[Bug 268760] lang/gcc12 build renders system unresponsive (12-core)
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Thu, 05 Jan 2023 03:38:32 UTC
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=268760
Bug ID: 268760
Summary: lang/gcc12 build renders system unresponsive (12-core)
Product: Ports & Packages
Version: Latest
Hardware: amd64
OS: Any
Status: New
Severity: Affects Only Me
Priority: ---
Component: Individual Port(s)
Assignee: salvadore@freebsd.org
Reporter: pmc@citylink.dinoex.sub.org
Flags: maintainer-feedback?(salvadore@freebsd.org)
Assignee: salvadore@freebsd.org
hw.ncpu: 12
hw.physmem: 19427618816
last pid: 16312; load averages: 105.17, 100.93, 76.14 up 0+02:23:05 21:18:12
218 processes: 144 running, 55 sleeping, 10 zombie, 9 lock
CPU: 0.0% user, 0.0% nice, 97.6% system, 2.4% interrupt, 0.0% idle
Mem: 9017M Active, 3139M Inact, 939M Laundry, 4536M Wired, 1412M Buf, 389M Free
ARC: 613M Total, 288M MFU, 125M MRU, 128K Anon, 3744K Header, 195M Other
313M Compressed, 504M Uncompressed, 1.61:1 Ratio
Swap: 20G Total, 20G Free
PID USERNAME THR PRI NICE SIZE RES STATE C TIME WCPU COMMAND
16221 root 1 40 0 1434M 1134M *bufob 8 1:59 18.15% lto1
16223 root 1 44 0 1434M 1134M *bufob 7 2:34 17.24% lto1
16212 root 1 49 0 1413M 1024M CPU11 11 2:42 15.81% lto1
16266 root 1 41 0 1478M 1156M *bufob 11 2:14 14.60% lto1
16215 root 1 38 0 1413M 1023M *bufob 4 2:34 14.41% lto1
16194 root 1 47 0 1413M 1029M RUN 4 3:04 14.41% lto1
16191 root 1 41 0 1418M 1040M *bufob 5 2:26 14.18% lto1
16210 root 1 44 0 1413M 1024M *bufob 9 2:13 14.16% lto1
16179 root 1 41 0 1413M 1030M *bufob 2 2:50 14.14% lto1
16205 root 1 41 0 1413M 1023M *bufob 3 2:30 13.90% lto1
16224 root 1 36 0 1434M 1133M *bufob 6 2:23 13.86% lto1
16181 root 1 46 0 1413M 1031M *bufob 11 2:21 13.72% lto1
16257 root 1 41 0 1478M 1158M *bufob 10 2:38 13.68% lto1
# ps ax | grep lto | wc ## and see the insane loadavg above!
153 1070 8638
It is also necessary to twiddle the OOM to get this running at all:
vm.pageout_oom_seq=5120
vm.oom_pf_secs=180
In comparison, building on only 8 cores does not show this pathologic behaviour
(both instances were run simultaneously on the same platform):
hw.ncpu: 8
hw.physmem: 14239264768
last pid: 31114; load averages: 9.94, 10.19, 10.00 up 0+02:46:40 21:42:02
73 processes: 9 running, 64 sleeping
CPU: 97.9% user, 0.0% nice, 1.6% system, 0.4% interrupt, 0.0% idle
Mem: 792M Active, 5544M Inact, 1675M Laundry, 3088M Wired, 1248M Buf, 2106M
Free
ARC: 859M Total, 529M MFU, 132M MRU, 128K Anon, 4725K Header, 193M Other
558M Compressed, 1107M Uncompressed, 1.99:1 Ratio
Swap: 20G Total, 447M Used, 20G Free, 2% Inuse
PID USERNAME THR PRI NICE SIZE RES STATE C TIME WCPU COMMAND
31108 root 1 102 0 190M 154M CPU6 6 0:34 103.85% lto1
31097 root 1 103 0 241M 199M CPU5 5 0:50 100.30% lto1
31066 root 1 103 0 285M 246M CPU7 7 2:07 100.28% lto1
31091 root 1 103 0 206M 170M CPU0 0 1:03 99.91% lto1
31119 root 1 88 0 152M 115M RUN 1 0:08 98.78% lto1
31103 root 1 102 0 211M 173M CPU4 4 0:39 97.62% lto1
31086 root 1 102 0 290M 251M CPU3 3 1:04 97.25% lto1
31114 root 1 100 0 149M 112M CPU2 2 0:26 97.02% lto1
31109 root 1 20 0 14M 3396K CPU1 1 0:01 0.98% top
638 ntpd 1 20 0 22M 1444K select 5 0:01 0.01% ntpd
# ps ax | grep lto | wc
9 62 434
The build is:
ports: ba2be1158c403a13cab32b446555e8d8b59cbc57 (2023Q1)
src: 12core 753d65a19a5541d532aff3cdd1089246fc5d2b8b (releng/13.1)
8core d290399b6ce99c99cb1b9d1cdadc605172723561 (stable/13)
Over all compute usage (the build takes longer on 12-core and consumes almost
twice as much cycles):
ncpu 12
real 22911.77
user 158470.70
sys 141767.89
2022732 maximum resident set size
29106 average shared memory size
1710 average unshared data size
232 average unshared stack size
275316289 page reclaims
106509 page faults
0 swaps
24970 block input operations
1587767 block output operations
351720 messages sent
461428 messages received
11651 signals received
3436058 voluntary context switches
6330784 involuntary context switches
ncpu 8
real 18661.43
user 150203.79
sys 12814.11
2082684 maximum resident set size
28908 average shared memory size
1797 average unshared data size
267 average unshared stack size
277207835 page reclaims
118424 page faults
112 swaps
41921 block input operations
1597336 block output operations
348368 messages sent
457959 messages received
12200 signals received
2049506 voluntary context switches
5806408 involuntary context switches
Besides, I fail to understand what this activity might have to do with
bootstrapping.
(From the book: Bootstrapping a compiler means to recompile the compiler with
itself after having it built with the available system compiler. It is
necessary to ensure that the compiler is not malfunctioning, whenever the
available system compiler is a different version than the build target.)
Furthermore, I also think that now having to build 8 (eight!) different
compilers in the process of a rather simple desktop deploy is quite insane -
but that's not directly related to this issue.
--
You are receiving this mail because:
You are the assignee for the bug.