Generate .labelobl files to speed-up labelled graph loading

Also see https://github.com/vigna/webgraph-big/pull/7

docs/compression.rst

@@ -131,7 +131,7 @@ The compression pipeline consists of the following steps:
 .. figure:: images/compression_steps.png
     :align: center
     :alt: Compression steps
-    :target: _images/compression_steps.png
+    :scale: 20%
 
     Compression steps
 
@@ -587,8 +587,23 @@ equivalent of these files:
 ``graph-transposed-labelled.{properties,labels,labeloffsets}``.
 
 
+23. EDGE_LABELS_OBL
+-------------------
+
+Cache the label offsets of the forward labelled graph to make loading faster.
+The resulting label offset big list is stored in the
+``graph-labelled.labelobl`` file.
+
+
+23. EDGE_LABELS_TRANSPOSE_OBL
+-----------------------------
+
+Same as EDGE_LABELS_OBL, but for the transposed labelled graph.
+The resulting label offset big list is stored in the
+``graph-transposed-labelled.labelobl`` file.
+
 
-22. CLEAN_TMP
+24. CLEAN_TMP
 -------------
 
 This step reclaims space by deleting the temporary directory, as well as all

docs/images/compression_steps.dot

@@ -31,6 +31,8 @@ digraph "Compression steps" {
     labels_fcl [label="graph.labels.fcl"];
     graph_labelled [label="graph-labelled.*"];
     graph_transposed_labelled [label="graph-transposed-labelled.*"];
+    graph_labelled_obl [label="graph-labelled.labelobl"];
+    graph_transposed_labelled [label="graph-transposed-labelled.labelobl"];
 
     subgraph {
         node [shape=box, fontname="Courier New"];
@@ -55,6 +57,8 @@ digraph "Compression steps" {
         MPH_LABELS;
         FCL_LABELS;
         EDGE_LABELS;
+        EDGE_LABELS_OBL;
+        EDGE_LABELS_TRANSPOSE_OBL;
     }
 
 
@@ -102,4 +106,6 @@ digraph "Compression steps" {
     graph_order -> EDGE_LABELS;
     EDGE_LABELS -> graph_labelled;
     EDGE_LABELS -> graph_transposed_labelled;
+    graph_labelled -> EDGE_LABELS_OBL -> graph_labelled_obl;
+    graph_transposed_labelled -> EDGE_LABELS_TRANSPOSE_OBL -> graph_transposed_labelled_obl;
 }

swh/graph/cli.py

@@ -176,7 +176,7 @@ def compress(ctx, input_dataset, output_directory, graph_name, steps):
     (10) obl, (11) compose_orders, (12) stats, (13) transpose, (14)
     transpose_obl, (15) maps, (16) extract_persons, (17) mph_persons, (18)
     node_properties, (19) mph_labels, (20) fcl_labels, (21) edge_labels, (22)
-    clean_tmp.
+    edge_labels_obl, (23) edge_labels_transpose_obl, (24) clean_tmp.
     Compression steps can be selected by name or number using --steps,
     separating them with commas; step ranges (e.g., 3-9, 6-, etc.) are also
     supported.

swh/graph/webgraph.py

@@ -42,7 +42,9 @@ class CompressionStep(Enum):
     MPH_LABELS = 19
     FCL_LABELS = 20
     EDGE_LABELS = 21
-    CLEAN_TMP = 22
+    EDGE_LABELS_OBL = 22
+    EDGE_LABELS_TRANSPOSE_OBL = 23
+    CLEAN_TMP = 24
 
     def __str__(self):
         return self.name
@@ -231,6 +233,18 @@ STEP_ARGV: Dict[CompressionStep, List[str]] = {
         "{in_dir}",
         "{out_dir}/{graph_name}",
     ],
+    CompressionStep.EDGE_LABELS_OBL: [
+        "{java}",
+        "it.unimi.dsi.big.webgraph.labelling.BitStreamArcLabelledImmutableGraph",
+        "--list",
+        "{out_dir}/{graph_name}-labelled",
+    ],
+    CompressionStep.EDGE_LABELS_TRANSPOSE_OBL: [
+        "{java}",
+        "it.unimi.dsi.big.webgraph.labelling.BitStreamArcLabelledImmutableGraph",
+        "--list",
+        "{out_dir}/{graph_name}-transposed-labelled",
+    ],
     CompressionStep.CLEAN_TMP: [
         "rm",
         "-rf",