From: Frank da Cruz <email@example.com>
Organization: The Kermit Project - Columbia University
Date: 21 Jul 2004 19:29:06 GMT
Subject: Case Study #29: Transaction processing revisited
It is an increasingly common question: how to upload a file in such a way that another process won't try to process it before the upload is complete? This sort of operation -- in which a variety of clients upload transactions (files) to a central service for processing -- is commonly used in EDI applications, insurance claims, etc. You want to be sure that each transaction is processed exactly once; not zero, or two or more times, and that processing takes place only when the upload is complete and successful.
Discussions of transaction processing have been available on the Kermit website for some time:
The second one concerns FTP but the principles are the same -- the transaction processing example is about halfway down the document, after the introductory tutorial material.
In these documents we talk about uploading files to a temporary working directory and the moving each file to its final destination only after the upload is complete and successful.
In some situations, however, it is possible that multiple clients might try to upload different files having the same name to the same working directory at the same time, and although Kermit's file collision features can be used to prevent one file from destroying the other, it complicates the moving or renaming process.
Another approach is for the server to ensure that each incoming file has a unique name. This way, no collisions will ever take place and no renaming within the temporary working directory will be necessary. This can be done with the following command:
set receive rename-to xxx
where xxx is a template, as explained here:
Let's say the incoming filename is ABCD1234.FIL, and you want the new name to incorporate the original name and still have the ".FIL" extension. Here are some variables you can use:
\v(filename) = the name the file was sent was (e.g. ABCD1234.FIL). \v(ntime) = current local time, seconds since midnight (e.g. 45369). \v(ndate) = current date, numeric (e.g. 20040716). \v(pid) = Kermit's numeric process ID.
The date, time, and PID combined would produce a guaranteed unique filename because at any given moment, every process (including every Kermit receiver) has a unique process ID.
How to construct the new filename? Let's assume that the filename contains only one dot (.). First use string functions to separate the basename from the extension:
\fstripx(\v(filename) = ABCD1234 \flop(\v(filename) = FIL
Then concatenate with the date, time, and pid; you can use \flpad() to left-pad variable-length fields (like \v(ntime) or \v(pid)) if you want to make them fixed-length. So here is your renaming function:
(This is a single line, which I've broken for benefit of netnews. Rejoin by removing the hyphen and the linebreak and indentation.)
I've separated the fields with underscores but of course you don't have to do that, or you can use a different separator. The command would be:
set receive rename-to -
which you would execute before putting the Kermit into receive or server mode. With this command in effect, a file called ABCD1234.FIL might be received as ABCD1234_20040721_04739_632.FIL. Your SET FILE COLLISION setting should be irrelevant because you won't have any collisions.
If you also want to move the file to a different directory at the same time you rename it, you can include that in the string, e.g.:
set receive rename-to ../Ready/\fstripx(\v(filename))_\v(ndate)-
Note that we use forward slash (/) as the directory separator in both Unix (C-Kermit) and Windows (Kermit 95). In other operating systems such as VMS, use the appropriate syntax.
[ Top ] [ Previous ] [ Next ] [ Index ] [ C-Kermit ] [ Kermit 95 ] [ Kermit Home ]