1 |
On Tuesday, 30 October 2018 06:30:23 GMT Håkon Alstadheim wrote: |
2 |
> I'm having fun enabling "test" in FEATURES on my gentoo-desktop. One |
3 |
> interesting failure, that brings to mind build failures I have had in |
4 |
> the past: |
5 |
> |
6 |
> Building sys-apps/mlocate-0.26-r2, I get |
7 |
> |
8 |
> 43: updatedb: Very deep hierarchy FAILED |
9 |
> (updatedb.at:261) |
10 |
> |
11 |
> Trying to reproduce, as root I do "make check" in the work/mlocate-0.26/ |
12 |
> , and the test passes. |
13 |
> |
14 |
> 43: updatedb: Very deep hierarchy ok |
15 |
> |
16 |
> I'd really like to get to the bottom of this, as I believe it must have |
17 |
> the same root-cause as issues I have had compiling large packages such |
18 |
> as firefox. |
19 |
> |
20 |
> Re-running both the emerge and the make check, I get the same results. |
21 |
> emerge fails, make check succeeds. I made a local copy of the ebuild and |
22 |
> inserted a "ulimit -a" in pre_src_test. |
23 |
> |
24 |
> ulimit from root-shell: |
25 |
> |
26 |
> # ulimit -a |
27 |
> core file size (blocks, -c) unlimited |
28 |
> data seg size (kbytes, -d) unlimited |
29 |
> scheduling priority (-e) 0 |
30 |
> file size (blocks, -f) unlimited |
31 |
> pending signals (-i) 59958 |
32 |
> max locked memory (kbytes, -l) 16384 |
33 |
> max memory size (kbytes, -m) unlimited |
34 |
> open files (-n) 1024 |
35 |
> pipe size (512 bytes, -p) 8 |
36 |
> POSIX message queues (bytes, -q) 819200 |
37 |
> real-time priority (-r) 0 |
38 |
> stack size (kbytes, -s) 8192 |
39 |
> cpu time (seconds, -t) unlimited |
40 |
> max user processes (-u) 10000 |
41 |
> virtual memory (kbytes, -v) unlimited |
42 |
> file locks (-x) unlimited |
43 |
> |
44 |
> ulimit from emerge: |
45 |
> >>> Source compiled. |
46 |
> |
47 |
> core file size (blocks, -c) unlimited |
48 |
> data seg size (kbytes, -d) unlimited |
49 |
> scheduling priority (-e) 0 |
50 |
> file size (blocks, -f) unlimited |
51 |
> pending signals (-i) 59958 |
52 |
> max locked memory (kbytes, -l) 16384 |
53 |
> max memory size (kbytes, -m) unlimited |
54 |
> open files (-n) 1024 |
55 |
> pipe size (512 bytes, -p) 8 |
56 |
> POSIX message queues (bytes, -q) 819200 |
57 |
> real-time priority (-r) 0 |
58 |
> stack size (kbytes, -s) 9788 |
59 |
> cpu time (seconds, -t) unlimited |
60 |
> max user processes (-u) 10000 |
61 |
> virtual memory (kbytes, -v) unlimited |
62 |
> file locks (-x) unlimited |
63 |
> |
64 |
> >>> Test phase: sys-apps/mlocate-0.26-r2 |
65 |
> |
66 |
> I have plenty of space in my portage temp directory (/pt): |
67 |
> |
68 |
> # df -hT ./ |
69 |
> Filsystem Type Størrelse Brukt Tilgj. Bruk% Montert på |
70 |
> /dev/xvdc ext4 163G 8,0G 147G 6% /pt |
71 |
> |
72 |
> Portage temp is at /pt due to the earlier mentioned issues with firefox. |
73 |
> |
74 |
> At my wits end here. Anyone ? |
75 |
|
76 |
I have not looked or used the test FEATURES of portage, but have also noticed |
77 |
over time certain packages fail to build on machines with low RAM. As low |
78 |
here I consider anything less than 4G. It seems each thread is now consuming |
79 |
so much memory they cumulatively use up all RAM available and then start |
80 |
swapping endlessly until the compilation invariably fails. Increasingly more |
81 |
and more packages have been suffering from this, the last two I noticed are |
82 |
qtwebkit and qtwebengine. |
83 |
|
84 |
My solution has been to create a package.env file in which I specify MAKEOPTS |
85 |
limiting the number of jobs and average load for any of these packages which |
86 |
chew up all the RAM. |
87 |
-- |
88 |
Regards, |
89 |
Mick |