pcp
[Top] [All Lists]

Re: [pcp] recent buld regression for inifinband PMDA on debian

To: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>, Nathan Scott <nathans@xxxxxxxxxx>
Subject: Re: [pcp] recent buld regression for inifinband PMDA on debian
From: Martins Innus <minnus@xxxxxxxxxxx>
Date: Thu, 03 Jul 2014 10:55:16 -0400
Cc: pcp@xxxxxxxxxxx
Delivered-to: pcp@xxxxxxxxxxx
In-reply-to: <53B48B8B.4090905@xxxxxxxxxxxxxxxx>
References: <006b01cf95ba$27501890$75f049b0$@internode.on.net> <853918576.1602669.1404285230879.JavaMail.zimbra@xxxxxxxxxx> <53B4052C.7010903@xxxxxxxxxxx> <53B48B8B.4090905@xxxxxxxxxxxxxxxx>
User-agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.9; rv:24.0) Gecko/20100101 Thunderbird/24.6.0
Ken,
Thanks for adding that in. As long as both builds completed with no errors, it looks good to me.

On 7/2/14, 6:45 PM, Ken McDonell wrote:
Debian (where it was failing) ...

src/include/pcp/platform_header.h:/* #undef HAVE_PORT_PERFORMANCE_QUERY_VIA */
src/include/pcp/config.h:#define HAVE_PORT_PERFORMANCE_QUERY_VIA 1
src/include/pcp/config.h:/* #undef HAVE_PMA_QUERY_VIA */
src/include/pcp/config.h.in:#undef HAVE_PORT_PERFORMANCE_QUERY_VIA
src/include/pcp/config.h.in:#undef HAVE_PMA_QUERY_VIA
kenj@bozo-vm:~/src/pcp$ cd src/pmdas/infiniband/
kenj@bozo-vm:~/src/pcp/src/pmdas/infiniband$ make clean; make
rm -rf domain.h *.o ib.log pmdaib dep dep.bak *.[1-9].gz *.[1-9].bz2 
*.[1-9].lzma *.[1-9].xz *.[1-9].tmp   ib.o pmda.o     pmdaib
for d in `echo  `; do if test -d "$d" -a -f "$d/GNUmakefile"; then /bin/echo 
=== $d ===; make --no-print-directory -C $d clean || exit $?; fi; done
gcc -fPIC -fno-strict-aliasing -D_GNU_SOURCE -fstack-protector-all -D_FORTIFY_SOURCE=2  
-Wall -O2 -g -DPCP_DEBUG -DPCP_VERSION=\"3.9.7\" -I../../../src/include 
-I../../../src/include/pcp   -c -o ib.o ib.c
gcc -fPIC -fno-strict-aliasing -D_GNU_SOURCE -fstack-protector-all -D_FORTIFY_SOURCE=2  
-Wall -O2 -g -DPCP_DEBUG -DPCP_VERSION=\"3.9.7\" -I../../../src/include 
-I../../../src/include/pcp   -c -o pmda.o pmda.c
gcc -fPIC -fno-strict-aliasing -D_GNU_SOURCE -fstack-protector-all -D_FORTIFY_SOURCE=2  
-Wall -O2 -g -DPCP_DEBUG -DPCP_VERSION=\"3.9.7\" -I../../../src/include 
-I../../../src/include/pcp -o pmdaib  -Wall -L../../../src/libpcp/src 
-L../../../src/libpcp_pmda/src   ib.o pmda.o     -libmad -libumad 
-L../../../src/libpcp/src -L../../../src/libpcp_pmda/src -lpcp_pmda -lpcp
This Debian version has the old interface, which doesn't allow querying of switch metrics.
Ubuntu (where the build was OK) ...

src/include/pcp/config.h.in:#undef HAVE_PORT_PERFORMANCE_QUERY_VIA
src/include/pcp/config.h.in:#undef HAVE_PMA_QUERY_VIA
src/include/pcp/config.h:/* #undef HAVE_PORT_PERFORMANCE_QUERY_VIA */
src/include/pcp/config.h:#define HAVE_PMA_QUERY_VIA 1
kenj@bozo:~/src/pcp$ cd src/pmdas/infiniband
kenj@bozo:~/src/pcp/src/pmdas/infiniband$ make clean; make
rm -rf domain.h *.o ib.log pmdaib dep dep.bak *.[1-9].gz *.[1-9].bz2 
*.[1-9].lzma *.[1-9].xz *.[1-9].tmp   ib.o pmda.o     pmdaib
for d in `echo  `; do if test -d "$d" -a -f "$d/GNUmakefile"; then /bin/echo 
=== $d ===; make --no-print-directory -C $d clean || exit $?; fi; done
gcc -fPIC -fno-strict-aliasing -D_GNU_SOURCE -fstack-protector-all -D_FORTIFY_SOURCE=2  
-Wall -O2 -g -DPCP_DEBUG -DPCP_VERSION=\"3.9.7\" -I../../../src/include 
-I../../../src/include/pcp   -c -o ib.o ib.c
gcc -fPIC -fno-strict-aliasing -D_GNU_SOURCE -fstack-protector-all -D_FORTIFY_SOURCE=2  
-Wall -O2 -g -DPCP_DEBUG -DPCP_VERSION=\"3.9.7\" -I../../../src/include 
-I../../../src/include/pcp   -c -o pmda.o pmda.c
gcc -fPIC -fno-strict-aliasing -D_GNU_SOURCE -fstack-protector-all -D_FORTIFY_SOURCE=2  
-Wall -O2 -g -DPCP_DEBUG -DPCP_VERSION=\"3.9.7\" -I../../../src/include 
-I../../../src/include/pcp -o pmdaib  -Wall -L../../../src/libpcp/src 
-L../../../src/libpcp_pmda/src   ib.o pmda.o     -libmad -libumad 
-L../../../src/libpcp/src -L../../../src/libpcp_pmda/src -lpcp_pmda -lpcp
The version in Ubuntu has the new interface which allows both port and switch metrics.

So something bad is happening in the configure logic ... Debian looks sort of 
right to me, but Ubuntu looks sort of busted because 
HAVE_PORT_PERFORMANCE_QUERY_VIA is NOT set, but (I assume the later/subset 
control) HAVE_PMA_QUERY_VIA is set.

I'll leave this to Martins to sort out ... I don't have hardware to test the 
PMDA, so I can't help beyond build issues.

Only one of those should be set, not both, as it marks a change in interface to the mad library calls. I am rebuilding now and will test, but may not finish before I head out for the holiday weekend.

Martins

<Prev in Thread] Current Thread [Next in Thread>