Hi Bryce,
I have been trying to isolate the problem (problems?) that you found here. I have not identified it yet, but I do have some information to pass along.
The problem is intermittent but reproducable. You reported that this fails:
"(1 to: 300) do: [:i | Transcript show: i; cr. CommandShellTestCase run: #testPipeline73]"
which I adapted as follows for testing:
"OSProcess debugMessage: 'start 300 iterations'. (1 to: 300) do: [:i | OSProcess debugMessage: i asString. CommandShell new pipeline: 'ls /no/such/file | stdout nextPutAll: stdin upToEnd !']"
The failure mode is that the image hangs completely, and the VM appears to be blocked on a low level file read. This is probably actually a file descriptor for an OS pipe (pipe readers are set non-blocking by OSProcess, but possibly there is a race condition that I have not identified).
I tested image versions and OSProcess versions, and found: - The problem is not related to image version 3.8 versus 3.9 (fails on both) - The problem is not related to OSProcess version 4.3 versus 4.0.1 (fails on both)
I tested VM and OSPP plugin versions and found: - The combination of recent VM and OSPP fails: Squeak3.8 of ''5 May 2005'' [latest update: #6665] UnixOSProcessPlugin 4 July 2004 (e) version 3.3 - An older combination of VM and OSPP does not fail: Squeak3.7beta of ''1 April 2004'' [latest update: #5923] UnixOSProcessPlugin 4 July 2004 (e) version 3.3
So far this suggests a problem associated with some change in the VM and/or plugins. I have not yet successfully built a combination of newer VM with the older OSPP (due to various annoyances in the build, and I'm out of spare time for now). Hopefully if I can do this, the problem can be narrowed down to either the VM or the plugins (but note that this would not necessarily mean the that VM or plugin has a fault; it could still be a race condition in the image that is just aggrivated by some change in the VM).
There was also some discussion in this thread of possible aio problems. This particular issue does not seem to be related to aoi in the VM or to the AIO plugin.
Hopefully more to follow later,
Dave
On Mon, Jan 29, 2007 at 10:45:57PM +0000, bryce@kampjes.demon.co.uk wrote:
I'm have some problems getting OSProcess 4.3 to run reliably, I'm using a VM with AioPlugin 2.0 and OSProcessPlugin 4.0.1. These are the latest versions availible via SqueakMap.
Most of my testing was done in my standard development image built on the Squeak 3.9 developer image using an Exupery VM. Exupery was loaded but not running. I'm trying to get Exupery's stress test to pass which runs almost all the tests in the image 3 times. Up until now the OSProcess tests have been useful for flushing out context switching bugs.
Any idea how to build a VM with OSProcess that will run 3.9 images reliably?
With both AioPlugin and OSProcessPlugin installed running:
OSPipeTestCase buildSuite run
May lock up the image consuming 100% CPU. This only happens when AioPlugin has been built and installed. It doesn't always happen.
Cannot find new threads: generic error ioFindExternalFunctionIn(primitiveGetThreadID, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveGetThreadID ioFindExternalFunctionIn(primitiveTestEndOfFileFlag, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveTestEndOfFileFlag ioFindExternalFunctionIn(primitiveTestEndOfFileFlag, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveTestEndOfFileFlag
Hitting Alt-. sometimes locks up the image after bringing up a notifier.
When the image has locked up interrupting with gdb shows: (gdb) p printCallStack () 1715565776 >idleProcess 1715534844 [] in >startUp 1715534936 [] in BlockContext>newProcess $5 = 10 (gdb) where #0 0xffffe405 in __kernel_vsyscall () #1 0xf7f0d08d in select () from /lib/libc.so.6 #2 0x0806ab6c in aioPoll (microSeconds=96000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/aio.c:226 #3 0xf7e51743 in display_ioRelinquishProcessorForMicroseconds ( microSeconds=96000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm-display-X11/sqUnixX11.c:2304 #4 0x080523fc in ioRelinquishProcessorForMicroseconds (us=1000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/sqUnixMain.c:477 #5 0x0805be14 in primitiveRelinquishProcessor () at gnu-interp.c:19142 #6 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0x805bde0) at gnu-interp.c:4093 #7 0x08066c57 in interpret () at gnu-interp.c:9080 #8 0x08052019 in main (argc=256, argv=0x0, envp=0x0) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/sqUnixMain.c:1390
Running:
300 timesRepeat: [CommandShellTestCase run: #testPipeline73]
Causes the image to lock up consuming 0% CPU. This happens even when AioPlugin isn't loaded and I've reproduced it in a clean 3.9 image with just OSProcess loaded running the latest Linux VM from squeak.org. My Squeak images normally idle consuming a few percent CPU due to polling.
Cannot find new threads: generic error ioFindExternalFunctionIn(primitiveGetThreadID, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveGetThreadID
0xffffe405 in __kernel_vsyscall () (gdb) where #0 0xffffe405 in __kernel_vsyscall () #1 0xf7e39143 in read () from /lib/libc.so.6 #2 0xf7deb658 in _IO_file_read () from /lib/libc.so.6 #3 0xf7dec83a in _IO_file_underflow () from /lib/libc.so.6 #4 0xf7dee3cd in __underflow () from /lib/libc.so.6 #5 0xf7deb15b in _IO_file_seek () from /lib/libc.so.6 #6 0xf7decfa8 in _IO_sgetn () from /lib/libc.so.6 #7 0xf7de19f0 in fread () from /lib/libc.so.6 #8 0xf7bb2787 in sqFileReadIntoAt (f=0x66412590, count=1, byteArrayIndex=0x66412814 "", startIndex=0) at /home/bryce/squeak/exuperyNew/platforms/Cross/plugins/FilePlugin/sqFilePluginBasicPrims.c:247 #9 0xf7bb12a4 in primitiveFileRead () at /home/bryce/squeak/exuperyNew/src/plugins/FilePlugin/FilePlugin.c:641 #10 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0xf7bb1160) at gnu-interp.c:4093 #11 0x0805c315 in primitiveExternalCall () at gnu-interp.c:15540 #12 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0x805c240) at gnu-interp.c:4093 #13 0x08066c57 in interpret () at gnu-interp.c:9080 #14 0x08052019 in main (argc=0, argv=0x0, envp=0x0) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/sqUnixMain.c:1390 (gdb) p printCallStack () 1715621928 StandardFileStream>basicNext 1715621836 StandardFileStream>next 1715622132 [] in OSPipe>next 1715621628 BlockContext>on:do: 1715621536 OSPipe>next 1715622040 [] in OSPipe>next: 1715621444 Interval>do: 1715621352 OSPipe>next: 1715621720 [] in OSPipe>upToEnd 1715621236 BlockContext>repeat 1715621144 OSPipe>upToEnd 1715620776 [] in PipeableEvaluator>? 1715620636 PipeableEvaluator>blockValue 1715620868 [] in PipeableEvaluator>value 1715620960 [] in PipeableEvaluator>handleRuntimeErrorFor: 1715620524 BlockContext>on:do: 1715620432 PipeableEvaluator>handleRuntimeErrorFor: 1715565264 PipeableEvaluator>value 1715550668 CommandShell>pipeProxy:toCommandList: 1715423904 CommandShell>pipeline: 1715423788 CommandShellTestCase>testPipeline73
Here it looks like the primitive is blocking waiting for input that doesn't come even though it looks to me like it's blocking on a non-blocking file.
Bryce
Hi Dave, Thanks, knowing for sure that the testPipeline bug is a real bug not due to something strange in my VMs (including the stock Linux one from squeak.org) is useful. It means that for my testing of Exupery, it's OK to work around it. I was running a test for 20 minutes that should have been pointless but was causing lockups removing it from the 2 hour stress test with a clear conscience is great.
I'd previously told the stress test to ignore these classes UnixProcessTestCase, AioEventHandlerTestCase, and UnixProcessFileLockingTestCase. So it's possible that the bug has been there for a while especially if UnixProcessUnixFileLockingTestCase was UnixProcessUnixFileLockingTestCase.
It's also worth noting that Exupery's stress test first runs each test class while profiling. If there's a timing issue then it may be effected by the profiler.
To work around the problems I've modified Exupery's stress test to ignore UnixProcessUnixFileLockingTestCase, AbstractUnixProcessFileLockingTestCase, and ExuperyStoryTests as well as the eight classes that it's aways ignored.
The AIO issues only came up when I installed that module trying to get the right configuration to sort out the testPipeline bug. I hadn't investigated that to the same extent as testPipeline.
From where I stand, it's probably better to just fix the bug in the
current development version if it'll be released in the next few months than worry about fixing it in the current versions.
Bryce
David T. Lewis writes:
Hi Bryce,
I have been trying to isolate the problem (problems?) that you found here. I have not identified it yet, but I do have some information to pass along.
The problem is intermittent but reproducable. You reported that this fails:
"(1 to: 300) do: [:i | Transcript show: i; cr. CommandShellTestCase run: #testPipeline73]"
which I adapted as follows for testing:
"OSProcess debugMessage: 'start 300 iterations'. (1 to: 300) do: [:i | OSProcess debugMessage: i asString. CommandShell new pipeline: 'ls /no/such/file | stdout nextPutAll: stdin upToEnd !']"
The failure mode is that the image hangs completely, and the VM appears to be blocked on a low level file read. This is probably actually a file descriptor for an OS pipe (pipe readers are set non-blocking by OSProcess, but possibly there is a race condition that I have not identified).
I tested image versions and OSProcess versions, and found:
- The problem is not related to image version 3.8 versus 3.9 (fails on both)
- The problem is not related to OSProcess version 4.3 versus 4.0.1 (fails on both)
I tested VM and OSPP plugin versions and found:
- The combination of recent VM and OSPP fails: Squeak3.8 of ''5 May 2005'' [latest update: #6665] UnixOSProcessPlugin 4 July 2004 (e) version 3.3
- An older combination of VM and OSPP does not fail: Squeak3.7beta of ''1 April 2004'' [latest update: #5923] UnixOSProcessPlugin 4 July 2004 (e) version 3.3
So far this suggests a problem associated with some change in the VM and/or plugins. I have not yet successfully built a combination of newer VM with the older OSPP (due to various annoyances in the build, and I'm out of spare time for now). Hopefully if I can do this, the problem can be narrowed down to either the VM or the plugins (but note that this would not necessarily mean the that VM or plugin has a fault; it could still be a race condition in the image that is just aggrivated by some change in the VM).
There was also some discussion in this thread of possible aio problems. This particular issue does not seem to be related to aoi in the VM or to the AIO plugin.
Hopefully more to follow later,
Dave
On Mon, Jan 29, 2007 at 10:45:57PM +0000, bryce@kampjes.demon.co.uk wrote:
I'm have some problems getting OSProcess 4.3 to run reliably, I'm using a VM with AioPlugin 2.0 and OSProcessPlugin 4.0.1. These are the latest versions availible via SqueakMap.
Most of my testing was done in my standard development image built on the Squeak 3.9 developer image using an Exupery VM. Exupery was loaded but not running. I'm trying to get Exupery's stress test to pass which runs almost all the tests in the image 3 times. Up until now the OSProcess tests have been useful for flushing out context switching bugs.
Any idea how to build a VM with OSProcess that will run 3.9 images reliably?
With both AioPlugin and OSProcessPlugin installed running:
OSPipeTestCase buildSuite run
May lock up the image consuming 100% CPU. This only happens when AioPlugin has been built and installed. It doesn't always happen.
Cannot find new threads: generic error ioFindExternalFunctionIn(primitiveGetThreadID, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveGetThreadID ioFindExternalFunctionIn(primitiveTestEndOfFileFlag, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveTestEndOfFileFlag ioFindExternalFunctionIn(primitiveTestEndOfFileFlag, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveTestEndOfFileFlag
Hitting Alt-. sometimes locks up the image after bringing up a notifier.
When the image has locked up interrupting with gdb shows: (gdb) p printCallStack () 1715565776 >idleProcess 1715534844 [] in >startUp 1715534936 [] in BlockContext>newProcess $5 = 10 (gdb) where #0 0xffffe405 in __kernel_vsyscall () #1 0xf7f0d08d in select () from /lib/libc.so.6 #2 0x0806ab6c in aioPoll (microSeconds=96000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/aio.c:226 #3 0xf7e51743 in display_ioRelinquishProcessorForMicroseconds ( microSeconds=96000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm-display-X11/sqUnixX11.c:2304 #4 0x080523fc in ioRelinquishProcessorForMicroseconds (us=1000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/sqUnixMain.c:477 #5 0x0805be14 in primitiveRelinquishProcessor () at gnu-interp.c:19142 #6 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0x805bde0) at gnu-interp.c:4093 #7 0x08066c57 in interpret () at gnu-interp.c:9080 #8 0x08052019 in main (argc=256, argv=0x0, envp=0x0) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/sqUnixMain.c:1390
Running:
300 timesRepeat: [CommandShellTestCase run: #testPipeline73]
Causes the image to lock up consuming 0% CPU. This happens even when AioPlugin isn't loaded and I've reproduced it in a clean 3.9 image with just OSProcess loaded running the latest Linux VM from squeak.org. My Squeak images normally idle consuming a few percent CPU due to polling.
Cannot find new threads: generic error ioFindExternalFunctionIn(primitiveGetThreadID, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveGetThreadID
0xffffe405 in __kernel_vsyscall () (gdb) where #0 0xffffe405 in __kernel_vsyscall () #1 0xf7e39143 in read () from /lib/libc.so.6 #2 0xf7deb658 in _IO_file_read () from /lib/libc.so.6 #3 0xf7dec83a in _IO_file_underflow () from /lib/libc.so.6 #4 0xf7dee3cd in __underflow () from /lib/libc.so.6 #5 0xf7deb15b in _IO_file_seek () from /lib/libc.so.6 #6 0xf7decfa8 in _IO_sgetn () from /lib/libc.so.6 #7 0xf7de19f0 in fread () from /lib/libc.so.6 #8 0xf7bb2787 in sqFileReadIntoAt (f=0x66412590, count=1, byteArrayIndex=0x66412814 "", startIndex=0) at /home/bryce/squeak/exuperyNew/platforms/Cross/plugins/FilePlugin/sqFilePluginBasicPrims.c:247 #9 0xf7bb12a4 in primitiveFileRead () at /home/bryce/squeak/exuperyNew/src/plugins/FilePlugin/FilePlugin.c:641 #10 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0xf7bb1160) at gnu-interp.c:4093 #11 0x0805c315 in primitiveExternalCall () at gnu-interp.c:15540 #12 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0x805c240) at gnu-interp.c:4093 #13 0x08066c57 in interpret () at gnu-interp.c:9080 #14 0x08052019 in main (argc=0, argv=0x0, envp=0x0) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/sqUnixMain.c:1390 (gdb) p printCallStack () 1715621928 StandardFileStream>basicNext 1715621836 StandardFileStream>next 1715622132 [] in OSPipe>next 1715621628 BlockContext>on:do: 1715621536 OSPipe>next 1715622040 [] in OSPipe>next: 1715621444 Interval>do: 1715621352 OSPipe>next: 1715621720 [] in OSPipe>upToEnd 1715621236 BlockContext>repeat 1715621144 OSPipe>upToEnd 1715620776 [] in PipeableEvaluator>? 1715620636 PipeableEvaluator>blockValue 1715620868 [] in PipeableEvaluator>value 1715620960 [] in PipeableEvaluator>handleRuntimeErrorFor: 1715620524 BlockContext>on:do: 1715620432 PipeableEvaluator>handleRuntimeErrorFor: 1715565264 PipeableEvaluator>value 1715550668 CommandShell>pipeProxy:toCommandList: 1715423904 CommandShell>pipeline: 1715423788 CommandShellTestCase>testPipeline73
Here it looks like the primitive is blocking waiting for input that doesn't come even though it looks to me like it's blocking on a non-blocking file.
Bryce
I am moving this discussion to vm-dev, because it is unlikely to be of interest on the general list.
To add to my last message, I have now build a newer VM with an old OSProcessPlugin: Squeak3.8 of ''5 May 2005'' [latest update: #6665] installed plugin UnixOSProcessPlugin 6 February 2007 (e) version 3.3
The problem still exists when running 300 iterations of: OSProcess debugMessage: 'start 300 iterations'. (1 to: 300) do: [:i | OSProcess debugMessage: i asString. CommandShell new pipeline: 'ls /no/such/file | stdout nextPutAll: stdin upToEnd !']
This give us:
Older VM plus older OSPP ==> no failures Recent VM plus recent OSPP ==> intermittent failure after about 100 iterations Recent VM plus older OSPP ==> intermittent failure after about 100 iterations
The problem is also unrelated to image version and unrelated to OSProcess version. That seems to narrow it down to something changing in the VM (support code and/or VMMaker). That's purely correlational; it does not mean there is something wrong in the VM, rather that something changed that either provokes the bug in OSProcess or that OSProcess puts enough load on the VM to provoke the bug. At this point, I have no clue as to which one it is.
Bryce's original diagnosis was right on target:
Here it looks like the primitive is blocking waiting for input that doesn't come even though it looks to me like it's blocking on a non-blocking file.
Bryce
After getting the VM to hang up like this, I can interrupt from gdb and get a stack trace that makes absolutely no sense to me (particularly since the test case does not do any file opens whatsoever). I'm no expert on gdb, so if anyone give me a pointer as to what this means, I'd be grateful. Line 600 of FilePlugin.c is "interpreterProxy->popthenPush(3, filePointer);" at the end of primitiveFileOpen(). It acts as if it's being called recursively, which is impossible, and by the time we get to sqFileReadIntoAt(), we are dealing with what looks like a bad case of scrambled brains. My test is running on Linux, so there should be no pthreads involved. The stack trace is attached below.
Dave
stack dump of VM interrupted from gdb:
(gdb) run TestImageHang.1 [New Thread 16384 (LWP 5753)] ^C Program received signal SIGINT, Interrupt. [Switching to Thread 16384 (LWP 5753)] 0x4013d088 in read () from /lib/i686/libc.so.6 (gdb) bt #0 0x4013d088 in read () from /lib/i686/libc.so.6 #1 0x4019dfd8 in __DTOR_END__ () from /lib/i686/libc.so.6 #2 0x400dc232 in _IO_file_read_internal () from /lib/i686/libc.so.6 #3 0x400dafa1 in _IO_new_file_underflow () from /lib/i686/libc.so.6 #4 0x400dcfc6 in __underflow () from /lib/i686/libc.so.6 #5 0x400dc70a in _IO_file_xsgetn_internal () from /lib/i686/libc.so.6 #6 0x400dd3e8 in _IO_sgetn_internal () from /lib/i686/libc.so.6 #7 0x400d0942 in fread () from /lib/i686/libc.so.6 #8 0x080a2499 in sqFileReadIntoAt (f=0x4019dfd8, count=1, byteArrayIndex=0x1 <Address 0x1 out of bounds>, startIndex=4294966784) at /data3/lewis/squeak/sq/Squeak64BitPort/images/platforms/Cross/plugins/FilePlugin/sqFilePluginBasicPrims.c:247 #9 0x080a19ad in primitiveFileRead () at /data3/lewis/squeak/sq/Squeak64BitPort/images/src/vm/intplugins/FilePlugin/FilePlugin.c:641 #10 0x00000010 in ?? () #11 0x41a6d770 in ?? () #12 0x00000001 in ?? () #13 0x40b1cc5c in ?? () #14 0x080a1880 in primitiveFileOpen () at /data3/lewis/squeak/sq/Squeak64BitPort/images/src/vm/intplugins/FilePlugin/FilePlugin.c:600 #15 0x40b1cc5c in ?? () #16 0x080a1880 in primitiveFileOpen () at /data3/lewis/squeak/sq/Squeak64BitPort/images/src/vm/intplugins/FilePlugin/FilePlugin.c:600 #17 0x41317be0 in ?? () #18 0x000003f2 in ?? () #19 0x080a1880 in primitiveFileOpen () at /data3/lewis/squeak/sq/Squeak64BitPort/images/src/vm/intplugins/FilePlugin/FilePlugin.c:600 #20 0x41a8c698 in ?? () #21 0x40b1c61e in ?? () #22 0x41a8c6c8 in ?? () #23 0x41a8c698 in ?? () #24 0x0805e4a7 in dispatchFunctionPointer (aFunctionPointer=0x8067b50) at gnu-interp.c:4024 #25 0x08062990 in interpret () at gnu-interp.c:7902 #26 0x0805c04c in main (argc=-1073745031, argv=0x0, envp=0xfffffe00) at /data3/lewis/squeak/sq/Squeak64BitPort/images/platforms/unix/vm/sqUnixMain.c:1388 (gdb)
On Tue, Feb 06, 2007 at 07:09:05AM -0500, David T. Lewis wrote:
Hi Bryce,
I have been trying to isolate the problem (problems?) that you found here. I have not identified it yet, but I do have some information to pass along.
The problem is intermittent but reproducable. You reported that this fails:
"(1 to: 300) do: [:i | Transcript show: i; cr. CommandShellTestCase run: #testPipeline73]"
which I adapted as follows for testing:
"OSProcess debugMessage: 'start 300 iterations'. (1 to: 300) do: [:i | OSProcess debugMessage: i asString. CommandShell new pipeline: 'ls /no/such/file | stdout nextPutAll: stdin upToEnd !']"
The failure mode is that the image hangs completely, and the VM appears to be blocked on a low level file read. This is probably actually a file descriptor for an OS pipe (pipe readers are set non-blocking by OSProcess, but possibly there is a race condition that I have not identified).
I tested image versions and OSProcess versions, and found:
- The problem is not related to image version 3.8 versus 3.9 (fails on both)
- The problem is not related to OSProcess version 4.3 versus 4.0.1 (fails on both)
I tested VM and OSPP plugin versions and found:
- The combination of recent VM and OSPP fails: Squeak3.8 of ''5 May 2005'' [latest update: #6665] UnixOSProcessPlugin 4 July 2004 (e) version 3.3
- An older combination of VM and OSPP does not fail: Squeak3.7beta of ''1 April 2004'' [latest update: #5923] UnixOSProcessPlugin 4 July 2004 (e) version 3.3
So far this suggests a problem associated with some change in the VM and/or plugins. I have not yet successfully built a combination of newer VM with the older OSPP (due to various annoyances in the build, and I'm out of spare time for now). Hopefully if I can do this, the problem can be narrowed down to either the VM or the plugins (but note that this would not necessarily mean the that VM or plugin has a fault; it could still be a race condition in the image that is just aggrivated by some change in the VM).
There was also some discussion in this thread of possible aio problems. This particular issue does not seem to be related to aoi in the VM or to the AIO plugin.
Hopefully more to follow later,
Dave
On Mon, Jan 29, 2007 at 10:45:57PM +0000, bryce@kampjes.demon.co.uk wrote:
I'm have some problems getting OSProcess 4.3 to run reliably, I'm using a VM with AioPlugin 2.0 and OSProcessPlugin 4.0.1. These are the latest versions availible via SqueakMap.
Most of my testing was done in my standard development image built on the Squeak 3.9 developer image using an Exupery VM. Exupery was loaded but not running. I'm trying to get Exupery's stress test to pass which runs almost all the tests in the image 3 times. Up until now the OSProcess tests have been useful for flushing out context switching bugs.
Any idea how to build a VM with OSProcess that will run 3.9 images reliably?
With both AioPlugin and OSProcessPlugin installed running:
OSPipeTestCase buildSuite run
May lock up the image consuming 100% CPU. This only happens when AioPlugin has been built and installed. It doesn't always happen.
Cannot find new threads: generic error ioFindExternalFunctionIn(primitiveGetThreadID, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveGetThreadID ioFindExternalFunctionIn(primitiveTestEndOfFileFlag, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveTestEndOfFileFlag ioFindExternalFunctionIn(primitiveTestEndOfFileFlag, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveTestEndOfFileFlag
Hitting Alt-. sometimes locks up the image after bringing up a notifier.
When the image has locked up interrupting with gdb shows: (gdb) p printCallStack () 1715565776 >idleProcess 1715534844 [] in >startUp 1715534936 [] in BlockContext>newProcess $5 = 10 (gdb) where #0 0xffffe405 in __kernel_vsyscall () #1 0xf7f0d08d in select () from /lib/libc.so.6 #2 0x0806ab6c in aioPoll (microSeconds=96000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/aio.c:226 #3 0xf7e51743 in display_ioRelinquishProcessorForMicroseconds ( microSeconds=96000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm-display-X11/sqUnixX11.c:2304 #4 0x080523fc in ioRelinquishProcessorForMicroseconds (us=1000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/sqUnixMain.c:477 #5 0x0805be14 in primitiveRelinquishProcessor () at gnu-interp.c:19142 #6 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0x805bde0) at gnu-interp.c:4093 #7 0x08066c57 in interpret () at gnu-interp.c:9080 #8 0x08052019 in main (argc=256, argv=0x0, envp=0x0) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/sqUnixMain.c:1390
Running:
300 timesRepeat: [CommandShellTestCase run: #testPipeline73]
Causes the image to lock up consuming 0% CPU. This happens even when AioPlugin isn't loaded and I've reproduced it in a clean 3.9 image with just OSProcess loaded running the latest Linux VM from squeak.org. My Squeak images normally idle consuming a few percent CPU due to polling.
Cannot find new threads: generic error ioFindExternalFunctionIn(primitiveGetThreadID, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/UnixOSProcessPlugin: undefined symbol: primitiveGetThreadID
0xffffe405 in __kernel_vsyscall () (gdb) where #0 0xffffe405 in __kernel_vsyscall () #1 0xf7e39143 in read () from /lib/libc.so.6 #2 0xf7deb658 in _IO_file_read () from /lib/libc.so.6 #3 0xf7dec83a in _IO_file_underflow () from /lib/libc.so.6 #4 0xf7dee3cd in __underflow () from /lib/libc.so.6 #5 0xf7deb15b in _IO_file_seek () from /lib/libc.so.6 #6 0xf7decfa8 in _IO_sgetn () from /lib/libc.so.6 #7 0xf7de19f0 in fread () from /lib/libc.so.6 #8 0xf7bb2787 in sqFileReadIntoAt (f=0x66412590, count=1, byteArrayIndex=0x66412814 "", startIndex=0) at /home/bryce/squeak/exuperyNew/platforms/Cross/plugins/FilePlugin/sqFilePluginBasicPrims.c:247 #9 0xf7bb12a4 in primitiveFileRead () at /home/bryce/squeak/exuperyNew/src/plugins/FilePlugin/FilePlugin.c:641 #10 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0xf7bb1160) at gnu-interp.c:4093 #11 0x0805c315 in primitiveExternalCall () at gnu-interp.c:15540 #12 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0x805c240) at gnu-interp.c:4093 #13 0x08066c57 in interpret () at gnu-interp.c:9080 #14 0x08052019 in main (argc=0, argv=0x0, envp=0x0) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/sqUnixMain.c:1390 (gdb) p printCallStack () 1715621928 StandardFileStream>basicNext 1715621836 StandardFileStream>next 1715622132 [] in OSPipe>next 1715621628 BlockContext>on:do: 1715621536 OSPipe>next 1715622040 [] in OSPipe>next: 1715621444 Interval>do: 1715621352 OSPipe>next: 1715621720 [] in OSPipe>upToEnd 1715621236 BlockContext>repeat 1715621144 OSPipe>upToEnd 1715620776 [] in PipeableEvaluator>? 1715620636 PipeableEvaluator>blockValue 1715620868 [] in PipeableEvaluator>value 1715620960 [] in PipeableEvaluator>handleRuntimeErrorFor: 1715620524 BlockContext>on:do: 1715620432 PipeableEvaluator>handleRuntimeErrorFor: 1715565264 PipeableEvaluator>value 1715550668 CommandShell>pipeProxy:toCommandList: 1715423904 CommandShell>pipeline: 1715423788 CommandShellTestCase>testPipeline73
Here it looks like the primitive is blocking waiting for input that doesn't come even though it looks to me like it's blocking on a non-blocking file.
Bryce
Mmm startIndex=4294966784 arrayIndex = 0x1?
Would those be valid?
#8 0x080a2499 in sqFileReadIntoAt (f=0x4019dfd8, count=1, byteArrayIndex=0x1 <Address 0x1 out of bounds>, startIndex=4294966784) at /data3/lewis/squeak/sq/Squeak64BitPort/ images/platforms/Cross/plugins/FilePlugin/sqFilePluginBasicPrims.c:247
arrayIndex = interpreterProxy->firstIndexableField(array); bytesRead = sqFileReadIntoAt(file, count * byteSize, arrayIndex, (startIndex - 1) * byteSize);
size_t sqFileReadIntoAt(SQFile *f, size_t count, char* byteArrayIndex, size_t startIndex) { /* Read count bytes from the given file into byteArray starting at startIndex. byteArray is the address of the first byte of a Squeak bytes object (e.g. String or ByteArray). startIndex is a zero-based index; that is a startIndex of 0 starts writing at the first byte of byteArray. */
char *dst; size_t bytesRead; FILE *file;
if (!sqFileValid(f)) return interpreterProxy->success(false); file= getFile(f); if (f->writable && (f->lastOp == WRITE_OP)) fseek(file, 0, SEEK_CUR); /* seek between writing and reading */ dst = byteArrayIndex + startIndex; bytesRead = fread(dst, 1, count, file);
On Feb 10, 2007, at 4:23 PM, David T. Lewis wrote:
I am moving this discussion to vm-dev, because it is unlikely to be of interest on the general list.
To add to my last message, I have now build a newer VM with an old OSProcessPlugin: Squeak3.8 of ''5 May 2005'' [latest update: #6665] installed plugin UnixOSProcessPlugin 6 February 2007 (e) version 3.3
The problem still exists when running 300 iterations of: OSProcess debugMessage: 'start 300 iterations'. (1 to: 300) do: [:i | OSProcess debugMessage: i asString. CommandShell new pipeline: 'ls /no/such/file | stdout nextPutAll: stdin upToEnd !']
This give us:
Older VM plus older OSPP ==> no failures Recent VM plus recent OSPP ==> intermittent failure after about 100 iterations Recent VM plus older OSPP ==> intermittent failure after about 100 iterations
The problem is also unrelated to image version and unrelated to OSProcess version. That seems to narrow it down to something changing in the VM (support code and/or VMMaker). That's purely correlational; it does not mean there is something wrong in the VM, rather that something changed that either provokes the bug in OSProcess or that OSProcess puts enough load on the VM to provoke the bug. At this point, I have no clue as to which one it is.
Bryce's original diagnosis was right on target:
Here it looks like the primitive is blocking waiting for input that doesn't come even though it looks to me like it's blocking on a non-blocking file.
Bryce
After getting the VM to hang up like this, I can interrupt from gdb and get a stack trace that makes absolutely no sense to me (particularly since the test case does not do any file opens whatsoever). I'm no expert on gdb, so if anyone give me a pointer as to what this means, I'd be grateful. Line 600 of FilePlugin.c is "interpreterProxy->popthenPush(3, filePointer);" at the end of primitiveFileOpen(). It acts as if it's being called recursively, which is impossible, and by the time we get to sqFileReadIntoAt(), we are dealing with what looks like a bad case of scrambled brains. My test is running on Linux, so there should be no pthreads involved. The stack trace is attached below.
Dave
stack dump of VM interrupted from gdb:
(gdb) run TestImageHang.1 [New Thread 16384 (LWP 5753)] ^C Program received signal SIGINT, Interrupt. [Switching to Thread 16384 (LWP 5753)] 0x4013d088 in read () from /lib/i686/libc.so.6 (gdb) bt #0 0x4013d088 in read () from /lib/i686/libc.so.6 #1 0x4019dfd8 in __DTOR_END__ () from /lib/i686/libc.so.6 #2 0x400dc232 in _IO_file_read_internal () from /lib/i686/libc.so.6 #3 0x400dafa1 in _IO_new_file_underflow () from /lib/i686/libc.so.6 #4 0x400dcfc6 in __underflow () from /lib/i686/libc.so.6 #5 0x400dc70a in _IO_file_xsgetn_internal () from /lib/i686/libc.so.6 #6 0x400dd3e8 in _IO_sgetn_internal () from /lib/i686/libc.so.6 #7 0x400d0942 in fread () from /lib/i686/libc.so.6 #8 0x080a2499 in sqFileReadIntoAt (f=0x4019dfd8, count=1, byteArrayIndex=0x1 <Address 0x1 out of bounds>, startIndex=4294966784) at /data3/lewis/squeak/sq/Squeak64BitPort/ images/platforms/Cross/plugins/FilePlugin/sqFilePluginBasicPrims.c:247 #9 0x080a19ad in primitiveFileRead () at /data3/lewis/squeak/sq/ Squeak64BitPort/images/src/vm/intplugins/FilePlugin/FilePlugin.c:641 #10 0x00000010 in ?? () #11 0x41a6d770 in ?? () #12 0x00000001 in ?? () #13 0x40b1cc5c in ?? () #14 0x080a1880 in primitiveFileOpen () at /data3/lewis/squeak/sq/ Squeak64BitPort/images/src/vm/intplugins/FilePlugin/FilePlugin.c:600 #15 0x40b1cc5c in ?? () #16 0x080a1880 in primitiveFileOpen () at /data3/lewis/squeak/sq/ Squeak64BitPort/images/src/vm/intplugins/FilePlugin/FilePlugin.c:600 #17 0x41317be0 in ?? () #18 0x000003f2 in ?? () #19 0x080a1880 in primitiveFileOpen () at /data3/lewis/squeak/sq/ Squeak64BitPort/images/src/vm/intplugins/FilePlugin/FilePlugin.c:600 #20 0x41a8c698 in ?? () #21 0x40b1c61e in ?? () #22 0x41a8c6c8 in ?? () #23 0x41a8c698 in ?? () #24 0x0805e4a7 in dispatchFunctionPointer (aFunctionPointer=0x8067b50) at gnu-interp.c:4024 #25 0x08062990 in interpret () at gnu-interp.c:7902 #26 0x0805c04c in main (argc=-1073745031, argv=0x0, envp=0xfffffe00) at /data3/lewis/squeak/sq/Squeak64BitPort/images/ platforms/unix/vm/sqUnixMain.c:1388 (gdb)
On Tue, Feb 06, 2007 at 07:09:05AM -0500, David T. Lewis wrote:
Hi Bryce,
I have been trying to isolate the problem (problems?) that you found here. I have not identified it yet, but I do have some information to pass along.
The problem is intermittent but reproducable. You reported that this fails:
"(1 to: 300) do: [:i | Transcript show: i; cr. CommandShellTestCase run: #testPipeline73]"
which I adapted as follows for testing:
"OSProcess debugMessage: 'start 300 iterations'. (1 to: 300) do: [:i | OSProcess debugMessage: i asString. CommandShell new pipeline: 'ls /no/such/file | stdout nextPutAll: stdin upToEnd !']"
The failure mode is that the image hangs completely, and the VM appears to be blocked on a low level file read. This is probably actually a file descriptor for an OS pipe (pipe readers are set non-blocking by OSProcess, but possibly there is a race condition that I have not identified).
I tested image versions and OSProcess versions, and found:
- The problem is not related to image version 3.8 versus 3.9
(fails on both)
- The problem is not related to OSProcess version 4.3 versus 4.0.1
(fails on both)
I tested VM and OSPP plugin versions and found:
- The combination of recent VM and OSPP fails: Squeak3.8 of ''5 May 2005'' [latest update: #6665] UnixOSProcessPlugin 4 July 2004 (e) version 3.3
- An older combination of VM and OSPP does not fail: Squeak3.7beta of ''1 April 2004'' [latest update: #5923] UnixOSProcessPlugin 4 July 2004 (e) version 3.3
So far this suggests a problem associated with some change in the VM and/or plugins. I have not yet successfully built a combination of newer VM with the older OSPP (due to various annoyances in the build, and I'm out of spare time for now). Hopefully if I can do this, the problem can be narrowed down to either the VM or the plugins (but note that this would not necessarily mean the that VM or plugin has a fault; it could still be a race condition in the image that is just aggrivated by some change in the VM).
There was also some discussion in this thread of possible aio problems. This particular issue does not seem to be related to aoi in the VM or to the AIO plugin.
Hopefully more to follow later,
Dave
On Mon, Jan 29, 2007 at 10:45:57PM +0000, bryce@kampjes.demon.co.uk wrote:
I'm have some problems getting OSProcess 4.3 to run reliably, I'm using a VM with AioPlugin 2.0 and OSProcessPlugin 4.0.1. These are the latest versions availible via SqueakMap.
Most of my testing was done in my standard development image built on the Squeak 3.9 developer image using an Exupery VM. Exupery was loaded but not running. I'm trying to get Exupery's stress test to pass which runs almost all the tests in the image 3 times. Up until now the OSProcess tests have been useful for flushing out context switching bugs.
Any idea how to build a VM with OSProcess that will run 3.9 images reliably?
With both AioPlugin and OSProcessPlugin installed running:
OSPipeTestCase buildSuite run
May lock up the image consuming 100% CPU. This only happens when AioPlugin has been built and installed. It doesn't always happen.
Cannot find new threads: generic error ioFindExternalFunctionIn(primitiveGetThreadID, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/ UnixOSProcessPlugin: undefined symbol: primitiveGetThreadID ioFindExternalFunctionIn(primitiveTestEndOfFileFlag, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/ UnixOSProcessPlugin: undefined symbol: primitiveTestEndOfFileFlag ioFindExternalFunctionIn(primitiveTestEndOfFileFlag, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/ UnixOSProcessPlugin: undefined symbol: primitiveTestEndOfFileFlag
Hitting Alt-. sometimes locks up the image after bringing up a notifier.
When the image has locked up interrupting with gdb shows: (gdb) p printCallStack () 1715565776 >idleProcess 1715534844 [] in >startUp 1715534936 [] in BlockContext>newProcess $5 = 10 (gdb) where #0 0xffffe405 in __kernel_vsyscall () #1 0xf7f0d08d in select () from /lib/libc.so.6 #2 0x0806ab6c in aioPoll (microSeconds=96000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/aio.c:226 #3 0xf7e51743 in display_ioRelinquishProcessorForMicroseconds ( microSeconds=96000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm-display- X11/sqUnixX11.c:2304 #4 0x080523fc in ioRelinquishProcessorForMicroseconds (us=1000) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/ sqUnixMain.c:477 #5 0x0805be14 in primitiveRelinquishProcessor () at gnu-interp.c: 19142 #6 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0x805bde0) at gnu-interp.c:4093 #7 0x08066c57 in interpret () at gnu-interp.c:9080 #8 0x08052019 in main (argc=256, argv=0x0, envp=0x0) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/sqUnixMain.c:1390
Running:
300 timesRepeat: [CommandShellTestCase run: #testPipeline73]
Causes the image to lock up consuming 0% CPU. This happens even when AioPlugin isn't loaded and I've reproduced it in a clean 3.9 image with just OSProcess loaded running the latest Linux VM from squeak.org. My Squeak images normally idle consuming a few percent CPU due to polling.
Cannot find new threads: generic error ioFindExternalFunctionIn(primitiveGetThreadID, 0x80a4e88): /home/bryce/squeak/exuperyNew/build/UnixOSProcessPlugin/.libs/ UnixOSProcessPlugin: undefined symbol: primitiveGetThreadID
0xffffe405 in __kernel_vsyscall () (gdb) where #0 0xffffe405 in __kernel_vsyscall () #1 0xf7e39143 in read () from /lib/libc.so.6 #2 0xf7deb658 in _IO_file_read () from /lib/libc.so.6 #3 0xf7dec83a in _IO_file_underflow () from /lib/libc.so.6 #4 0xf7dee3cd in __underflow () from /lib/libc.so.6 #5 0xf7deb15b in _IO_file_seek () from /lib/libc.so.6 #6 0xf7decfa8 in _IO_sgetn () from /lib/libc.so.6 #7 0xf7de19f0 in fread () from /lib/libc.so.6 #8 0xf7bb2787 in sqFileReadIntoAt (f=0x66412590, count=1, byteArrayIndex=0x66412814 "", startIndex=0) at /home/bryce/squeak/exuperyNew/platforms/Cross/plugins/ FilePlugin/sqFilePluginBasicPrims.c:247 #9 0xf7bb12a4 in primitiveFileRead () at /home/bryce/squeak/exuperyNew/src/plugins/FilePlugin/ FilePlugin.c:641 #10 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0xf7bb1160) at gnu-interp.c:4093 #11 0x0805c315 in primitiveExternalCall () at gnu-interp.c:15540 #12 0x080535f7 in dispatchFunctionPointer (aFunctionPointer=0x805c240) at gnu-interp.c:4093 #13 0x08066c57 in interpret () at gnu-interp.c:9080 #14 0x08052019 in main (argc=0, argv=0x0, envp=0x0) at /home/bryce/squeak/exuperyNew/platforms/unix/vm/ sqUnixMain.c:1390 (gdb) p printCallStack () 1715621928 StandardFileStream>basicNext 1715621836 StandardFileStream>next 1715622132 [] in OSPipe>next 1715621628 BlockContext>on:do: 1715621536 OSPipe>next 1715622040 [] in OSPipe>next: 1715621444 Interval>do: 1715621352 OSPipe>next: 1715621720 [] in OSPipe>upToEnd 1715621236 BlockContext>repeat 1715621144 OSPipe>upToEnd 1715620776 [] in PipeableEvaluator>? 1715620636 PipeableEvaluator>blockValue 1715620868 [] in PipeableEvaluator>value 1715620960 [] in PipeableEvaluator>handleRuntimeErrorFor: 1715620524 BlockContext>on:do: 1715620432 PipeableEvaluator>handleRuntimeErrorFor: 1715565264 PipeableEvaluator>value 1715550668 CommandShell>pipeProxy:toCommandList: 1715423904 CommandShell>pipeline: 1715423788 CommandShellTestCase>testPipeline73
Here it looks like the primitive is blocking waiting for input that doesn't come even though it looks to me like it's blocking on a non-blocking file.
Bryce
-- ======================================================================== === John M. McIntosh johnmci@smalltalkconsulting.com Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================== ===
mmm I'll futher note that the file primitives haven't really been reviewed closely enough for issues with 64bit math since
primitiveFileRead
count = interpreterProxy->positive32BitValueOf(interpreterProxy-
stackValue(0));
startIndex = interpreterProxy->positive32BitValueOf(interpreterProxy-
stackValue(1));
Technically I could want to read with an count more than 32bits of data into a location offset > 32bits....
On Feb 10, 2007, at 4:23 PM, David T. Lewis wrote:
-- ======================================================================== === John M. McIntosh johnmci@smalltalkconsulting.com Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================== ===
vm-dev@lists.squeakfoundation.org