make -j failure on MacPro....

Question

Level 1

20 points

make -j failure on MacPro....

Hi,

I've many failures running large parrallel builds on my 4 core MacPro. For example:

$ make -j 6
...
cp treeplayer/inc/TTreeFormulaManager.h include/TTreeFormulaManager.h
cp treeplayer/inc/TTreeIndex.h include/TTreeIndex.h
cp treeplayer/inc/TTreePlayer.h include/TTreePlayer.h
cp treeplayer/inc/TTreeProxyGenerator.h include/TTreeProxyGenerator.h
cp treeviewer/inc/TTreeViewer.h include/TTreeViewer.h
cp treeviewer/inc/TTVSession.h include/TTVSession.h
make: cp: Command not found
make: * [include/TTVSession.h] Error 127
make: * Waiting for unfinished jobs....

real 0m1.447s
user 0m0.505s
sys 0m2.086s

How can MacOS X suddenly not find the cp command anymore. Also often the compiler exits with BTP and bus errors.

After a reboot things work a gain for a while, but especially after interrupting a make -j (ctrl-c) things go quickly wrong and I need to reboot. Is this a kown locking problem?

The same makefile build perfectly fine in parallel on any other platform (Linux, Solaris, etc).

Cheers, Fons.

MacPro, Mac OS X (10.4.7)

Posted on Sep 14, 2006 7:31 AM

Reply

Answer 1

salobaas Author

Level 1

20 points

Sep 14, 2006 10:16 AM in response to salobaas

Some more info, during compilation with -j 2 I get:

********

Host Name: proof
Date/Time: 2006-09-14 19:11:06.072 +0200
OS Version: 10.4.7 (Build 8K1079)
Report Version: 4

Command: make
Path: /usr/bin/make
Parent: bash [25842]

Version: ??? (???)

PID: 26414
Thread: 0

Exception: EXC BADACCESS (0x0001)
Codes: KERN INVALIDADDRESS (0x0001) at 0x78632e77

Thread 0 Crashed:
0 make 0x00015462 0x1000 + 83042
1 make 0x000175aa 0x1000 + 91562
2 make 0x00016613 0x1000 + 87571
3 make 0x0000a417 0x1000 + 37911
4 make 0x0000ad32 0x1000 + 40242
5 make 0x0000b145 0x1000 + 41285
6 make 0x00013f63 0x1000 + 77667
7 make 0x0001302f 0x1000 + 73775
8 make 0x00013792 0x1000 + 75666
9 make 0x0001302f 0x1000 + 73775
10 make 0x00013792 0x1000 + 75666
11 make 0x0001302f 0x1000 + 73775
12 make 0x00013792 0x1000 + 75666
13 make 0x0001302f 0x1000 + 73775
14 make 0x00013792 0x1000 + 75666
15 make 0x0001426c 0x1000 + 78444
16 make 0x0000db4a 0x1000 + 52042
17 make 0x00001772 0x1000 + 1906
18 make 0x00001699 0x1000 + 1689

Thread 0 crashed with X86 Thread State (32-bit):
eax: 0x00000006 ebx: 0x00301000 ecx: 0x78632e73 edx: 0x00000006
edi: 0xbfffd764 esi: 0x01d17310 ebp: 0xbfffd6c8 esp: 0xbfffd6b0
ss: 0x0000001f efl: 0x00010282 eip: 0x00015462 cs: 0x00000017
ds: 0x0000001f es: 0x0000001f fs: 0x00000000 gs: 0x00000037

Binary Images Description:
0x1000 - 0x1efff make /usr/bin/make
0x8fe00000 - 0x8fe49fff dyld 46.8 /usr/lib/dyld
0x90000000 - 0x9016ffff libSystem.B.dylib /usr/lib/libSystem.B.dylib
0x901bf000 - 0x901c1fff libmathCommon.A.dylib /usr/lib/system/libmathCommon.A.dylib
0x90bce000 - 0x90bd5fff libgcc_s.1.dylib /usr/lib/libgcc_s.1.dylib

But this is not reproducable. Sometimes it happens sometimes not.

Very worrisome.

-- Fons

Reply

Answer 2

Sep 15, 2006 11:18 PM in response to salobaas

make: cp: Command not found

Hi.

After you get that error, try to run the cp command manually from the command line. What did you get?
Also, did you have fink installed?

Mihalis.

Reply

Answer 3

salobaas Author

Level 1

20 points

Sep 16, 2006 2:11 AM in response to Mihalis Tsoukalos

Hi,

as far as I can tell now the problem turned out to be due to unsufficient ulimit settings. Using a parallel build with more than 6 compiles in parallel and using the -pipe gcc option caused too many open files. Setting: ulimit -n 2048 fixed the problem (MacOS X default is 256, while linux default is 1024).

Cheers, Fons.

Reply