On Thu, 2011-02-17 at 13:39 +0200, Panu Matilainen wrote:
> On 02/17/2011 12:15 PM, Panu Matilainen wrote:
> > At least hasattr() calls for _cache_prco_names_* and _prco_lookup
> > were causing headers to be loaded when not wanted at all, causing
> > headers for all to-be updated packages to be kept in memory throughout
> > the entire transaction.
> 
> Just FYI, this isn't the only problem of this kind. For example the 
> changelog plugin, despite trying to conserve memory by only storing 
> changelog times, has the same effect of keeping headers of all to-be 
> updated packages loaded in memory.
> 
> Dunno yet what to do about it though...

 Yeh, we still have the same problem with summary/description (although
we just load it all the time atm). In theory callers can call _get_hdr()
directly, the problem is that after a couple of rpmdb lookups it's
noticeable CPU wise so we get hurt that way.
 I'd thought a couple of times of trying to have a "these are the
attributes I'm interested in" type of API, which could work for sqlite
too. But it's kind of hard to use and feels like such a hack (and the
improvements are only noticeable "relatively").

_______________________________________________
Yum-devel mailing list
[email protected]
http://lists.baseurl.org/mailman/listinfo/yum-devel

Reply via email to