Summary: | [PATCH] portage 2.2 uses ***way*** too much VM | ||
---|---|---|---|
Product: | Portage Development | Reporter: | James Cloos <cloos> |
Component: | Core | Assignee: | Portage team <dev-portage> |
Status: | RESOLVED OBSOLETE | ||
Severity: | normal | CC: | dschridde+gentoobugs, kingjon3377, tiago |
Priority: | High | ||
Version: | 2.2 | ||
Hardware: | All | ||
OS: | Linux | ||
Whiteboard: | |||
Package list: | Runtime testing required: | --- | |
Bug Depends on: | |||
Bug Blocks: | 240320 | ||
Attachments: | break Package -> depgraph references for proper garbage collection |
Description
James Cloos
2008-06-23 15:59:20 UTC
(In reply to comment #0) > But 2.2 is unusable. It allocates well over 200 megs durring emerges, and > even 175 for a simple emerge -f --nodeps run. You neglected to post emerge --info so I'm not sure shich version you're referring to. The patch from bug 228075 should help, and it's included in 2.2_rc1. For comparison, I'm seeing approx. 120 megs with current svn. portageq match / "" | wc -l 2130 That's 2130 installed packages. grep ^PACKAGES "$(portageq envvar PKGDIR)"/Packages PACKAGES: 3384 That's 3384 binary packages. How do your counts compare? If you have too many binary packages then you may want to use eclean (from the gentoolkit package) to remove some of them. There is only one version of 2.2 in the tree as of my last sync (Sunday), 2.2_rc1. I don't have a Packages directory under PKGDIR; ${PKGDIR}/All has 9 tbz2 files. prtageq match / ""|wc -l says 5647. Turns out I had to go down to 2.1.4.4 to get back to only 75 megs of heap (according to /proc/$PID/smaps). I thought 2.1.5.2 was in the 70ish range, but 2.1.5.6 and 2.1.5.5 were as bad as 2.2_rc1. Created attachment 158349 [details, diff]
break Package -> depgraph references for proper garbage collection
This patch should help a lot. I'll keep looking to see if I can find any other obvious fixes.
The memory consumption should be better in 2.2_rc3, so please try again and see how we're doing. VM usage is still way too high. ps(1) says VM is 190656, 1030 major faults, 2:35.61 of CPU time (up to the prompt) for emerge -av --nodeps app-text/chm2pdf. That is actually more VM than I saw for a --nodeps run when I first reported the bug. Has this bug been solved? I use portage 2.1.6.7 and have not tested unstable portage, but portage is already quite slow and I want 2.2 to be an improvement in this aspect. That is why I voted for this bug. By the way, can I vote for bugs that are important to me but that I have not reproduced on my system (such as this one) or will this confuse developers as if I was implicitly saying I had reproduced the bug? Thank you for your time and contribution, Jorge Let's call this obsolete. If it's still happening (and I've not seen anything like this), we need a new bug with new info. |