Fatal Error: Caught ORCHESTRATE exception in parallel proces

[martinkirouac]

Hi Everyone, I have a problem with the install I have on a 2 nodes windows server 2003. Install looks fine (rsh -l dstage IS8node2 ls works on both node) and ssh RSA key is good. I build a basic job with a row gen, transformer and peek ... it works fine on 1 node but with 2 nodes I have the following error :

1 ) A WARNING :
Transformer_8,1: Warning: library "V0S8_TransformerMultinode_Transformer_8" (for class "APT_TransformOperatorImplV0S8_TransformerMultinode_Transformer_8") failed to load: Could not find "V0S8_TransformerMultinode_Transformer_8" on the library search path defined by the environment variable PATH;
path: /C=/IBM/InformationServer/Server/PXEngine/bin:
/C=/IBM/InformationServer/Server/PXEngine/osh_wrappers:
/C=/IBM/InformationServer/Server/PXEngine/user_osh_wrappers:
/C=/IBM/InformationServer/Server/PXEngine/user_osh_wrappers:
/C=/IBM/InformationServer/Server/DSComponents/bin:
/C=/IBM/InformationServer/Server/DSParallel:
/C=/IBM/InformationServer/Server/Projects/tmpproject/RT_BP5.O:
/C=/IBM/InformationServer/Server/DSEngine/bin:
/C=/IBM/InformationServer/Server/Projects/tmpproject/buildop:
/C=/IBM/InformationServer/Server/Projects/tmpproject/wrapped:
/C=/PROGRA~1/MKSTOO~1/mksnt:
/C=/PROGRA~1/MKSTOO~1/bin:
/C=/PROGRA~1/MKSTOO~1/bin/X11:
/C=/PROGRA~1/MKSTOO~1/mksnt:
/C=/IBM/InformationServer/ASBNode/apps/jre/bin/classic:
/C=/IBM/InformationServer/ASBNode/lib/cpp:
/C=/IBM/InformationServer/ASBNode/apps/proxy/cpp/vc60/MT_dll/bin:
/C=/WINDOWS/system32:
/C=/WINDOWS:
/C=/WINDOWS/System32/Wbem:
/C=/IBM/SQLLIB/BIN:
/C=/IBM/SQLLIB/FUNCTION:
/C=/IBM/SQLLIB/SAMPLES/REPL:
/C=/IBM/SQLLIB/BIN.

2 ) ERROR
Transformer_8,1: Fatal Error: Caught ORCHESTRATE exception in parallel process: APT_Archive::BadClass: From: Loading memory archive: Could not find "APT_TransformOperatorImplV0S8_TransformerMultinode_Transformer_8".

node_node2: Player 1 terminated unexpectedly.

It seems like the compiled transformer is not seen by the second node.
If anyone has any idea, I would appreciate.

Thank you very much,

[ray.wurlod]

Welcome aboard.

I think your diagnosis is correct. Is the DataStage project visible from the second machine? That's where the compiled Transformer code is. If the project is not visible, you need to propagate the code to all machines in the configuration - this is explained in (from memory) the Parallel Job Developer's Guide.

[martinkirouac]

Hi Ray,

Thanks for the tip, I did found some information on p.604 of the Parallel Job Developer's Guide (Chapter about remote deploying). I've been re-reading it and some things are still unclear to me. I was hoping you could help me understand the process.

1 ) When you create a new project, do you have to copy it to each node or is there a better way to setup this on a multi-nodes configuration ? Doing that this way, I don't see the project when I open the designer (ie. IS8NODE2/testproject). I am going through the whole doc looking for this topic... I probably miss this somewhere. Maybe that's what you said by propagating the project... I did try just in case and I had an error at compile time. I will send details if this is supposed to be the way to do it.

2 ) I am developing at the moment and trying to avoid bad behavior in the job due to bad partitioning or anything that I could do by accident and results in bad calculation for instance. So I would prefer to stay inside the designer/director to design, launch and analyze the jobs. The only way I found last night to make it work is to copy manually the content of the "deployment base directory" (as specified in the remote tab in the administrator) to the node 2. This seems a bit unorthodox to me so I think I might be wrong on the process since I've read that the only thing I should care as a DS developer is the configuration file. Is this normal or I shouldn't have to do anything manually (or using command prompt).

Thank you very much!

[martinkirouac]

Hi Everyone. Thanks to my friends from IBM, I've managed to solve the problem. Since I still didn't find that in the documentation, I'd figure it would be nice to share my experience.

So, first thing to know... DataStage doesn't propagate the compile code through the MMP. So when you compile a job, if you have 2 nodes it seems that project on node 1 is not copied on project on node 2. I found that by copying the project folder across the nodes. The IBM Sales Engineer told me that most companies are creating a network share between the nodes. That would solve the problem. So far, it works for me but can't find anything specified about this in the documentation. There might be a better solution.

If anyone has a constructive comment or alternate solution, you are more than welcome to contact me.

Thanks!