Re: ZODB eating CPU but RAM is free

On Wed, Jul 19, 2017, at 18:53, Jim Fulton wrote:
> On Wed, Jul 19, 2017 at 7:28 AM, Hanno Schlichting <[email protected]> wrote:__>> I'd still consider it to be experimental. I did my last tests with it 8 or so years ago and it didn't work reliably back than.> 
> How so?

This was 8 years ago, so I'm spotty on the details.

IIRC I tried this on a couple different Plone sites back than. After some fairly short time of normal usage (maybe a couple hours) the ZODB connection cache managed to report a negative size. After that the site became unusable. One way to get there earlier was to do frequent manual connection cache GC or minimize.
As a second issue the target cache byte value was almost uncorrelated to the actual memory used on the sites.
Back than I concluded that the way the size estimation works wasn't really good enough yet. I think ZODB/Connection uses the byte length of the pickle of a persistent object as an estimated size. But at least back than in a Plone site that wasn't a good way to predict memory usage of unghosted objects.
So you could either put in a magic count like 30000 and figure out through monitoring what kind of actual memory usage that would result in for your application. Or you could put in 100mb and use monitoring to figure out what the unknown multiplier was to get the actual memory usage.
Since in either case you had an unknown multiplier to figure out, there wasn't really any advantage of using the less proven byte based cache size rather than the traditional count based one.
And of course I was bad and didn't try to distill this further or make bug reports about it back than, sorry!
Hanno

-- 
You received this message because you are subscribed to the Google Groups "zodb" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
For more options, visit https://groups.google.com/d/optout.