I think using filter is not possible since I cannot detect the end of the transmission. An implementation of IdempotentRepository should work because
- contains() works like filter.accept(): directories and files are handed over but since they are all strings I have to know the structure to distinguish between them. Anyway I can count (or even remember) the number of files
- add() only receives the files, not the directories. I know that I am finished when I receive al files I accepted in contains().
Anyway I am still astound that there is no better solution or that it is not possible to receive the end of the ftp polling.
Kind regards, Christian
Re: AW: IdempotentRepository
For repository contains() is called for directories and files, while add() is called only for files.
But it is much easier to use filter instead of repository as you recommended, so I switched to filter.
However the original problem is still there: I mark directories as processed, since these contain the date-information and I do this when I switch to a new directory: because the input is sorted I know that the old directory is handled and I mark it as processed.
But when I read the last directory I don't receive a "it is over" message inside the filter, so the next time the route is executed, it copies the last directory again and I would like to solve this.
Von: Claus Ibsen
Gesendet: Freitag, 06. Juli 2012 15:04
Betreff: Re: IdempotentRepository
You can use a filter, and then some custom logic (eg a pojo) and return true|false to accept the file.
Then you dont need the idempotent pattern.
And from Camel 2.10 onwards the filter is now also invoked for directories. So you can skip traversing down in directories you do not want to.
On Fri, Jul 6, 2012 at 8:17 AM, Christian Lipp wrote:
I am fetching files from a FTP server (severall GB for the next
years). The files are produced daily in directories which correspond
to the date, like
I have only read rights and I am not the only consumer. This means that they keep the last month or so on the server and I fetch on a daily base. To avoid that I am fetching files twice I want to use an IdempotentRepository implementation. I don't want to save each file in a database or in a text file because the service will run for years and this is just unnecessary data.
What I want to store is the last processed date only. This handles just the directories and would mean that I need some other strategy for the files. I could combine this approach with the default in memory based store. But let just stick to the directories:
I read the directory sorted by file name. The IdempotentRepository is
called by the FtpConsumer with
- contains() for every directory and file
- add() for files only
and that's it. No stop(), no confirm(). When I have errors, sometimes remove() is called. Since the repository is called only with String (the full path) I have not information if I deal with directories or files. I know it from the structure, but I am not able to implement a generic solution.
Anyway the idea is:
- Store the LastProcessedDate inside the repository
- contains(): if the path contains an already processed date (<LastProcessedDate) then I skip it (return true) otherwise return false.
- add(): if add() jumps to the next directory I set the
LastProcessedDate to the directory before
The only problem is the last processed directory: even if it is finished I do not get the chance to mark it as processed (set LastProcessedDate to its value).
So finally my questions: do you think this approach makes sense and if yes: how would you deal with the last processed directory?
If no, how would you solve it?
Thanks and kind regards, Christian
Twitter: davsclaus, fusenews
Author of Camel in Action: http://www.manning.com/ibsen