1 |
Peter Humphrey <prh@××××××××××.uk> posted |
2 |
200611231821.13640.prh@××××××××××.uk, excerpted below, on Thu, 23 Nov |
3 |
2006 18:21:13 +0000: |
4 |
|
5 |
> Does anyone here know of a signal I can pass to portage to make it stop at |
6 |
> the end of the current package? Quite often I find I'm emerging quite a lot |
7 |
> of packages, and I'd like to shut the machine down for the night and resume |
8 |
> in the morning. (I have been running it all night, but it's getting to be a |
9 |
> bit noisy in the fan department.) |
10 |
|
11 |
I don't believe there's a way to do it other than manually (watching for |
12 |
the right moment and stopping it then), but using the --resume switch will |
13 |
resume it at the same package, and if you use FEATURES=ccache and have it |
14 |
setup correctly, the compile to the point of interruption will have been |
15 |
cached so that package should merge faster, as well. It'll still take |
16 |
awhile as the configure steps repeat, and the linking and other |
17 |
non-compile steps repeat as well, but the actual compilation should be |
18 |
cached, so that part will be faster, on the package it's doing twice. |
19 |
|
20 |
Of course, there are some I'd still not want to interrupt. a gcc compile, |
21 |
or glibc, come to mind, as does kdelibs if you have it merged, and from |
22 |
what I read, openoffice, unless they are just barely started. However, an |
23 |
emerge --pretend or --ask can give you an idea of if and where they are in |
24 |
the list, and you can schedule accordingly. |
25 |
|
26 |
Alternatively, do a pretend, to get a list, and then emerge individual |
27 |
packages. I often do this with --tree, to get an idea of the |
28 |
dependencies, and then do several merges in parallel (such that there |
29 |
aren't any conflicting dependencies, thus the --tree), since portage |
30 |
otherwise doesn't make very efficient use of multiple CPUs for much of the |
31 |
emerge run. The configure step for instance is generally single-job |
32 |
serial, as are most of the other steps other than the actual compile. |
33 |
Even the compile is often forced to -j1 for individual packages due to job |
34 |
ordering issues if -jX were allowed, so the /only/ way I've found to |
35 |
efficiently make use of even TWO-way SMP is to invoke up to five emerges |
36 |
in parallel. The problem will be even worse once I upgrade to dual-core |
37 |
Opteron 285s, thus four-way SMP (I see prices are down to $1200-ish for |
38 |
the pair, now). |
39 |
|
40 |
-- |
41 |
Duncan - List replies preferred. No HTML msgs. |
42 |
"Every nonfree program has a lord, a master -- |
43 |
and if you use the program, he is your master." Richard Stallman |
44 |
|
45 |
-- |
46 |
gentoo-amd64@g.o mailing list |