pkg-1.7.0 is an order of magnitude slower than pkg-1.6.4
trashcan at ellael.org
Fri Apr 8 20:22:43 UTC 2016
Baptiste Daroussin <bapt at FreeBSD.org> wrote:
> On Sat, Apr 02, 2016 at 02:59:06PM +0200, Michael Grimm wrote:
>> Baptiste Daroussin <bapt at FreeBSD.org> wrote:
>>> On Sat, Apr 02, 2016 at 02:42:06PM +0200, Michael Grimm wrote:
>>>> 26 seconds for 74 ports within a jail and pkg-1.6.4:
>>>> 309 seconds for the very same 74 ports within the very same jail and pkg-1.7.0:
>>>> Is this an expected slow-down? /usr/ports/UPGRADE and https://svnweb.freebsd.org/ports/head/ports-mgmt/pkg/?view=log are not indicating that behavior.
>>>> But I might have missed something.
>>>> Any feedback is highly appriciated, thanks, and regards,
>>> pkg 1.7 is IO intensive that may explain.
>> Ok, understood.
>> JFTR: perl (24s), python27 (44s), and ruby (125s) take the longest time to reinstall.
>>> I plan to readd some improvements on this side before 1.8
>> Good to know, thanks for your feedback.
> Just to follow up on the performance issue, there is a regression that happened
> on FreeBSD 10.3-RELEASE (also HEAD) that causes pkg extraction process to be 10
> times slower as it should. r297626 fixes it in head. We are working on bringing
> that into the 10 branch:
Thanks. I added the relevant patch  to 10.3-STABLE (r297721) and I can confirm that that fixed the performance issue.
Now, with pkg-1.7.2 for the very same 74 ports and jail reinstallation is back to 26 seconds:
Apr 8 20:04:42 <user.notice> pkg: indexinfo reinstalled: 0.2.4 -> 0.2.4
Apr 8 20:05:08 <user.notice> pkg: pkg reinstalled: 1.7.2 -> 1.7.2
Up to now, 10.3-STABLE runs rock solid. I will report if that might change.
Thanks for your engagement and with kind regards,
More information about the freebsd-ports