EC2 Instance Segmentation Fault Houdini Engine Sim

[kaffee, junge]

Hi all,

I am running sims with houdini engine on amazon aws ec2 instances. Using Thinkbox Deadline and Usage Based Licensing.

It does work, but there is a ~75% chance, the sim won’t start and throw a Segmentation Fault immediately after a message about OpenCL:

2019-11-15 14:26:17:  0: STDOUT: Rendering frame 1 to 240
2019-11-15 14:26:17:  0: STDOUT: OUT_whitewater at time 0
2019-11-15 14:26:17:  0: STDOUT: Bytes sent: 0
2019-11-15 14:26:17:  0: STDOUT: Bytes received: 0
2019-11-15 14:26:17:  0: STDOUT: Seconds taken: 0
2019-11-15 14:26:17:  0: STDOUT: ALF_PROGRESS 0%
2019-11-15 14:26:18:  0: STDOUT: No OpenCL platform has the specified device type (HOUDINI_OCL_DEVICETYPE): GPU.  Falling back to built-in CPU OpenCL driver.
2019-11-15 14:26:18:  0: STDOUT: 8395: Fatal error: Segmentation fault (sent by pid 0)
2019-11-15 14:26:18:  0: STDOUT: -- TRACEBACK BEGIN --
2019-11-15 14:26:18:  0: STDOUT: Traceback from hython 17.5.360 (Compiled on linux-x86_64-gcc6.3):
2019-11-15 14:26:18:  0: STDOUT: stackTrace() <libHoudiniUT.so>
2019-11-15 14:26:18:  0: STDOUT: signalCallback(UTsignalHandlerArg) <libHoudiniUT.so>
2019-11-15 14:26:18:  0: STDOUT: UT_Signal::UT_ComboSignalHandler::operator()(int, siginfo_t*, void*) const <libHoudiniUT.so>
2019-11-15 14:26:18:  0: STDOUT: UT_Signal::processSignal(int, siginfo_t*, void*) <libHoudiniUT.so>
2019-11-15 14:26:18:  0: STDOUT: __funlockfile <libpthread.so.0>
2019-11-15 14:26:18:  0: STDOUT: -- TRACEBACK END --
2019-11-15 14:26:19:  0: INFO: Process exit code: 139
2019-11-15 14:26:19:  0: INFO: Sending EndTaskRequest to S3BackedCacheClient.
2019-11-15 14:26:19:  0: DEBUG: Request:
2019-11-15 14:26:19:  0: DEBUG: 	JobId: 5dceb41cfc813313c877a56e
2019-11-15 14:26:19:  0: Done executing plugin command of type 'Render Task

Since I dont want to run this on a GPU instance, installing Nvidia OpenCL drivers wouldnt do much.
So my guess is, it has something to do with the CPU OpenCL driver.
The instance used here was c5.18xlarge.

Here is some info on the CPUs I found on AWS for the C5 Instance type:

Quote

custom 2nd generation Intel Xeon Scalable Processors (Cascade Lake) or the 1st generation Intel Xeon Platinum 8000 series (Skylake-SP)

 

I also asked this over at the Thinkbox Forums, but I think it is more of a houdini issue.

Thanks!

 

EDIT1: I added

HOUDINI_OCL_DEVICETYPE=CPU

to houdini.env. This gets rid of the Warning:

No OpenCL Platform...

The segmentation Fault remains though.

Here is the full LOG

Edited November 19, 2019 by kaffee, junge
more info