> -----Oorspronkelijk bericht-----
> Van: [EMAIL PROTECTED]
> [mailto:[EMAIL PROTECTED]]Namens Johannes Verelst
> Verzonden: woensdag 18 september 2002 16:31
> Aan: [EMAIL PROTECTED]
> Onderwerp: Bugfix for MMFlash
>
>
> Hi,
>
> This afternoon I found a rather critical performance bug in the flash
> generator in MMbase. The idea is as follows:
>
> 1. User does a request for a flash-generated page
> 2. Servflash.request() is called
> 3. MMFlash module is called to do all the work
> 4. Servflash.request() returns data and exists
>
> The problem: the request() method is synchronized, meaning that only ONE
> thread can generate flash, all others must wait. With one of our sites
> this had the effect that if one client does a difficult search (taking
> 1.5 minutes), the entire site was unreachable during that time. When
> somebody requested any page, the flash generator was triggered, but
> since there already was a thread in the Servflash.request() method, he
> had to wait.
>
> I have fixed this the following way:
> - nothing in servflash and mmflash is synchronized anymore
> - mmflash used to always use the same filename (temppath +
> "import.sws"), this filename is now randomized so that concurrent
> instances of MMFlash do not interfere with eachother.
>
> I have attached the unified-diff for this problem. I have also fixed
> several other issues in MMFlash:
> - general code cleanup (diff doesn't show since I diffed with -w) to
> comply with coding standards
> - comments for all methods
> - several instances of '/' were changed into 'File.separator'
> - better checking on the specified value of 'generatortemppath'
>
> The one issue I haven't implemented yet is cleanup of all temporary
> data, I will add this before checking it in. I also want to check it in
> into the MMBase 1.5 branch since we are going to run this version of
> mmbase for the website in question. Since it is a performance bugfix, I
> think it is allowed to be checked in, but I wanted to make that sure
> before committing my files.
>
> If nobody objects, I'll check it in somewhere next week.
>
> Johannes
>
>
if the delete is implemented its fine by me... tnx for the fix.
Daniel.
