with the branch; see:
https://issues.apache.org/jira/browse/WOOKIE-145
We will also need to review the documentation to ensure it is consistent with
the new persistence mechanism and its configuration.
However at this point I would prefer to manage these issues within the trunk
rather than continue to maintain a separate branch, so I am proposing to merge
in the pluggable persistence branch.
What do you think?
S
