[Yum-devel] Current yum git desolver is a little slow

[Florian Festi]

Tim Lauridsen wrote:
> Florian Festi wrote:
>> Tim Lauridsen wrote:
>>> I did some basic testing of the current yum git HEAD vs. yum 3.2.2.
>>
>> Yes, current git head is a bit slower than 3.2.2. There is a simple 
>> reason for that: Linear searches on disk are even slower than linear 
>> searches in memory.
>>
>> In other words: We are still missing some indexes in the sqlite 
>> database to really gain speed from the changes. For performance runs 
>> you can temporary remove patch 
>> 3c2621ba8f8070f24ad3e979f6bd1699f6f6b394 or readd 
>> 42283902f929ac131cda7b3497ae047b497e02bc.
>>
>> There are two possible permanent solutions:
>>
>> Readd the patch mentions above and extend it that all indexes are 
>> created if they are not present yet. I posted timings for creating 
>> these indexes some weeks ago (2. Aug "Sqlite DB Indexes").
>>
>> As alternative or in addition we could patch yum-metadata-parser to 
>> just add the needed indexes.
>>
>> I can provide patches for both as soon as there is a decision with 
>> way(s) we want to go.
>>
>>
> I added this
> http://devel.linux.duke.edu/gitweb/?p=yum.git;a=commitdiff;h=42283902f929ac131cda7b3497ae047b497e02bc 
> 
> It speed up things a lot, but makes yum load the metadata every time, 
> because the index creation changes the checksum i think.
> I we want to create the indexes locally, how do we know when to download 
> the sqlite tarball from the repo ?

Hmpf, one shouldn't test with local mirrors only...

Ehm, don't we have the checksum of the XML file in the sqlite database? That 
might be a better thing to check against. The fact that sqlite doesn't alter 
the db files during reads is probably not something we should rely on.

I guess I have to take a deeper look into that topic and come up with a patch...

Thanks for testing and pointing that out!

Florian