sablevm-developer

[Sablevm-developer] SableVM hangs during database connection

[Roman K. <ro...@on...>]

Hi list,

I found another issue, this time related to a database connection with
MySQL. During sending an update, the hole connection seems to hang
infinitely. I waited >10minutes, which should me more than enough. As
far as I worked this out, the issue seems to be in some native code.
Here is the stacktrace from CTRL+\:

(PlainSocketImpl.java:-1)  gnu/java/net/PlainSocketImpl.read  n
(PlainSocketImpl.java:431) 
gnu/java/net/PlainSocketImpl$SocketInputStream.read 
(FilterInputStream.java:189)  java/io/FilterInputStream.read
(BufferedInputStream.java:370)  java/io/BufferedInputStream.refill
(BufferedInputStream.java:231)  java/io/BufferedInputStream.read
(MysqlIO.java:1399)  com/mysql/jdbc/MysqlIO.reuseAndReadPacket
(MysqlIO.java:1775)  com/mysql/jdbc/MysqlIO.checkErrorPacket
(MysqlIO.java:1020)  com/mysql/jdbc/MysqlIO.sendCommand
(MysqlIO.java:1109)  com/mysql/jdbc/MysqlIO.sqlQueryDirect
(Connection.java:2030)  com/mysql/jdbc/Connection.execSQL
(PreparedStatement.java:1750) 
com/mysql/jdbc/PreparedStatement.executeUpdate
(PreparedStatement.java:1596) 
com/mysql/jdbc/PreparedStatement.executeUpdate
(JDBCStorage.java:138)  com/knowledgetracking/JDBCStorage.storeXML
(NBEngine.java:78)  com/knowledgetracking/nb/NBEngine.learn
(NBEngine.java:37)  com/knowledgetracking/nb/NBEngine.learn
(learn.java:29)  com/knowledgetracking/cli/learn.invoke
(learn.java:45)  com/knowledgetracking/cli/learn.invoke
(Method.java:-1)  java/lang/reflect/Method.invokeNative  n
(Method.java:556)  java/lang/reflect/Method.invoke
((null):-1)  bsh/Reflect.invokeOnMethod
((null):-1)  bsh/Reflect.invokeStaticMethod
((null):-1)  bsh/Reflect.invokeCompiledCommand
((null):-1)  bsh/Name.invokeLocalMethod
((null):-1)  bsh/Name.invokeMethod
((null):-1)  bsh/BSHMethodInvocation.eval
((null):-1)  bsh/BSHPrimaryExpression.eval
((null):-1)  bsh/BSHPrimaryExpression.eval
((null):-1)  bsh/Interpreter.eval
((null):-1)  bsh/Interpreter.source
((null):-1)  bsh/Interpreter.main
(VirtualMachine.java:-1)  java/lang/VirtualMachine.invokeMain  n
(VirtualMachine.java:92)  java/lang/VirtualMachine.main

I'm trying to put together a simple test case to make this reproducible.
Is there any more information I can send (like a stacktrace of the
native code-> I don't know how to get this)?

Cheers, Roman

Re: [Sablevm-developer] SableVM hangs during database connection

[Grzegorz B. P. <ga...@de...>]

W liście z wto, 15-06-2004, godz. 14:18, Roman Kennke pisze: 
> Hi list,
> 
> I found another issue, this time related to a database connection with
> MySQL. During sending an update, the hole connection seems to hang
> infinitely. I waited >10minutes, which should me more than enough. As
> far as I worked this out, the issue seems to be in some native code.
> Here is the stacktrace from CTRL+\:

[...]

> I'm trying to put together a simple test case to make this reproducible.
> Is there any more information I can send (like a stacktrace of the
> native code-> I don't know how to get this)?

You need to get PID of the sablevm process, then you do:
$ gdb /path/to/bin/sablevm $PID
at gdb command prompt you simply do 'bt'

If your app is multithreaded, you might want to attach to each thread
and get backtrace it.

HTH

				Grzegorz B. Prokopski
-- 
Grzegorz B. Prokopski      <ga...@de...>
Debian GNU/Linux           http://www.debian.org
SableVM - LGPL'ed Java VM  http://www.sablevm.org
Why SableVM ?!?            http://devel.sablevm.org/wiki/Features

Re: [Sablevm-developer] SableVM hangs during database connection

[Roman K. <ro...@on...>]

Am Di, den 15.06.2004 schrieb Roman Kennke um 20:18:
> Hi list,
>=20
> I found another issue, this time related to a database connection with
> MySQL. During sending an update, the hole connection seems to hang
> infinitely.

Update: I've done a lot of tests, during which I got it running on
Linux. I used a new MySQL driver for this. Unfortunately my testprog
does still not run under FreeBSD. I will attach my testprog. The result
under FreeBSD is, that I can transmit a byte array of size <21488 bytes,
not 1 byte more (in this case the whole thing hangs). The gdb backtrace
is surprisingly short:
(gdb) bt
#0  0x2817f2af in poll () from /lib/libc.so.5
#1  0x28123a51 in _thread_kern_sched_state_unlock () from
/usr/lib/libc_r.so.5
#2  0x28123445 in _thread_kern_scheduler () from /usr/lib/libc_r.so.5
(gdb)

I have not yet solved this, and have no idea.

But I have solved another small problem, which I will write in another
email. :-)

Cheers, Roman

Re: [Sablevm-developer] SableVM hangs during database connection

[Chris P. <chr...@ma...>]

Roman Kennke wrote:
> Am Di, den 15.06.2004 schrieb Roman Kennke um 20:18:
> 
>>Hi list,
>>
>>I found another issue, this time related to a database connection with
>>MySQL. During sending an update, the hole connection seems to hang
>>infinitely.
> 
> 
> Update: I've done a lot of tests, during which I got it running on
> Linux. I used a new MySQL driver for this. Unfortunately my testprog
> does still not run under FreeBSD. I will attach my testprog. 

Hi Roman,

I think you forgot to attach the test case.

Does anyone in our group have an account on a FreeBSD machine?

Cheers,
Chris

The result
> under FreeBSD is, that I can transmit a byte array of size <21488 bytes,
> not 1 byte more (in this case the whole thing hangs). The gdb backtrace
> is surprisingly short:
> (gdb) bt
> #0  0x2817f2af in poll () from /lib/libc.so.5
> #1  0x28123a51 in _thread_kern_sched_state_unlock () from
> /usr/lib/libc_r.so.5
> #2  0x28123445 in _thread_kern_scheduler () from /usr/lib/libc_r.so.5
> (gdb)
> 
> I have not yet solved this, and have no idea.
> 
> But I have solved another small problem, which I will write in another
> email. :-)
> 
> Cheers, Roman
> 

Re: [Sablevm-developer] SableVM hangs during database connection

[Roman K. <ro...@on...>]

Hi,

> >>I found another issue, this time related to a database connection with
> >>MySQL. During sending an update, the hole connection seems to hang
> >>infinitely.
> > 
> > 
> > Update: I've done a lot of tests, during which I got it running on
> > Linux. I used a new MySQL driver for this. Unfortunately my testprog
> > does still not run under FreeBSD. I will attach my testprog. 
> 
> Hi Roman,
> 
> I think you forgot to attach the test case.

Oh, sorry. As always, I forgot the attachment ;-)

In order to run the test, you need a MySQL database set up this way:

CREATE DATABASE test;
GRANT ALL PRIVILEGES ON test.* TO test IDENTIFIED BY 'test';
GRANT ALL PRIVILEGES ON test.* TO test@localhost IDENTIFIED BY 'test';
USE test;
CREATE TABLE (
  test LONGBLOB
);

and of course you need the MySQL JDBC driver (I used version 3.0.14).

I hope, someone is willing and able to run this.

Cheers, Roman

Re: [Sablevm-developer] SableVM hangs during database connection

[Chris P. <chr...@ma...>]

Roman Kennke wrote:
> Hi,
> 
> 
>>>>I found another issue, this time related to a database connection with
>>>>MySQL. During sending an update, the hole connection seems to hang
>>>>infinitely.
>>>
>>>
>>>Update: I've done a lot of tests, during which I got it running on
>>>Linux. I used a new MySQL driver for this. Unfortunately my testprog
>>>does still not run under FreeBSD. I will attach my testprog. 
>>
>>Hi Roman,
>>
>>I think you forgot to attach the test case.
> 
> 
> Oh, sorry. As always, I forgot the attachment ;-)
> 
> In order to run the test, you need a MySQL database set up this way:
> 
> CREATE DATABASE test;
> GRANT ALL PRIVILEGES ON test.* TO test IDENTIFIED BY 'test';
> GRANT ALL PRIVILEGES ON test.* TO test@localhost IDENTIFIED BY 'test';
> USE test;
> CREATE TABLE (
>   test LONGBLOB
> );
> 
> and of course you need the MySQL JDBC driver (I used version 3.0.14).
> 
> I hope, someone is willing and able to run this.

Eventually someone will get to it.  I don't know if we have a FreeBSD 
machine.  If you were able to provide shell access to somebody it might 
be faster.  Anyway, can you try

./configure --enable-debugging-features --with-threading=switch

?

I have no idea either, but it's apparently another threading bug.  If 
you have linuxthreads available for FreeBSD and it's not a pain to use 
them, that would be worth trying out.

Cheers,
Chris

Re: [Sablevm-developer] SableVM hangs during database connection

[Roman K. <ro...@on...>]

Hi,

> >>>>I found another issue, this time related to a database connection with
> >>>>MySQL. During sending an update, the hole connection seems to hang
> >>>>infinitely.
> >>>
> >>>
> >>>Update: I've done a lot of tests, during which I got it running on
> >>>Linux. I used a new MySQL driver for this. Unfortunately my testprog
> >>>does still not run under FreeBSD. I will attach my testprog. 
> >>
> I have no idea either, but it's apparently another threading bug.  If 
> you have linuxthreads available for FreeBSD and it's not a pain to use 
> them, that would be worth trying out.

Grzegorz Prokopski has helped me to narrow the bug. It seems to be an
issue in Classpaths network handling. I'll send a new test-case: a
simple server and a simple client. When trying to send big arrays >22k,
the network layer hangs. At least in FreeBSD.

Running sablevm through strace gives the following:
(last few lines only)
write(2, "sending 220000 bytes...\n", 24) = 24
break(0x9408000)                        = 0
sendto(6, "\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1\1"...,
220000, 0, NULL, 0) = 220000
break(0x93d2000)                        = 0
write(2, "receiving status bytes...\n", 26) = 26
recvfrom(6, 0x91b2070, 1, 0, 0, 0)      = -1 EAGAIN (Resource
temporarily unavailable)
poll([{fd=6, events=POLLRDNORM}], 1, 0) = 0
poll(

Grzegorz Prokopski assumed, that Classpath does not handle EAGAIN
correcly, and doesn't try to recvfrom again, as it should.

I hope this is helpful.

/Roman