Yan Zhu wrote: > Ok, here we go. > > I rebooted the machine and reclaimed all my memory. > And I ran test.pl on a db that has buffered logging. > > I let 50K records passed before I stopped the program. > > according to test.pl, the size stablized to 827 very quickly and stayed there > till the end. > > # 8 R 0 2433 1645 12 59 20 ? 827 pts/1 1:03 test.pl > > However, according to the top command, I lost 20 megs, I watched it > closely as the test.pl executes, it's losing the memory very evenly, in fact, > it loses one meg almost at the exact time when the PS info prints out, but > not all the time. So 20 meg slowly leaks out after 50K records updated. > > now what?
First - please include DBD:Informix in the subject line (as requested in the documentation); I fed up tracking random messages that are actually related to a problem in DBD::Informix. Second: which process is gobbling the memory? Does 'top' say that it is the Perl process that is chewing up the memory? If so, we have a contradiction; the 'ps' information says that Perl stabilizes, the 'top' information says it doesn't. If (a big IF) that's the case, then at least one of the programs is reporting incorrectly, or its report is being interpreted incorrectly. If 'top' doesn't say that Perl is chewing up the memory, which process is chewing it up? I'd have to guess that it is the server (oninit processes), but you must deduce which it is, and be able to justify that deduction. You should be able to simply stop IDS (onmode -ky) and restart it (oninit) and see the memory jump all over the place if it is indeed oninit that is chewing up the memory. > Jonathan Leffler wrote: > > I should also have added "or maybe the test is not tickling the bug". It appears that the code is tickling the bug, but not reporting on the problem. Also, if the server is growing, it is most likely a server bug which should be dealt with through IBM/Informix tech support channels. However, we would need to remove Perl from the equation, probably by converting the Perl script into ESQL/C and demonstrating that the same effect occurs with pure ESQL/C code. -- Jonathan Leffler ([EMAIL PROTECTED], [EMAIL PROTECTED]) Guardian of DBD::Informix 1.00.PC2 -- see http://dbi.perl.org/ #include <disclaimer.h>
