canonical_etypes should be used in dgl.distributed.partition_graph

[Sherler]

🐛 Bug

To Reproduce

Steps to reproduce the behavior:

1. define two edge types with same 'middle name': (person,loves,person) ,(person, loves, pet)
2. create partition_graph using dgl.distributed.partition_graph
3. there will be an exception:

dgl._ffi.base.DGLError: Edge type "loves" is ambiguous. Please use canonical edge type in the form of (srctype, etype, dsttype)

Environment

• DGL Version (e.g., 1.0): 0.7.1
• Backend Library & Version (e.g., PyTorch 0.4.1, MXNet/Gluon 1.3):
• OS (e.g., Linux):
• How you installed DGL (conda, pip, source):
• Build command you used (if compiling from source):
• Python version:
• CUDA/cuDNN version (if applicable): 10.2
• GPU models and configuration (e.g. V100):
• Any other relevant information:

Additional context

[VoVAllen]

Thanks for reporting this. We're planning to fix this at https://github.com/dmlc/dgl/pull/3249

[github-actions[bot]]

This issue has been automatically marked as stale due to lack of activity. It will be closed if no further activity occurs. Thank you

[Rhett-Ying]

As canonical etypes are not used in partition_graph() yet. Let me re-open this ticket.

• [ ] use canonical etypes when partition graph including partition_graph() and dist partition pipeline.
• [ ] load partitions and parse canonical etypes directly in DistGraph instead of going through different partitions.
• [ ] throw exception if loading deprecated configuration json. update corresponding doc.