You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@qpid.apache.org by "Jan Sarenik (JIRA)" <qp...@incubator.apache.org> on 2010/04/27 11:43:32 UTC
[jira] Created: (QPID-2552) broker memory leaks
broker memory leaks
-------------------
Key: QPID-2552
URL: https://issues.apache.org/jira/browse/QPID-2552
Project: Qpid
Issue Type: Bug
Components: C++ Broker
Affects Versions: 0.7
Environment: Fedora Rawhide x86_64
glibc-2.11.90-20.x86_64
valgrind-3.5.0-16.fc14.x86_64
gcc-c++-4.4.3-18.fc14.x86_64
boost-devel-1.41.0-8.fc14.x86_64
Reporter: Jan Sarenik
Priority: Critical
Fix For: 0.7
There was a recent patch to Glibc which added another sanity check for
fastbins (I personally do not understand the whole fastbin concept yet and
when the error first occured, I was mislead by myself and thought it has
something to do with prelink, but now I have clean system without prelink
with all the binaries exactly the same as they come in packages). See
http://repo.or.cz/w/glibc.git/commit/90a3055e8bdd9308eceeadc0b37278f324ec6b5d
The problem here is, that qpidd is very easily (though randomly) caught
by this check. This often happens during 'make check' and it successfully
renders any of the tests as false-positive.
I was discussing this issue with Jakub Jelinek (Glibc hacker) as I was
not sure if the problem resides on our side or not and he suggested to
use Valgrind or ElectricFence. So I installed Valgrind and I am kindly
asking you, the developers to fix the memory leaks in Qpid.
Have a look at the attached log of 'make check' on r938298 of qpid trunk
(with little clean-up patches also included in attachment).
The problem should arise in any of the modern GNU/Linux
distributions. Even the beta RHEL6 contains new Glibc with mentioned
patch so I am setting critical priority for this.
I did not try to compile the whole thing against older (stable?) boost
version, but can do that easily if it helps in extracting the root cause
of the problem. No idea how suspicious boost can be in this case.
Get back to me if there is anything else I can help with or more
info to add.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
[jira] Updated: (QPID-2552) broker memory leaks
Posted by "Jan Sarenik (JIRA)" <qp...@incubator.apache.org>.
[ https://issues.apache.org/jira/browse/QPID-2552?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Jan Sarenik updated QPID-2552:
------------------------------
Attachment: qpid-r938298.diff
Mentioned clean-up patches for some tests.
Plus it contains patch that is vital to compile qpidd on current Fedora Rawhide (see the -lpthread part and -lboost_* part)
> broker memory leaks
> -------------------
>
> Key: QPID-2552
> URL: https://issues.apache.org/jira/browse/QPID-2552
> Project: Qpid
> Issue Type: Bug
> Components: C++ Broker
> Affects Versions: 0.7
> Environment: Fedora Rawhide x86_64
> glibc-2.11.90-20.x86_64
> valgrind-3.5.0-16.fc14.x86_64
> gcc-c++-4.4.3-18.fc14.x86_64
> boost-devel-1.41.0-8.fc14.x86_64
> Reporter: Jan Sarenik
> Priority: Critical
> Fix For: 0.7
>
> Attachments: qpid-cpp-r938298-makecheck.log, qpid-r938298.diff
>
>
> There was a recent patch to Glibc which added another sanity check for
> fastbins (I personally do not understand the whole fastbin concept yet and
> when the error first occured, I was mislead by myself and thought it has
> something to do with prelink, but now I have clean system without prelink
> with all the binaries exactly the same as they come in packages). See
> http://repo.or.cz/w/glibc.git/commit/90a3055e8bdd9308eceeadc0b37278f324ec6b5d
> The problem here is, that qpidd is very easily (though randomly) caught
> by this check. This often happens during 'make check' and it successfully
> renders any of the tests as false-positive.
> I was discussing this issue with Jakub Jelinek (Glibc hacker) as I was
> not sure if the problem resides on our side or not and he suggested to
> use Valgrind or ElectricFence. So I installed Valgrind and I am kindly
> asking you, the developers to fix the memory leaks in Qpid.
> Have a look at the attached log of 'make check' on r938298 of qpid trunk
> (with little clean-up patches also included in attachment).
> The problem should arise in any of the modern GNU/Linux
> distributions. Even the beta RHEL6 contains new Glibc with mentioned
> patch so I am setting critical priority for this.
> I did not try to compile the whole thing against older (stable?) boost
> version, but can do that easily if it helps in extracting the root cause
> of the problem. No idea how suspicious boost can be in this case.
> Get back to me if there is anything else I can help with or more
> info to add.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
[jira] Commented: (QPID-2552) broker memory leaks
Posted by "Jan Sarenik (JIRA)" <qp...@incubator.apache.org>.
[ https://issues.apache.org/jira/browse/QPID-2552?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12862337#action_12862337 ]
Jan Sarenik commented on QPID-2552:
-----------------------------------
Still valid on qpid-0.7-r939184, still random
> broker memory leaks
> -------------------
>
> Key: QPID-2552
> URL: https://issues.apache.org/jira/browse/QPID-2552
> Project: Qpid
> Issue Type: Bug
> Components: C++ Broker
> Affects Versions: 0.7
> Environment: Fedora Rawhide x86_64
> glibc-2.11.90-20.x86_64
> valgrind-3.5.0-16.fc14.x86_64
> gcc-c++-4.4.3-18.fc14.x86_64
> boost-devel-1.41.0-8.fc14.x86_64
> Reporter: Jan Sarenik
> Assignee: Gordon Sim
> Priority: Critical
> Fix For: 0.7
>
> Attachments: qpid-cpp-r938298-makecheck.log, qpid-r938298.diff
>
>
> There was a recent patch to Glibc which added another sanity check for
> fastbins (I personally do not understand the whole fastbin concept yet and
> when the error first occured, I was mislead by myself and thought it has
> something to do with prelink, but now I have clean system without prelink
> with all the binaries exactly the same as they come in packages). See
> http://repo.or.cz/w/glibc.git/commit/90a3055e8bdd9308eceeadc0b37278f324ec6b5d
> The problem here is, that qpidd is very easily (though randomly) caught
> by this check. This often happens during 'make check' and it successfully
> renders any of the tests as false-positive.
> I was discussing this issue with Jakub Jelinek (Glibc hacker) as I was
> not sure if the problem resides on our side or not and he suggested to
> use Valgrind or ElectricFence. So I installed Valgrind and I am kindly
> asking you, the developers to fix the memory leaks in Qpid.
> Have a look at the attached log of 'make check' on r938298 of qpid trunk
> (with little clean-up patches also included in attachment).
> The problem should arise in any of the modern GNU/Linux
> distributions. Even the beta RHEL6 contains new Glibc with mentioned
> patch so I am setting critical priority for this.
> I did not try to compile the whole thing against older (stable?) boost
> version, but can do that easily if it helps in extracting the root cause
> of the problem. No idea how suspicious boost can be in this case.
> Get back to me if there is anything else I can help with or more
> info to add.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
[jira] Assigned: (QPID-2552) broker memory leaks
Posted by "Gordon Sim (JIRA)" <qp...@incubator.apache.org>.
[ https://issues.apache.org/jira/browse/QPID-2552?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Gordon Sim reassigned QPID-2552:
--------------------------------
Assignee: Gordon Sim
> broker memory leaks
> -------------------
>
> Key: QPID-2552
> URL: https://issues.apache.org/jira/browse/QPID-2552
> Project: Qpid
> Issue Type: Bug
> Components: C++ Broker
> Affects Versions: 0.7
> Environment: Fedora Rawhide x86_64
> glibc-2.11.90-20.x86_64
> valgrind-3.5.0-16.fc14.x86_64
> gcc-c++-4.4.3-18.fc14.x86_64
> boost-devel-1.41.0-8.fc14.x86_64
> Reporter: Jan Sarenik
> Assignee: Gordon Sim
> Priority: Critical
> Fix For: 0.7
>
> Attachments: qpid-cpp-r938298-makecheck.log, qpid-r938298.diff
>
>
> There was a recent patch to Glibc which added another sanity check for
> fastbins (I personally do not understand the whole fastbin concept yet and
> when the error first occured, I was mislead by myself and thought it has
> something to do with prelink, but now I have clean system without prelink
> with all the binaries exactly the same as they come in packages). See
> http://repo.or.cz/w/glibc.git/commit/90a3055e8bdd9308eceeadc0b37278f324ec6b5d
> The problem here is, that qpidd is very easily (though randomly) caught
> by this check. This often happens during 'make check' and it successfully
> renders any of the tests as false-positive.
> I was discussing this issue with Jakub Jelinek (Glibc hacker) as I was
> not sure if the problem resides on our side or not and he suggested to
> use Valgrind or ElectricFence. So I installed Valgrind and I am kindly
> asking you, the developers to fix the memory leaks in Qpid.
> Have a look at the attached log of 'make check' on r938298 of qpid trunk
> (with little clean-up patches also included in attachment).
> The problem should arise in any of the modern GNU/Linux
> distributions. Even the beta RHEL6 contains new Glibc with mentioned
> patch so I am setting critical priority for this.
> I did not try to compile the whole thing against older (stable?) boost
> version, but can do that easily if it helps in extracting the root cause
> of the problem. No idea how suspicious boost can be in this case.
> Get back to me if there is anything else I can help with or more
> info to add.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
[jira] Updated: (QPID-2552) broker memory leaks
Posted by "Jan Sarenik (JIRA)" <qp...@incubator.apache.org>.
[ https://issues.apache.org/jira/browse/QPID-2552?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Jan Sarenik updated QPID-2552:
------------------------------
Attachment: qpid-cpp-r938298-makecheck.log
The log of 'make check' run on r938298 svn trunk (only the little patch applied, see other attachment)
> broker memory leaks
> -------------------
>
> Key: QPID-2552
> URL: https://issues.apache.org/jira/browse/QPID-2552
> Project: Qpid
> Issue Type: Bug
> Components: C++ Broker
> Affects Versions: 0.7
> Environment: Fedora Rawhide x86_64
> glibc-2.11.90-20.x86_64
> valgrind-3.5.0-16.fc14.x86_64
> gcc-c++-4.4.3-18.fc14.x86_64
> boost-devel-1.41.0-8.fc14.x86_64
> Reporter: Jan Sarenik
> Priority: Critical
> Fix For: 0.7
>
> Attachments: qpid-cpp-r938298-makecheck.log
>
>
> There was a recent patch to Glibc which added another sanity check for
> fastbins (I personally do not understand the whole fastbin concept yet and
> when the error first occured, I was mislead by myself and thought it has
> something to do with prelink, but now I have clean system without prelink
> with all the binaries exactly the same as they come in packages). See
> http://repo.or.cz/w/glibc.git/commit/90a3055e8bdd9308eceeadc0b37278f324ec6b5d
> The problem here is, that qpidd is very easily (though randomly) caught
> by this check. This often happens during 'make check' and it successfully
> renders any of the tests as false-positive.
> I was discussing this issue with Jakub Jelinek (Glibc hacker) as I was
> not sure if the problem resides on our side or not and he suggested to
> use Valgrind or ElectricFence. So I installed Valgrind and I am kindly
> asking you, the developers to fix the memory leaks in Qpid.
> Have a look at the attached log of 'make check' on r938298 of qpid trunk
> (with little clean-up patches also included in attachment).
> The problem should arise in any of the modern GNU/Linux
> distributions. Even the beta RHEL6 contains new Glibc with mentioned
> patch so I am setting critical priority for this.
> I did not try to compile the whole thing against older (stable?) boost
> version, but can do that easily if it helps in extracting the root cause
> of the problem. No idea how suspicious boost can be in this case.
> Get back to me if there is anything else I can help with or more
> info to add.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
[jira] Commented: (QPID-2552) broker memory leaks
Posted by "Jan Sarenik (JIRA)" <qp...@incubator.apache.org>.
[ https://issues.apache.org/jira/browse/QPID-2552?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12861334#action_12861334 ]
Jan Sarenik commented on QPID-2552:
-----------------------------------
In the log file look for lines simillar to following one:
*** glibc detected *** /home/jsarenik/pub/git/qpid/cpp/src/.libs/lt-qpidd: invalid fastbin entry (free): 0x0000000001bdde80 ***
> broker memory leaks
> -------------------
>
> Key: QPID-2552
> URL: https://issues.apache.org/jira/browse/QPID-2552
> Project: Qpid
> Issue Type: Bug
> Components: C++ Broker
> Affects Versions: 0.7
> Environment: Fedora Rawhide x86_64
> glibc-2.11.90-20.x86_64
> valgrind-3.5.0-16.fc14.x86_64
> gcc-c++-4.4.3-18.fc14.x86_64
> boost-devel-1.41.0-8.fc14.x86_64
> Reporter: Jan Sarenik
> Priority: Critical
> Fix For: 0.7
>
> Attachments: qpid-cpp-r938298-makecheck.log, qpid-r938298.diff
>
>
> There was a recent patch to Glibc which added another sanity check for
> fastbins (I personally do not understand the whole fastbin concept yet and
> when the error first occured, I was mislead by myself and thought it has
> something to do with prelink, but now I have clean system without prelink
> with all the binaries exactly the same as they come in packages). See
> http://repo.or.cz/w/glibc.git/commit/90a3055e8bdd9308eceeadc0b37278f324ec6b5d
> The problem here is, that qpidd is very easily (though randomly) caught
> by this check. This often happens during 'make check' and it successfully
> renders any of the tests as false-positive.
> I was discussing this issue with Jakub Jelinek (Glibc hacker) as I was
> not sure if the problem resides on our side or not and he suggested to
> use Valgrind or ElectricFence. So I installed Valgrind and I am kindly
> asking you, the developers to fix the memory leaks in Qpid.
> Have a look at the attached log of 'make check' on r938298 of qpid trunk
> (with little clean-up patches also included in attachment).
> The problem should arise in any of the modern GNU/Linux
> distributions. Even the beta RHEL6 contains new Glibc with mentioned
> patch so I am setting critical priority for this.
> I did not try to compile the whole thing against older (stable?) boost
> version, but can do that easily if it helps in extracting the root cause
> of the problem. No idea how suspicious boost can be in this case.
> Get back to me if there is anything else I can help with or more
> info to add.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org
[jira] Commented: (QPID-2552) broker memory leaks
Posted by "Jan Sarenik (JIRA)" <qp...@incubator.apache.org>.
[ https://issues.apache.org/jira/browse/QPID-2552?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12865110#action_12865110 ]
Jan Sarenik commented on QPID-2552:
-----------------------------------
Exactly the same fastbin catch happens with
glibc-2.12-1.x86_64
gcc-4.4.4-2.fc14.x86_64
boost-1.41.0-9.fc14.x86_64
qpid r941968
> broker memory leaks
> -------------------
>
> Key: QPID-2552
> URL: https://issues.apache.org/jira/browse/QPID-2552
> Project: Qpid
> Issue Type: Bug
> Components: C++ Broker
> Affects Versions: 0.7
> Environment: Fedora Rawhide x86_64
> glibc-2.11.90-20.x86_64
> valgrind-3.5.0-16.fc14.x86_64
> gcc-c++-4.4.3-18.fc14.x86_64
> boost-devel-1.41.0-8.fc14.x86_64
> Reporter: Jan Sarenik
> Assignee: Gordon Sim
> Priority: Critical
> Fix For: 0.7
>
> Attachments: qpid-cpp-r938298-makecheck.log, qpid-r938298.diff
>
>
> There was a recent patch to Glibc which added another sanity check for
> fastbins (I personally do not understand the whole fastbin concept yet and
> when the error first occured, I was mislead by myself and thought it has
> something to do with prelink, but now I have clean system without prelink
> with all the binaries exactly the same as they come in packages). See
> http://repo.or.cz/w/glibc.git/commit/90a3055e8bdd9308eceeadc0b37278f324ec6b5d
> The problem here is, that qpidd is very easily (though randomly) caught
> by this check. This often happens during 'make check' and it successfully
> renders any of the tests as false-positive.
> I was discussing this issue with Jakub Jelinek (Glibc hacker) as I was
> not sure if the problem resides on our side or not and he suggested to
> use Valgrind or ElectricFence. So I installed Valgrind and I am kindly
> asking you, the developers to fix the memory leaks in Qpid.
> Have a look at the attached log of 'make check' on r938298 of qpid trunk
> (with little clean-up patches also included in attachment).
> The problem should arise in any of the modern GNU/Linux
> distributions. Even the beta RHEL6 contains new Glibc with mentioned
> patch so I am setting critical priority for this.
> I did not try to compile the whole thing against older (stable?) boost
> version, but can do that easily if it helps in extracting the root cause
> of the problem. No idea how suspicious boost can be in this case.
> Get back to me if there is anything else I can help with or more
> info to add.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org