Hi:
I haven't create a JIRA yet, I thinks I should illustrate my
question more clearly to see if the end-marker is worth to add.
My question could be illustrated by a simplified application:
consume a very large file which store record , and save all the record of
the file to database.
First: following is my first solution using current camel:
<from uri="file://source?consumer.regexPattern="*\.dat">
<doTry>
<split stopOnException="true">
<method bean="fileSplitBean" method="split"/>
<bean ref="saveToDatabase">
</split>
<doFinally>
<bean ref="fileSplitBean" method="release"/>
</doFinally>
</doTry>
The bean "fileSplitBean" has two method : method "split" used by
split-pattern will open the file, and create a iterator to enumeration the
record in file. Method "release" will close the opened file;
But for me, what I feel uncomfortable is "normal process action"(here is
fileSplitBean.split) and "release action"(here is fileSplitBean.release) is
separated in two place. The sample I give is very simple, but if route is
more complex, at just a glance, people cannot realized the fact that: the
bean used for splitting also has a concerned "release action"; In another
words, the solution is too close to the program-language.
Next , I tried camel to let the "normal-process action"and "release action"
appeared more closely in the route. So the solution will appears as
following: 1)change the iterator, and let the iterator return a very special
end-flag message after all the record of file has been enumerated;2) the
bean "saveToDatabase" knows the end-flag message,and do nothing for it. 3)
give a aggregate for the splitter and the aggregator will test if the
message is end-flag to see. But this solution look too odd, and even cannot
works while exception occurs;
So in order to enhance the solution , I thought about to let split-pattern
give the end-flag, and using a aggregator to close file; the solution as
following:
<from uri="file://source?consumer.regexPattern="*\.dat">
<split stopOnException="true" strategyRef="releaseStrategy">
<method bean="fileSplitBean" method="split"/>
<bean ref="saveToDatabase">
</split>
<bean id="releaseStrategy" class="...">
<property name="fileSplitter" ref="fileSplitBean/>
</bean>
Class ReleaseStrategy{
FileSplitBean fileSplitter;
public Exchange aggregate(Exchange oldExchange, Exchange
newExchange) {
if(Boolean.True.equals(newExchange.getIn().getHeader(SPLIT_FINISHED)))
fileSplitter.rease();
}
}
But the solution still have some-things unresolved: 1:cannot invoke release
method while error occurs; 2:if split-pattern is nested, how to distinguish
the end-flag.
Finally , what I really want is give the split-patter another strategy:
release-strategy; but this solution maybe is too close to
special-application. And I doesn't know if it worth to be adopted by camel.
I illustrate it as following
<from uri="file://source?consumer.regexPattern="*\.dat">
<split stopOnException="true" releaseStrategyRef="releaseStrategy">
<method bean="fileSplitBean" method="split"/>
<bean ref="saveToDatabase">
</split>
<bean id="releaseStrategy" class="...">
<property name="fileSplitter" ref="fileSplitBean/>
</bean>
Class ReleaseStrategy{
FileSplitBean fileSplitter;
Public void release(Exchange exchange) {
//to do check if exchange failed
fileSplitter.release();
}
}
Willem Jiang wrote:
+1 for this requirement. Please feel free to create a JIRA[1] for it.
[1] https://issues.apache.org/activemq/browse/CAMEL
Claus Ibsen wrote:
> Hi
>
> No there is currently no such end marker, but I do think its possible
> to add such one in the Splitter.
>
> Could you create a ticket and attach a small sample / unit test etc.
> what you are doing?
>
>
> On Mon, Jan 11, 2010 at 7:26 AM, ext2 <[email protected]> wrote:
>> Hi:
>> By checking the camel's document, I just find a property tell the
>> index of message, but cannot find a property to indicate the end of
>> splitting.
>> But some-times, it will be very use-full for stream-based
splitter.
>> For example:
>> When I using stream-based splitter to process all the records
stored
>> in a large file sequence-ly. I will use a bean which return a iterator
which
>> read record sequence from the file(which store a lot of record ).After
>> processed all the record, I must close the file. So I need a flag to
>> indicate the end of record and close the file;
>>
>> If without such a property to indicate end of splitting, I must
>> write a very special iterator which give a "End-Flag" message to say :
"the
>> record is all over". But the program of iterator looks very odd;
>>
>> Does any-one knows about this? Or a better suggest to resolve
>> my-question?
>>
>> Thanks for any-suggestion;
>>
>>
>>
>>
>
>
>
