You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by Shai Erera <se...@gmail.com> on 2015/02/04 11:02:20 UTC
FST.pack() J9 bug
Hi
I was asked by someone in IBM about this bug which is listed under our
JavaBugs page:
*FST.pack() produces corrupt index (Lucene 4) because a loop is miscompiled
*
The J9 team would like to investigate that and they I've asked me for some
more details. Since we don't have an issue reported, I searched and came up
with several links:
http://mail-archives.apache.org/mod_mbox/lucene-dev/201404.mbox/%3Calpine.DEB.2.02.1404031005110.14297@frisbee%3E
http://marc.info/?l=solr-dev&m=134454323703970
Since the Wiki page describes a work around to disable JIT optimization for
FST.pack(), I wanted to ask:
1. Are our Jenkins builds with J9 disable this optimization currently?
2. Do we know if this bug still happens?
3. Is there a test which can reproduce the bug (even if it doesn't
always reproduce it) so the J9 folks can debug?
I am not sure if the tests in the above link are still relevant or were
good at reproducing the bug.
If it's not easily reproducible, is there additional information besides
what's written on the Wiki page that you think can help them investigate
this?
Shai
RE: FST.pack() J9 bug
Posted by Uwe Schindler <uw...@thetaphi.de>.
Hi Shai,
just install J9 and the run *all* core tests in a loop, ideally with “-Dtests.jvms=1”. It is important to run *all* tests, because it does not happen on a single test case alone.
In fact, it could be that some recent changes in FSTs may have worked around that issue. I testes the last time approx. half a year ago when I installed the last J9 update. Maybe search for occurrences in the mailing list archive. I actually ran Jenkins a few days without the disabled optimization and failures happened quite often.
In addition to the TAR.GZ issue: What the status about the full encryption support? Why does Oracle offers downloads with full encryption support (we need that for Lucene), but J9 has to be patched first? With unpatched policy files you cannot even run the build at all, because it already fails while trying to connect to issues.apache.org with SSL/TLS or some of the Maven repositories with https:// URLs. It looks like J9 only supports 1024 bits certificates, which are no longer used and got replaced by most providers recently (so affecting Maven and issues.apache.org).
Unfortunately I have no time to test J9 locally at the moment, sorry.
Uwe
-----
Uwe Schindler
H.-H.-Meier-Allee 63, D-28213 Bremen
<http://www.thetaphi.de/> http://www.thetaphi.de
eMail: uwe@thetaphi.de
From: Shai Erera [mailto:serera@gmail.com]
Sent: Wednesday, February 04, 2015 2:16 PM
To: dev@lucene.apache.org
Subject: Re: FST.pack() J9 bug
Thanks guys, I understand and I recently raised this issue again w/ the J9 team. I am told that the TAR.GZ is probably not an option, for legal reasons which I do not understand. I am taking it one step at a time, and currently they would like to investigate this FST.pack() issue since it can lead to index corruption, and this is not acceptable.
I also asked them to figure out a process which would allow us (the community) to file bug reports and have them work with us on these issues. I hope they will be able to figure out such a process. I will update the list when such process will be in place.
Back to the FST.pack(), from what Uwe says it seems it reproduces very easily (he tries new J9 versions, they fail and he disables the optimization). If so, does someone have an example such failure which I can ask them to reproduce?
Shai
On Wed, Feb 4, 2015 at 12:20 PM, Dawid Weiss <da...@cs.put.poznan.pl> wrote:
I agree with Uwe -- even trying to get J9 (on Windows, for example)
used to be a headache. Not to mention there is no clear channel to
report bugs to (and receive feedback from)... it makes testing on J9 a
burden. Some kind of (official or semi-official) help from IBM to get
this rolling more smoothly would be great.
Dawid
On Wed, Feb 4, 2015 at 11:12 AM, Uwe Schindler <uw...@thetaphi.de> wrote:
> Hi Shai,
>
>
>
> the current Jenkins disables the optimization, with the listed flag.
> Whenever I update the version, I check if the bug still happens. And it
> still happens at least with the version currently installed.
>
>
>
> serv1:/var/lib/jenkins/tools/java/64bit/ibm-j9-jdk7# bin/java -version
>
> java version "1.7.0"
>
> Java(TM) SE Runtime Environment (build pxa6470_27sr1-20140411_01(SR1))
>
> IBM J9 VM (build 2.7, JRE 1.7.0 Linux amd64-64 Compressed References
> 20140410_195893 (JIT enabled, AOT enabled)
>
> J9VM - R27_Java727_SR1_20140410_1931_B195893
>
> JIT - tr.r13.java_20140410_61421
>
> GC - R27_Java727_SR1_20140410_1931_B195893_CMPRSS
>
> J9CL - 20140410_195893)
>
> JCL - 20140409_01 based on Oracle 7u55-b13
>
>
>
> I have not yet updated to last J9 version, but the reason for this is the
> brokenness of the installation (InstallShield in a console, haha). If IBM
> would provide a simple TAR.GZ file and also a version with the “extended
> encryption java policy files”, I would do this more often. But currently
> this costs me half an hour per platform to install this and it is all
> voluntary!
>
>
>
> In addition there is no IBM J9 for Java 8, so we don’t run TRUNK tests
> anymore with J9 (requires Java 8).
>
>
>
> See my other message about this a while back on this mailing list.
>
> Uwe
>
>
>
> -----
>
> Uwe Schindler
>
> H.-H.-Meier-Allee 63, D-28213 Bremen
>
> http://www.thetaphi.de
>
> eMail: uwe@thetaphi.de
>
>
>
> From: Shai Erera [mailto:serera@gmail.com]
> Sent: Wednesday, February 04, 2015 11:02 AM
> To: dev@lucene.apache.org
> Subject: FST.pack() J9 bug
>
>
>
> Hi
>
> I was asked by someone in IBM about this bug which is listed under our
> JavaBugs page:
>
> FST.pack() produces corrupt index (Lucene 4) because a loop is miscompiled
>
> The J9 team would like to investigate that and they I've asked me for some
> more details. Since we don't have an issue reported, I searched and came up
> with several links:
>
> http://mail-archives.apache.org/mod_mbox/lucene-dev/201404.mbox/%3Calpine.DEB.2.02.1404031005110.14297@frisbee%3E
> http://marc.info/?l=solr-dev <http://marc.info/?l=solr-dev&m=134454323703970> &m=134454323703970
>
> Since the Wiki page describes a work around to disable JIT optimization for
> FST.pack(), I wanted to ask:
>
> Are our Jenkins builds with J9 disable this optimization currently?
> Do we know if this bug still happens?
> Is there a test which can reproduce the bug (even if it doesn't always
> reproduce it) so the J9 folks can debug?
> I am not sure if the tests in the above link are still relevant or were good
> at reproducing the bug.
>
> If it's not easily reproducible, is there additional information besides
> what's written on the Wiki page that you think can help them investigate
> this?
>
> Shai
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org
Re: FST.pack() J9 bug
Posted by Shai Erera <se...@gmail.com>.
Thanks guys, I understand and I recently raised this issue again w/ the J9
team. I am told that the TAR.GZ is probably not an option, for legal
reasons which I do not understand. I am taking it one step at a time, and
currently they would like to investigate this FST.pack() issue since it can
lead to index corruption, and this is not acceptable.
I also asked them to figure out a process which would allow us (the
community) to file bug reports and have them work with us on these issues.
I hope they will be able to figure out such a process. I will update the
list when such process will be in place.
Back to the FST.pack(), from what Uwe says it seems it reproduces very
easily (he tries new J9 versions, they fail and he disables the
optimization). If so, does someone have an example such failure which I can
ask them to reproduce?
Shai
On Wed, Feb 4, 2015 at 12:20 PM, Dawid Weiss <da...@cs.put.poznan.pl>
wrote:
> I agree with Uwe -- even trying to get J9 (on Windows, for example)
> used to be a headache. Not to mention there is no clear channel to
> report bugs to (and receive feedback from)... it makes testing on J9 a
> burden. Some kind of (official or semi-official) help from IBM to get
> this rolling more smoothly would be great.
>
> Dawid
>
> On Wed, Feb 4, 2015 at 11:12 AM, Uwe Schindler <uw...@thetaphi.de> wrote:
> > Hi Shai,
> >
> >
> >
> > the current Jenkins disables the optimization, with the listed flag.
> > Whenever I update the version, I check if the bug still happens. And it
> > still happens at least with the version currently installed.
> >
> >
> >
> > serv1:/var/lib/jenkins/tools/java/64bit/ibm-j9-jdk7# bin/java -version
> >
> > java version "1.7.0"
> >
> > Java(TM) SE Runtime Environment (build pxa6470_27sr1-20140411_01(SR1))
> >
> > IBM J9 VM (build 2.7, JRE 1.7.0 Linux amd64-64 Compressed References
> > 20140410_195893 (JIT enabled, AOT enabled)
> >
> > J9VM - R27_Java727_SR1_20140410_1931_B195893
> >
> > JIT - tr.r13.java_20140410_61421
> >
> > GC - R27_Java727_SR1_20140410_1931_B195893_CMPRSS
> >
> > J9CL - 20140410_195893)
> >
> > JCL - 20140409_01 based on Oracle 7u55-b13
> >
> >
> >
> > I have not yet updated to last J9 version, but the reason for this is the
> > brokenness of the installation (InstallShield in a console, haha). If IBM
> > would provide a simple TAR.GZ file and also a version with the “extended
> > encryption java policy files”, I would do this more often. But currently
> > this costs me half an hour per platform to install this and it is all
> > voluntary!
> >
> >
> >
> > In addition there is no IBM J9 for Java 8, so we don’t run TRUNK tests
> > anymore with J9 (requires Java 8).
> >
> >
> >
> > See my other message about this a while back on this mailing list.
> >
> > Uwe
> >
> >
> >
> > -----
> >
> > Uwe Schindler
> >
> > H.-H.-Meier-Allee 63, D-28213 Bremen
> >
> > http://www.thetaphi.de
> >
> > eMail: uwe@thetaphi.de
> >
> >
> >
> > From: Shai Erera [mailto:serera@gmail.com]
> > Sent: Wednesday, February 04, 2015 11:02 AM
> > To: dev@lucene.apache.org
> > Subject: FST.pack() J9 bug
> >
> >
> >
> > Hi
> >
> > I was asked by someone in IBM about this bug which is listed under our
> > JavaBugs page:
> >
> > FST.pack() produces corrupt index (Lucene 4) because a loop is
> miscompiled
> >
> > The J9 team would like to investigate that and they I've asked me for
> some
> > more details. Since we don't have an issue reported, I searched and came
> up
> > with several links:
> >
> >
> http://mail-archives.apache.org/mod_mbox/lucene-dev/201404.mbox/%3Calpine.DEB.2.02.1404031005110.14297@frisbee%3E
> > http://marc.info/?l=solr-dev&m=134454323703970
> >
> > Since the Wiki page describes a work around to disable JIT optimization
> for
> > FST.pack(), I wanted to ask:
> >
> > Are our Jenkins builds with J9 disable this optimization currently?
> > Do we know if this bug still happens?
> > Is there a test which can reproduce the bug (even if it doesn't always
> > reproduce it) so the J9 folks can debug?
> > I am not sure if the tests in the above link are still relevant or were
> good
> > at reproducing the bug.
> >
> > If it's not easily reproducible, is there additional information besides
> > what's written on the Wiki page that you think can help them investigate
> > this?
> >
> > Shai
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
> For additional commands, e-mail: dev-help@lucene.apache.org
>
>
Re: FST.pack() J9 bug
Posted by Dawid Weiss <da...@cs.put.poznan.pl>.
I agree with Uwe -- even trying to get J9 (on Windows, for example)
used to be a headache. Not to mention there is no clear channel to
report bugs to (and receive feedback from)... it makes testing on J9 a
burden. Some kind of (official or semi-official) help from IBM to get
this rolling more smoothly would be great.
Dawid
On Wed, Feb 4, 2015 at 11:12 AM, Uwe Schindler <uw...@thetaphi.de> wrote:
> Hi Shai,
>
>
>
> the current Jenkins disables the optimization, with the listed flag.
> Whenever I update the version, I check if the bug still happens. And it
> still happens at least with the version currently installed.
>
>
>
> serv1:/var/lib/jenkins/tools/java/64bit/ibm-j9-jdk7# bin/java -version
>
> java version "1.7.0"
>
> Java(TM) SE Runtime Environment (build pxa6470_27sr1-20140411_01(SR1))
>
> IBM J9 VM (build 2.7, JRE 1.7.0 Linux amd64-64 Compressed References
> 20140410_195893 (JIT enabled, AOT enabled)
>
> J9VM - R27_Java727_SR1_20140410_1931_B195893
>
> JIT - tr.r13.java_20140410_61421
>
> GC - R27_Java727_SR1_20140410_1931_B195893_CMPRSS
>
> J9CL - 20140410_195893)
>
> JCL - 20140409_01 based on Oracle 7u55-b13
>
>
>
> I have not yet updated to last J9 version, but the reason for this is the
> brokenness of the installation (InstallShield in a console, haha). If IBM
> would provide a simple TAR.GZ file and also a version with the “extended
> encryption java policy files”, I would do this more often. But currently
> this costs me half an hour per platform to install this and it is all
> voluntary!
>
>
>
> In addition there is no IBM J9 for Java 8, so we don’t run TRUNK tests
> anymore with J9 (requires Java 8).
>
>
>
> See my other message about this a while back on this mailing list.
>
> Uwe
>
>
>
> -----
>
> Uwe Schindler
>
> H.-H.-Meier-Allee 63, D-28213 Bremen
>
> http://www.thetaphi.de
>
> eMail: uwe@thetaphi.de
>
>
>
> From: Shai Erera [mailto:serera@gmail.com]
> Sent: Wednesday, February 04, 2015 11:02 AM
> To: dev@lucene.apache.org
> Subject: FST.pack() J9 bug
>
>
>
> Hi
>
> I was asked by someone in IBM about this bug which is listed under our
> JavaBugs page:
>
> FST.pack() produces corrupt index (Lucene 4) because a loop is miscompiled
>
> The J9 team would like to investigate that and they I've asked me for some
> more details. Since we don't have an issue reported, I searched and came up
> with several links:
>
> http://mail-archives.apache.org/mod_mbox/lucene-dev/201404.mbox/%3Calpine.DEB.2.02.1404031005110.14297@frisbee%3E
> http://marc.info/?l=solr-dev&m=134454323703970
>
> Since the Wiki page describes a work around to disable JIT optimization for
> FST.pack(), I wanted to ask:
>
> Are our Jenkins builds with J9 disable this optimization currently?
> Do we know if this bug still happens?
> Is there a test which can reproduce the bug (even if it doesn't always
> reproduce it) so the J9 folks can debug?
> I am not sure if the tests in the above link are still relevant or were good
> at reproducing the bug.
>
> If it's not easily reproducible, is there additional information besides
> what's written on the Wiki page that you think can help them investigate
> this?
>
> Shai
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org
RE: FST.pack() J9 bug
Posted by Uwe Schindler <uw...@thetaphi.de>.
Hi Shai,
the current Jenkins disables the optimization, with the listed flag. Whenever I update the version, I check if the bug still happens. And it still happens at least with the version currently installed.
serv1:/var/lib/jenkins/tools/java/64bit/ibm-j9-jdk7# bin/java -version
java version "1.7.0"
Java(TM) SE Runtime Environment (build pxa6470_27sr1-20140411_01(SR1))
IBM J9 VM (build 2.7, JRE 1.7.0 Linux amd64-64 Compressed References 20140410_195893 (JIT enabled, AOT enabled)
J9VM - R27_Java727_SR1_20140410_1931_B195893
JIT - tr.r13.java_20140410_61421
GC - R27_Java727_SR1_20140410_1931_B195893_CMPRSS
J9CL - 20140410_195893)
JCL - 20140409_01 based on Oracle 7u55-b13
I have not yet updated to last J9 version, but the reason for this is the brokenness of the installation (InstallShield in a console, haha). If IBM would provide a simple TAR.GZ file and also a version with the “extended encryption java policy files”, I would do this more often. But currently this costs me half an hour per platform to install this and it is all voluntary!
In addition there is no IBM J9 for Java 8, so we don’t run TRUNK tests anymore with J9 (requires Java 8).
See my other message about this a while back on this mailing list.
Uwe
-----
Uwe Schindler
H.-H.-Meier-Allee 63, D-28213 Bremen
<http://www.thetaphi.de/> http://www.thetaphi.de
eMail: uwe@thetaphi.de
From: Shai Erera [mailto:serera@gmail.com]
Sent: Wednesday, February 04, 2015 11:02 AM
To: dev@lucene.apache.org
Subject: FST.pack() J9 bug
Hi
I was asked by someone in IBM about this bug which is listed under our JavaBugs page:
FST.pack() produces corrupt index (Lucene 4) because a loop is miscompiled
The J9 team would like to investigate that and they I've asked me for some more details. Since we don't have an issue reported, I searched and came up with several links:
http://mail-archives.apache.org/mod_mbox/lucene-dev/201404.mbox/%3Calpine.DEB.2.02.1404031005110.14297@frisbee%3E
http://marc.info/?l=solr-dev <http://marc.info/?l=solr-dev&m=134454323703970> &m=134454323703970
Since the Wiki page describes a work around to disable JIT optimization for FST.pack(), I wanted to ask:
1. Are our Jenkins builds with J9 disable this optimization currently?
2. Do we know if this bug still happens?
3. Is there a test which can reproduce the bug (even if it doesn't always reproduce it) so the J9 folks can debug?
I am not sure if the tests in the above link are still relevant or were good at reproducing the bug.
If it's not easily reproducible, is there additional information besides what's written on the Wiki page that you think can help them investigate this?
Shai