[Distributed] Fix a bug in graph partition. (#2869)

* update distributed training doc.

* explain data split.

* fix message passing.

* id mapping.

* fix.

* test data reshuffling.

* fix a bug.

* fix test.

* Revert "fix."

This reverts commit 2d025e9e1a5c05c3da9b803a035a788ced59bd77.

* Revert "id mapping."

This reverts commit 2a6a93ceb81fbdff86e6e9e5a58e1ace1e9d9882.

* Revert "fix message passing."

This reverts commit ed8a86bf2b015e5e4f64ba160e81b207ad2a1d65.

* Revert "explain data split."

This reverts commit 4338ddf8a336014cf92d4cb9a1db02b9badc0e55.

* Revert "update distributed training doc."

This reverts commit dda1c35c44536934c19715534f01f832afda6ad2.

* add more tests.

* fix.

* fix.

* fix.
Co-authored-by: Zheng <dzzhen@3c22fba32af5.ant.amazon.com>
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>

python/dgl/partition.py

@@ -7,6 +7,7 @@ from .heterograph import DGLHeteroGraph
 from . import backend as F
 from . import utils
 from .base import EID, NID, NTYPE, ETYPE
+from .subgraph import edge_subgraph
 
 __all__ = ["metis_partition", "metis_partition_assignment",
            "partition_graph_with_halo"]
@@ -173,6 +174,14 @@ def partition_graph_with_halo(g, node_part, extra_cached_hops, reshuffle=False):
     # This creaets a subgraph from subgraphs returned from the CAPI above.
     def create_subgraph(subg, induced_nodes, induced_edges):
         subg1 = DGLHeteroGraph(gidx=subg.graph, ntypes=['_N'], etypes=['_E'])
+        # If IDs are shuffled, we should shuffled edges. This will help us collect edge data
+        # from the distributed graph after training.
+        if reshuffle:
+            sorted_edges, index = F.sort_1d(induced_edges[0])
+            subg1 = edge_subgraph(subg1, index, preserve_nodes=True)
+            subg1.ndata[NID] = induced_nodes[0]
+            subg1.edata[EID] = sorted_edges
+        else:
             subg1.ndata[NID] = induced_nodes[0]
             subg1.edata[EID] = induced_edges[0]
         return subg1

tests/distributed/test_partition.py

@@ -39,7 +39,7 @@ def create_random_graph(n):
     return dgl.from_scipy(arr)
 
 def create_random_hetero():
-    num_nodes = {'n1': 10000, 'n2': 10010, 'n3': 10020}
+    num_nodes = {'n1': 1000, 'n2': 1010, 'n3': 1020}
     etypes = [('n1', 'r1', 'n2'),
               ('n1', 'r2', 'n3'),
               ('n2', 'r3', 'n3')]
@@ -120,24 +120,50 @@ def verify_hetero_graph(g, parts):
         assert len(uniq_ids) == g.number_of_edges(etype)
     # TODO(zhengda) this doesn't check 'part_id'
 
-def verify_graph_feats(g, part, node_feats):
+def verify_graph_feats(g, gpb, part, node_feats, edge_feats):
     for ntype in g.ntypes:
         ntype_id = g.get_ntype_id(ntype)
-        for name in g.nodes[ntype].data:
-            if name in [dgl.NID, 'inner_node']:
-                continue
         inner_node_mask = _get_inner_node_mask(part, ntype_id)
         inner_nids = F.boolean_mask(part.ndata[dgl.NID],inner_node_mask)
-            min_nids = F.min(inner_nids, 0)
+        ntype_ids, inner_type_nids = gpb.map_to_per_ntype(inner_nids)
+        partid = gpb.nid2partid(inner_type_nids, ntype)
+        assert np.all(F.asnumpy(ntype_ids) == ntype_id)
+        assert np.all(F.asnumpy(partid) == gpb.partid)
+
         orig_id = F.boolean_mask(part.ndata['orig_id'], inner_node_mask)
+        local_nids = gpb.nid2localnid(inner_type_nids, gpb.partid, ntype)
+
+        for name in g.nodes[ntype].data:
+            if name in [dgl.NID, 'inner_node']:
+                continue
             true_feats = F.gather_row(g.nodes[ntype].data[name], orig_id)
-            ndata = F.gather_row(node_feats[ntype + '/' + name], inner_nids - min_nids)
+            ndata = F.gather_row(node_feats[ntype + '/' + name], local_nids)
             assert np.all(F.asnumpy(ndata == true_feats))
 
+    for etype in g.etypes:
+        etype_id = g.get_etype_id(etype)
+        inner_edge_mask = _get_inner_edge_mask(part, etype_id)
+        inner_eids = F.boolean_mask(part.edata[dgl.EID],inner_edge_mask)
+        etype_ids, inner_type_eids = gpb.map_to_per_etype(inner_eids)
+        partid = gpb.eid2partid(inner_type_eids, etype)
+        assert np.all(F.asnumpy(etype_ids) == etype_id)
+        assert np.all(F.asnumpy(partid) == gpb.partid)
+
+        orig_id = F.boolean_mask(part.edata['orig_id'], inner_edge_mask)
+        local_eids = gpb.eid2localeid(inner_type_eids, gpb.partid, etype)
+
+        for name in g.edges[etype].data:
+            if name in [dgl.EID, 'inner_edge']:
+                continue
+            true_feats = F.gather_row(g.edges[etype].data[name], orig_id)
+            edata = F.gather_row(edge_feats[etype + '/' + name], local_eids)
+            assert np.all(F.asnumpy(edata == true_feats))
+
 def check_hetero_partition(hg, part_method):
     hg.nodes['n1'].data['labels'] = F.arange(0, hg.number_of_nodes('n1'))
     hg.nodes['n1'].data['feats'] = F.tensor(np.random.randn(hg.number_of_nodes('n1'), 10), F.float32)
     hg.edges['r1'].data['feats'] = F.tensor(np.random.randn(hg.number_of_edges('r1'), 10), F.float32)
+    hg.edges['r1'].data['labels'] = F.arange(0, hg.number_of_edges('r1'))
     num_parts = 4
     num_hops = 1
 
@@ -150,6 +176,8 @@ def check_hetero_partition(hg, part_method):
     for etype in hg.etypes:
         assert len(orig_eids[etype]) == hg.number_of_edges(etype)
     parts = []
+    shuffled_labels = []
+    shuffled_elabels = []
     for i in range(num_parts):
         part_g, node_feats, edge_feats, gpb, _, ntypes, etypes = load_partition('/tmp/partition/test.json', i)
         # Verify the mapping between the reshuffled IDs and the original IDs.
@@ -181,9 +209,21 @@ def check_hetero_partition(hg, part_method):
             assert len(orig_eids1) == len(orig_eids2)
             assert np.all(F.asnumpy(orig_eids1) == F.asnumpy(orig_eids2))
         parts.append(part_g)
-        verify_graph_feats(hg, part_g, node_feats)
+        verify_graph_feats(hg, gpb, part_g, node_feats, edge_feats)
+
+        shuffled_labels.append(node_feats['n1/labels'])
+        shuffled_elabels.append(edge_feats['r1/labels'])
     verify_hetero_graph(hg, parts)
 
+    shuffled_labels = F.asnumpy(F.cat(shuffled_labels, 0))
+    shuffled_elabels = F.asnumpy(F.cat(shuffled_elabels, 0))
+    orig_labels = np.zeros(shuffled_labels.shape, dtype=shuffled_labels.dtype)
+    orig_elabels = np.zeros(shuffled_elabels.shape, dtype=shuffled_elabels.dtype)
+    orig_labels[F.asnumpy(orig_nids['n1'])] = shuffled_labels
+    orig_elabels[F.asnumpy(orig_eids['r1'])] = shuffled_elabels
+    assert np.all(orig_labels == F.asnumpy(hg.nodes['n1'].data['labels']))
+    assert np.all(orig_elabels == F.asnumpy(hg.edges['r1'].data['labels']))
+
 def check_partition(g, part_method, reshuffle):
     g.ndata['labels'] = F.arange(0, g.number_of_nodes())
     g.ndata['feats'] = F.tensor(np.random.randn(g.number_of_nodes(), 10), F.float32)
@@ -196,6 +236,8 @@ def check_partition(g, part_method, reshuffle):
     orig_nids, orig_eids = partition_graph(g, 'test', num_parts, '/tmp/partition', num_hops=num_hops,
                                            part_method=part_method, reshuffle=reshuffle, return_mapping=True)
     part_sizes = []
+    shuffled_labels = []
+    shuffled_edata = []
     for i in range(num_parts):
         part_g, node_feats, edge_feats, gpb, _, ntypes, etypes = load_partition('/tmp/partition/test.json', i)
 
@@ -251,6 +293,7 @@ def check_partition(g, part_method, reshuffle):
         else:
             part_g.ndata['feats'] = F.gather_row(g.ndata['feats'], part_g.ndata[dgl.NID])
             part_g.edata['feats'] = F.gather_row(g.edata['feats'], part_g.edata[dgl.NID])
+
         part_g.update_all(fn.copy_src('feats', 'msg'), fn.sum('msg', 'h'))
         part_g.update_all(fn.copy_edge('feats', 'msg'), fn.sum('msg', 'eh'))
         assert F.allclose(F.gather_row(g.ndata['h'], local_nodes),
@@ -261,11 +304,31 @@ def check_partition(g, part_method, reshuffle):
         for name in ['labels', 'feats']:
             assert '_N/' + name in node_feats
             assert node_feats['_N/' + name].shape[0] == len(local_nodes)
-            assert np.all(F.asnumpy(g.ndata[name])[F.asnumpy(local_nodes)] == F.asnumpy(node_feats['_N/' + name]))
+            true_feats = F.gather_row(g.ndata[name], local_nodes)
+            ndata = F.gather_row(node_feats['_N/' + name], local_nid)
+            assert np.all(F.asnumpy(true_feats) == F.asnumpy(ndata))
         for name in ['feats']:
             assert '_E/' + name in edge_feats
             assert edge_feats['_E/' + name].shape[0] == len(local_edges)
-            assert np.all(F.asnumpy(g.edata[name])[F.asnumpy(local_edges)] == F.asnumpy(edge_feats['_E/' + name]))
+            true_feats = F.gather_row(g.edata[name], local_edges)
+            edata = F.gather_row(edge_feats['_E/' + name], local_eid)
+            assert np.all(F.asnumpy(true_feats) == F.asnumpy(edata))
+
+        # This only works if node/edge IDs are shuffled.
+        if reshuffle:
+            shuffled_labels.append(node_feats['_N/labels'])
+            shuffled_edata.append(edge_feats['_E/feats'])
+
+    # Verify that we can reconstruct node/edge data for original IDs.
+    if reshuffle:
+        shuffled_labels = F.asnumpy(F.cat(shuffled_labels, 0))
+        shuffled_edata = F.asnumpy(F.cat(shuffled_edata, 0))
+        orig_labels = np.zeros(shuffled_labels.shape, dtype=shuffled_labels.dtype)
+        orig_edata = np.zeros(shuffled_edata.shape, dtype=shuffled_edata.dtype)
+        orig_labels[F.asnumpy(orig_nids)] = shuffled_labels
+        orig_edata[F.asnumpy(orig_eids)] = shuffled_edata
+        assert np.all(orig_labels == F.asnumpy(g.ndata['labels']))
+        assert np.all(orig_edata == F.asnumpy(g.edata['feats']))
 
     if reshuffle:
         node_map = []
@@ -284,7 +347,7 @@ def check_partition(g, part_method, reshuffle):
 
 @unittest.skipIf(os.name == 'nt', reason='Do not support windows yet')
 def test_partition():
-    g = create_random_graph(10000)
+    g = create_random_graph(1000)
     check_partition(g, 'metis', False)
     check_partition(g, 'metis', True)
     check_partition(g, 'random', False)