[OpenIndiana-discuss] Questions about /var/pkg

[Matthew R. Trower]

On 1/13/24 10:55, Alan Coopersmith wrote:

> Someone looking to improve that might want to check out the recent 
> performance improving commits to 
> https://github.com/oracle/solaris-ips/commits/master/ and see if
> they cherry-pick to OI's fork.
> 
> -alan-

Thanks for pointing that out, Alan.  My OI machines are getting rather
long in the tooth and I'm dearly interested in any IPS performance
improvements to be had.


On 1/13/24 18:36, Joshua M. Clulow via openindiana-discuss wrote:
> The original question that started the thread is obviously very
> reasonable: in theory much of the bulk data that IPS stores should
> not need to be tied to a particular boot environment -- provided it
> is stored in a long-term stable format, or it can reliably be
> destroyed and recreated from upstream sources when that format needs
> to change.
> 
> I don't think it's ever going to be wise to try and mount datasets 
> from outside the boot environment on top of subdirectories for 
> arbitrary system components, IPS included.  Instead, if you would
> like it to work that way, it would be best to get IPS to directly
> support doing this.

I would go further and say the bulk of the issue is just in the cached 
files.  It should be possible for IPS to cache its files to some 
user-directed central location.  At a glance, the files appear to be 
stored according to their hash, so there shouldn't really be an issue 
with this.  Of course, I haven't looked; nothing is certain until I do. 
Maybe some IPS fork has even done this already.


On 1/13/24 18:29, Goetz T. Fischer wrote:
> i would assume it's eating most of it because on an old machine, the assigned cpu core is at 100% for 
> several minutes while the rotating slash keeps, well, rotating.
> disk writes wouldn't cause 100% cpu. in fact, having a slow disk brings the cpu load down if the disk 
> can't keep up.

I have my own thoughts, but I'm staying out of the speculation until I 
have something to show for them.  If it makes you feel any better, I 
understand your frustration.  Believe me, nobody is more frustrated than 
me when pkg pegs a core on my T5240 for 10 straight minutes, while the 
other 127 cores sit idle...

I imagine I'll end up having to look at this sooner rather than later, 
as pkg is central to a lot of the work I'm trying to do right now.



-- Matthew R. Trower