They have made the web site read only while they are investigating a security incident. Until they enable writes, we won't know whether the tests are broken or it is the web site problem. Rajeev On Apr 12, 2011, at 2:11 PM, Pavan Balaji wrote:
http://www.mcs.anl.gov/research/projects/mpich2/nightly/old
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
Looks like they have enabled writes now. You are right. The MPD and SMPD tests are broken. Rajeev On Apr 12, 2011, at 2:19 PM, Rajeev Thakur wrote:
They have made the web site read only while they are investigating a security incident. Until they enable writes, we won't know whether the tests are broken or it is the web site problem.
Rajeev
On Apr 12, 2011, at 2:11 PM, Pavan Balaji wrote:
http://www.mcs.anl.gov/research/projects/mpich2/nightly/old
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
Given that: (1) we no longer have any maintainers for mpd, and (2) it has been deprecated through the 1.3 series, I propose that we remove MPD in 1.4. Thoughts? -- Pavan On 04/12/2011 02:21 PM, Rajeev Thakur wrote:
Looks like they have enabled writes now. You are right. The MPD and SMPD tests are broken.
Rajeev
On Apr 12, 2011, at 2:19 PM, Rajeev Thakur wrote:
They have made the web site read only while they are investigating a security incident. Until they enable writes, we won't know whether the tests are broken or it is the web site problem.
Rajeev
On Apr 12, 2011, at 2:11 PM, Pavan Balaji wrote:
http://www.mcs.anl.gov/research/projects/mpich2/nightly/old
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
I dislike working on MPD as much as the next guy, but is MPD actually broken? Or is something just a bit off in the way the tests are being run or the way that our make/install steps have changed? I suspect this is a simpler issue, although I haven't done any investigation. -Dave On Apr 12, 2011, at 3:37 PM CDT, Pavan Balaji wrote:
Given that: (1) we no longer have any maintainers for mpd, and (2) it has been deprecated through the 1.3 series, I propose that we remove MPD in 1.4. Thoughts?
-- Pavan
On 04/12/2011 02:21 PM, Rajeev Thakur wrote:
Looks like they have enabled writes now. You are right. The MPD and SMPD tests are broken.
Rajeev
On Apr 12, 2011, at 2:19 PM, Rajeev Thakur wrote:
They have made the web site read only while they are investigating a security incident. Until they enable writes, we won't know whether the tests are broken or it is the web site problem.
Rajeev
On Apr 12, 2011, at 2:11 PM, Pavan Balaji wrote:
http://www.mcs.anl.gov/research/projects/mpich2/nightly/old
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
On 04/12/2011 03:55 PM, Dave Goodell wrote:
I dislike working on MPD as much as the next guy, but is MPD actually broken? Or is something just a bit off in the way the tests are being run or the way that our make/install steps have changed? I suspect this is a simpler issue, although I haven't done any investigation.
I should have clarified. I meant, I don't want to spend the time to investigate it. But if someone else wants to, that's OK with me. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji
There should be helpful log messages in the output from "maint/checkbuilds", but I don't see anything that matches what I expect to see. It looks like there's some sketchy perl code that doesn't make a lot of sense: ----8<---- 93 $OUTFD = OUTFD; ... 980 sub RunMPD { 981 # We need to background the mpd. 982 # If there is already an mpd running (for us), we want to stop it 983 my $mpdlocalsock = "/tmp/mpd2.console_". $ENV{"LOGNAME"}; 984 my $rc = 0; 985 if (-S $mpdlocalsock) { 986 system "$instdir/bin/mpdallexit"; 987 $rc = $? >> 8; 988 if ($rc != 0) { 989 print $OUTFD "Error running $instdir/bin/mpdallexit before starting MPD ($mpdlocalsock exists)\n"; 990 } 991 } 992 #system "$instdir/bin/mpd -d"; 993 # Try to background mpd 994 system "$instdir/bin/mpd &"; 995 $rc = $? >> 8; 996 if ($rc != 0) { 997 print $OUTFD "Error running $instdir/bin/mpd -d\n"; 998 } 999 else { 1000 print $OUTFD "Started mpd in demon mode\n"; 1001 } 1002 # Wait for the mpd to start 1003 sleep 1; 1004 return; 1005 } ----8<---- Is that bareword initialization on line 93 some sort of long-broken Perl4-ism? Or was it left that way so that "debug logging" could be enabled by a carefully placed "open(OUTFD,...)" call? I suppose a cheesier version is to change the OUTFD bareword to an STDOUT bareword, which will have the intended effect when not under "use strict", although in a totally gross way. Bill, because it's an older perl script, I'm guessing you wrote this. Is the output intentionally disabled because it causes a problem in normal use? Is there any reason that I shouldn't just replace line 93 with this? ----8<---- my $OUTFD = *STDOUT{IO}; ----8<---- -Dave On Apr 12, 2011, at 3:55 PM CDT, Dave Goodell wrote:
I dislike working on MPD as much as the next guy, but is MPD actually broken? Or is something just a bit off in the way the tests are being run or the way that our make/install steps have changed? I suspect this is a simpler issue, although I haven't done any investigation.
-Dave
On Apr 12, 2011, at 3:37 PM CDT, Pavan Balaji wrote:
Given that: (1) we no longer have any maintainers for mpd, and (2) it has been deprecated through the 1.3 series, I propose that we remove MPD in 1.4. Thoughts?
-- Pavan
On 04/12/2011 02:21 PM, Rajeev Thakur wrote:
Looks like they have enabled writes now. You are right. The MPD and SMPD tests are broken.
Rajeev
On Apr 12, 2011, at 2:19 PM, Rajeev Thakur wrote:
They have made the web site read only while they are investigating a security incident. Until they enable writes, we won't know whether the tests are broken or it is the web site problem.
Rajeev
On Apr 12, 2011, at 2:11 PM, Pavan Balaji wrote:
http://www.mcs.anl.gov/research/projects/mpich2/nightly/old
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
The tests ran until Friday night. I suspect the problem is because of some commit that went in on Saturday about location of mpiexec and friends. The tests run if I configure and build with mpd manually, but not if I run the nightly testing script manually. Rajeev On Apr 12, 2011, at 5:04 PM, Dave Goodell wrote:
There should be helpful log messages in the output from "maint/checkbuilds", but I don't see anything that matches what I expect to see. It looks like there's some sketchy perl code that doesn't make a lot of sense:
----8<---- 93 $OUTFD = OUTFD; ... 980 sub RunMPD { 981 # We need to background the mpd. 982 # If there is already an mpd running (for us), we want to stop it 983 my $mpdlocalsock = "/tmp/mpd2.console_". $ENV{"LOGNAME"}; 984 my $rc = 0; 985 if (-S $mpdlocalsock) { 986 system "$instdir/bin/mpdallexit"; 987 $rc = $? >> 8; 988 if ($rc != 0) { 989 print $OUTFD "Error running $instdir/bin/mpdallexit before starting MPD ($mpdlocalsock exists)\n"; 990 } 991 } 992 #system "$instdir/bin/mpd -d"; 993 # Try to background mpd 994 system "$instdir/bin/mpd &"; 995 $rc = $? >> 8; 996 if ($rc != 0) { 997 print $OUTFD "Error running $instdir/bin/mpd -d\n"; 998 } 999 else { 1000 print $OUTFD "Started mpd in demon mode\n"; 1001 } 1002 # Wait for the mpd to start 1003 sleep 1; 1004 return; 1005 } ----8<----
Is that bareword initialization on line 93 some sort of long-broken Perl4-ism? Or was it left that way so that "debug logging" could be enabled by a carefully placed "open(OUTFD,...)" call? I suppose a cheesier version is to change the OUTFD bareword to an STDOUT bareword, which will have the intended effect when not under "use strict", although in a totally gross way.
Bill, because it's an older perl script, I'm guessing you wrote this. Is the output intentionally disabled because it causes a problem in normal use? Is there any reason that I shouldn't just replace line 93 with this?
----8<---- my $OUTFD = *STDOUT{IO}; ----8<----
-Dave
On Apr 12, 2011, at 3:55 PM CDT, Dave Goodell wrote:
I dislike working on MPD as much as the next guy, but is MPD actually broken? Or is something just a bit off in the way the tests are being run or the way that our make/install steps have changed? I suspect this is a simpler issue, although I haven't done any investigation.
-Dave
On Apr 12, 2011, at 3:37 PM CDT, Pavan Balaji wrote:
Given that: (1) we no longer have any maintainers for mpd, and (2) it has been deprecated through the 1.3 series, I propose that we remove MPD in 1.4. Thoughts?
-- Pavan
On 04/12/2011 02:21 PM, Rajeev Thakur wrote:
Looks like they have enabled writes now. You are right. The MPD and SMPD tests are broken.
Rajeev
On Apr 12, 2011, at 2:19 PM, Rajeev Thakur wrote:
They have made the web site read only while they are investigating a security incident. Until they enable writes, we won't know whether the tests are broken or it is the web site problem.
Rajeev
On Apr 12, 2011, at 2:11 PM, Pavan Balaji wrote:
http://www.mcs.anl.gov/research/projects/mpich2/nightly/old
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
I could not find the problem, so I have made the change you suggested in maint/checkbuilds.in. Let's see if we get more information out of the tests tonight. Rajeev On Apr 12, 2011, at 5:04 PM, Dave Goodell wrote:
There should be helpful log messages in the output from "maint/checkbuilds", but I don't see anything that matches what I expect to see. It looks like there's some sketchy perl code that doesn't make a lot of sense:
----8<---- 93 $OUTFD = OUTFD; ... 980 sub RunMPD { 981 # We need to background the mpd. 982 # If there is already an mpd running (for us), we want to stop it 983 my $mpdlocalsock = "/tmp/mpd2.console_". $ENV{"LOGNAME"}; 984 my $rc = 0; 985 if (-S $mpdlocalsock) { 986 system "$instdir/bin/mpdallexit"; 987 $rc = $? >> 8; 988 if ($rc != 0) { 989 print $OUTFD "Error running $instdir/bin/mpdallexit before starting MPD ($mpdlocalsock exists)\n"; 990 } 991 } 992 #system "$instdir/bin/mpd -d"; 993 # Try to background mpd 994 system "$instdir/bin/mpd &"; 995 $rc = $? >> 8; 996 if ($rc != 0) { 997 print $OUTFD "Error running $instdir/bin/mpd -d\n"; 998 } 999 else { 1000 print $OUTFD "Started mpd in demon mode\n"; 1001 } 1002 # Wait for the mpd to start 1003 sleep 1; 1004 return; 1005 } ----8<----
Is that bareword initialization on line 93 some sort of long-broken Perl4-ism? Or was it left that way so that "debug logging" could be enabled by a carefully placed "open(OUTFD,...)" call? I suppose a cheesier version is to change the OUTFD bareword to an STDOUT bareword, which will have the intended effect when not under "use strict", although in a totally gross way.
Bill, because it's an older perl script, I'm guessing you wrote this. Is the output intentionally disabled because it causes a problem in normal use? Is there any reason that I shouldn't just replace line 93 with this?
----8<---- my $OUTFD = *STDOUT{IO}; ----8<----
-Dave
On Apr 12, 2011, at 3:55 PM CDT, Dave Goodell wrote:
I dislike working on MPD as much as the next guy, but is MPD actually broken? Or is something just a bit off in the way the tests are being run or the way that our make/install steps have changed? I suspect this is a simpler issue, although I haven't done any investigation.
-Dave
On Apr 12, 2011, at 3:37 PM CDT, Pavan Balaji wrote:
Given that: (1) we no longer have any maintainers for mpd, and (2) it has been deprecated through the 1.3 series, I propose that we remove MPD in 1.4. Thoughts?
-- Pavan
On 04/12/2011 02:21 PM, Rajeev Thakur wrote:
Looks like they have enabled writes now. You are right. The MPD and SMPD tests are broken.
Rajeev
On Apr 12, 2011, at 2:19 PM, Rajeev Thakur wrote:
They have made the web site read only while they are investigating a security incident. Until they enable writes, we won't know whether the tests are broken or it is the web site problem.
Rajeev
On Apr 12, 2011, at 2:11 PM, Pavan Balaji wrote:
http://www.mcs.anl.gov/research/projects/mpich2/nightly/old
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
That was ancient perl code. However, the perl code is directing all of the output, including any error messages, into the output file that passed to checkbuilds from testbuild. It must not got to STDOUT. Due to bugs in at least perl 4, it also needs to close and reopen that file (really - it took a while to find that problem). Bill On Apr 12, 2011, at 6:47 PM, Rajeev Thakur wrote:
I could not find the problem, so I have made the change you suggested in maint/checkbuilds.in. Let's see if we get more information out of the tests tonight.
Rajeev
On Apr 12, 2011, at 5:04 PM, Dave Goodell wrote:
There should be helpful log messages in the output from "maint/ checkbuilds", but I don't see anything that matches what I expect to see. It looks like there's some sketchy perl code that doesn't make a lot of sense:
----8<---- 93 $OUTFD = OUTFD; ... 980 sub RunMPD { 981 # We need to background the mpd. 982 # If there is already an mpd running (for us), we want to stop it 983 my $mpdlocalsock = "/tmp/mpd2.console_". $ENV{"LOGNAME"}; 984 my $rc = 0; 985 if (-S $mpdlocalsock) { 986 system "$instdir/bin/mpdallexit"; 987 $rc = $? >> 8; 988 if ($rc != 0) { 989 print $OUTFD "Error running $instdir/bin/mpdallexit before starting MPD ($mpdlocalsock exists)\n"; 990 } 991 } 992 #system "$instdir/bin/mpd -d"; 993 # Try to background mpd 994 system "$instdir/bin/mpd &"; 995 $rc = $? >> 8; 996 if ($rc != 0) { 997 print $OUTFD "Error running $instdir/bin/mpd -d\n"; 998 } 999 else { 1000 print $OUTFD "Started mpd in demon mode\n"; 1001 } 1002 # Wait for the mpd to start 1003 sleep 1; 1004 return; 1005 } ----8<----
Is that bareword initialization on line 93 some sort of long-broken Perl4-ism? Or was it left that way so that "debug logging" could be enabled by a carefully placed "open(OUTFD,...)" call? I suppose a cheesier version is to change the OUTFD bareword to an STDOUT bareword, which will have the intended effect when not under "use strict", although in a totally gross way.
Bill, because it's an older perl script, I'm guessing you wrote this. Is the output intentionally disabled because it causes a problem in normal use? Is there any reason that I shouldn't just replace line 93 with this?
----8<---- my $OUTFD = *STDOUT{IO}; ----8<----
-Dave
On Apr 12, 2011, at 3:55 PM CDT, Dave Goodell wrote:
I dislike working on MPD as much as the next guy, but is MPD actually broken? Or is something just a bit off in the way the tests are being run or the way that our make/install steps have changed? I suspect this is a simpler issue, although I haven't done any investigation.
-Dave
On Apr 12, 2011, at 3:37 PM CDT, Pavan Balaji wrote:
Given that: (1) we no longer have any maintainers for mpd, and (2) it has been deprecated through the 1.3 series, I propose that we remove MPD in 1.4. Thoughts?
-- Pavan
On 04/12/2011 02:21 PM, Rajeev Thakur wrote:
Looks like they have enabled writes now. You are right. The MPD and SMPD tests are broken.
Rajeev
On Apr 12, 2011, at 2:19 PM, Rajeev Thakur wrote:
They have made the web site read only while they are investigating a security incident. Until they enable writes, we won't know whether the tests are broken or it is the web site problem.
Rajeev
On Apr 12, 2011, at 2:11 PM, Pavan Balaji wrote:
http://www.mcs.anl.gov/research/projects/mpich2/nightly/old
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
William Gropp Director, Parallel Computing Institute Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign
Good to know. I somehow missed the "open" call that does that redirect on line 1101, so my suggestion would have no real effect anyway. I'll take a closer look soon to see why we aren't getting any output from the RunMPD subroutine. -Dave On Apr 13, 2011, at 10:45 AM CDT, William Gropp wrote:
That was ancient perl code. However, the perl code is directing all of the output, including any error messages, into the output file that passed to checkbuilds from testbuild. It must not got to STDOUT. Due to bugs in at least perl 4, it also needs to close and reopen that file (really - it took a while to find that problem).
Bill
On Apr 12, 2011, at 6:47 PM, Rajeev Thakur wrote:
I could not find the problem, so I have made the change you suggested in maint/checkbuilds.in. Let's see if we get more information out of the tests tonight.
Rajeev
On Apr 12, 2011, at 5:04 PM, Dave Goodell wrote:
There should be helpful log messages in the output from "maint/checkbuilds", but I don't see anything that matches what I expect to see. It looks like there's some sketchy perl code that doesn't make a lot of sense:
----8<---- 93 $OUTFD = OUTFD; ... 980 sub RunMPD { 981 # We need to background the mpd. 982 # If there is already an mpd running (for us), we want to stop it 983 my $mpdlocalsock = "/tmp/mpd2.console_". $ENV{"LOGNAME"}; 984 my $rc = 0; 985 if (-S $mpdlocalsock) { 986 system "$instdir/bin/mpdallexit"; 987 $rc = $? >> 8; 988 if ($rc != 0) { 989 print $OUTFD "Error running $instdir/bin/mpdallexit before starting MPD ($mpdlocalsock exists)\n"; 990 } 991 } 992 #system "$instdir/bin/mpd -d"; 993 # Try to background mpd 994 system "$instdir/bin/mpd &"; 995 $rc = $? >> 8; 996 if ($rc != 0) { 997 print $OUTFD "Error running $instdir/bin/mpd -d\n"; 998 } 999 else { 1000 print $OUTFD "Started mpd in demon mode\n"; 1001 } 1002 # Wait for the mpd to start 1003 sleep 1; 1004 return; 1005 } ----8<----
Is that bareword initialization on line 93 some sort of long-broken Perl4-ism? Or was it left that way so that "debug logging" could be enabled by a carefully placed "open(OUTFD,...)" call? I suppose a cheesier version is to change the OUTFD bareword to an STDOUT bareword, which will have the intended effect when not under "use strict", although in a totally gross way.
Bill, because it's an older perl script, I'm guessing you wrote this. Is the output intentionally disabled because it causes a problem in normal use? Is there any reason that I shouldn't just replace line 93 with this?
----8<---- my $OUTFD = *STDOUT{IO}; ----8<----
-Dave
On Apr 12, 2011, at 3:55 PM CDT, Dave Goodell wrote:
I dislike working on MPD as much as the next guy, but is MPD actually broken? Or is something just a bit off in the way the tests are being run or the way that our make/install steps have changed? I suspect this is a simpler issue, although I haven't done any investigation.
-Dave
On Apr 12, 2011, at 3:37 PM CDT, Pavan Balaji wrote:
Given that: (1) we no longer have any maintainers for mpd, and (2) it has been deprecated through the 1.3 series, I propose that we remove MPD in 1.4. Thoughts?
-- Pavan
On 04/12/2011 02:21 PM, Rajeev Thakur wrote:
Looks like they have enabled writes now. You are right. The MPD and SMPD tests are broken.
Rajeev
On Apr 12, 2011, at 2:19 PM, Rajeev Thakur wrote:
They have made the web site read only while they are investigating a security incident. Until they enable writes, we won't know whether the tests are broken or it is the web site problem.
Rajeev
On Apr 12, 2011, at 2:11 PM, Pavan Balaji wrote:
http://www.mcs.anl.gov/research/projects/mpich2/nightly/old
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
William Gropp Director, Parallel Computing Institute Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign
participants (4)
-
Dave Goodell -
Pavan Balaji -
Rajeev Thakur -
William Gropp