A recent conversation in the Developerworks ITM 6.x forum dealt with an unusual problem with Universal Agents.
A few rows were showing up cropped - only the first part of the line was making it's way from the UA to the TEMS/TEPS database.
The root of the problem turned out to be the UA's "unflushed buffer". What is a buffer and why should it be flushed?
A computer program, be it a simple "hello world" application or a complex missile control system, can often be summarized like this:
Get Input -> Do Something -> Write Output and repeat.
Now, things get complicated when we try to be more efficient. Say we've got a script which is checking a series of files/disks/servers/anything. If it wrote the result of each check immediately then the hard disk writing heads would constantly be starting and stopping - which is not efficient. What happens (behind the scenes) is that the operating system creates a Buffer of information which holds the lines temporarily. Once enough lines have been written into the buffer, the buffer is written (flushed) to the disk. This translates as much less starting and stopping of the disk heads. If you run the script yourself then you'll see everything displayed on the screen because the operating system will flush the buffer at the end of the script - at the latest.
Now, I don't know what the exact reason for the UA losing parts of lines, but I assume that the way the script UA works is reading the buffer that the script writes. If the UA reads the buffer BEFORE the operating system flushes the buffer then the UA will only get part of the information. This will not affect all UAs, but it might come and bite a few.
There is a full solution in the works, but a good workaround (for Perl) in the meantime is to add the following line to the beginning of the script:
BEGIN { $| = 1; }
This will make sure that the operating system flushes the buffer at the end of every output operation and that guarantees that the script and the UA are synchronized. If you're using other script languages, you'll have to find the equivalent function.
This also happened to me way-back-when when I was writing DCL scripts for the OpenVMS operating system. Just goes to show that what comes around, goes around! It also demonstrates that we Tivoli types need a good background in general computer science knowledge to help us solve fiddly little problems.
-- Robert
2 comments:
I was going crazy over this problem when I upgraded a script. I finally decided to revert to the old version.
Thanks for the explanation, even if it is a little overdone.
I would add some more tags to it.
Thanks for the feedback - what sort of tags do you mean?
To many sites either give solutions without explaining why, or give explanations without practical applications. I tried to balance the two, which is why it's a little... long.
Post a Comment