Skip flattening for mutations across imports

Author: vish-cs

pipeline/differ/src/main/java/org/datacommons/ingestion/differ/DifferPipeline.java

@@ -30,12 +30,12 @@ public static void main(String[] args) {
     PCollection<McfGraph> currentNodes;
     if (options.getUseOptimizedGraphFormat()) {
       LOGGER.info("Using tfrecord file format");
-      currentNodes = PipelineUtils.readMcfGraph(options.getCurrentData(), p);
-      previousNodes = PipelineUtils.readMcfGraph(options.getPreviousData(), p);
+      currentNodes = PipelineUtils.readMcfGraph("differ", options.getCurrentData(), p);
+      previousNodes = PipelineUtils.readMcfGraph("differ", options.getPreviousData(), p);
     } else {
       LOGGER.info("Using mcf file format");
-      previousNodes = PipelineUtils.readMcfFiles(options.getPreviousData(), p);
-      currentNodes = PipelineUtils.readMcfFiles(options.getCurrentData(), p);
+      previousNodes = PipelineUtils.readMcfFiles("differ", options.getPreviousData(), p);
+      currentNodes = PipelineUtils.readMcfFiles("differ", options.getCurrentData(), p);
     }
 
     // Process the input and perform diff operation.

pipeline/differ/src/test/java/org/datacommons/ingestion/differ/DifferTest.java

@@ -36,9 +36,9 @@ public void testDiffer() {
 
     // Process the input.
     PCollection<McfGraph> currentGraph =
-        PipelineUtils.readMcfFiles(Paths.get(currentFile, "*.mcf").toString(), p);
+        PipelineUtils.readMcfFiles("test", Paths.get(currentFile, "*.mcf").toString(), p);
     PCollection<McfGraph> previousGraph =
-        PipelineUtils.readMcfFiles(Paths.get(previousFile, "*.mcf").toString(), p);
+        PipelineUtils.readMcfFiles("test", Paths.get(previousFile, "*.mcf").toString(), p);
     PCollectionTuple currentNodesTuple = DifferUtils.processGraph(currentGraph);
     PCollectionTuple previousNodesTuple = DifferUtils.processGraph(previousGraph);
 

pipeline/ingestion/src/main/java/org/datacommons/ingestion/pipeline/GraphIngestionPipeline.java

@@ -4,18 +4,18 @@
 import com.google.gson.JsonArray;
 import com.google.gson.JsonElement;
 import com.google.gson.JsonParser;
-import java.util.ArrayList;
 import java.util.List;
 import org.apache.beam.sdk.Pipeline;
 import org.apache.beam.sdk.io.gcp.spanner.SpannerWriteResult;
 import org.apache.beam.sdk.metrics.Counter;
 import org.apache.beam.sdk.metrics.Metrics;
 import org.apache.beam.sdk.options.PipelineOptionsFactory;
-import org.apache.beam.sdk.transforms.Flatten;
+import org.apache.beam.sdk.transforms.Create;
 import org.apache.beam.sdk.transforms.Values;
+import org.apache.beam.sdk.transforms.Wait;
 import org.apache.beam.sdk.values.PCollection;
-import org.apache.beam.sdk.values.PCollectionList;
 import org.apache.beam.sdk.values.PCollectionTuple;
+import org.apache.beam.sdk.values.TypeDescriptor;
 import org.datacommons.ingestion.spanner.SpannerClient;
 import org.datacommons.ingestion.util.GraphReader;
 import org.datacommons.ingestion.util.PipelineUtils;
@@ -70,12 +70,6 @@ public static void buildPipeline(
       Pipeline pipeline, IngestionPipelineOptions options, SpannerClient spannerClient) {
     LOGGER.info("Running import pipeline for imports: {}", options.getImportList());
 
-    // Initialize lists to hold mutations from all imports.
-    List<PCollection<Void>> deleteOpsList = new ArrayList<>();
-    List<PCollection<Mutation>> obsMutationList = new ArrayList<>();
-    List<PCollection<Mutation>> edgeMutationList = new ArrayList<>();
-    List<PCollection<Mutation>> nodeMutationList = new ArrayList<>();
-
     // Parse the input import list JSON.
     JsonElement jsonElement = JsonParser.parseString(options.getImportList());
     JsonArray jsonArray = jsonElement.getAsJsonArray();
@@ -97,37 +91,8 @@ public static void buildPipeline(
       String graphPath = pathElement.getAsString();
 
       // Process the individual import.
-      processImport(
-          pipeline,
-          spannerClient,
-          importName,
-          graphPath,
-          options.getSkipDelete(),
-          deleteOpsList,
-          nodeMutationList,
-          edgeMutationList,
-          obsMutationList);
-    }
-    // Finally, aggregate all collected mutations and write them to Spanner.
-    // 1. Process Deletes:
-    // First, execute all delete mutations to clear old data for the imports.
-    PCollection<Void> deleted =
-        PCollectionList.of(deleteOpsList).apply("DeleteOps", Flatten.pCollections());
-
-    // 2. Process Observations:
-    // Write observation mutations after deletes are complete.
-    if (options.getWriteObsGraph()) {
-      spannerClient.writeMutations(pipeline, "Observations", obsMutationList, deleted);
+      processImport(pipeline, spannerClient, importName, graphPath, options.getSkipDelete());
     }
-
-    // 3. Process Nodes:
-    // Write node mutations after deletes are complete.
-    SpannerWriteResult writtenNodes =
-        spannerClient.writeMutations(pipeline, "Nodes", nodeMutationList, deleted);
-
-    // 4. Process Edges:
-    // Write edge mutations only after node mutations are complete to ensure referential integrity.
-    spannerClient.writeMutations(pipeline, "Edges", edgeMutationList, writtenNodes.getOutput());
   }
 
   /**
@@ -138,67 +103,94 @@ public static void buildPipeline(
    * @param importName The name of the import.
    * @param graphPath The full path to the graph data.
    * @param skipDelete Whether to skip delete operations.
-   * @param deleteOpsList List to collect delete Ops.
-   * @param nodeMutationList List to collect node mutations.
-   * @param edgeMutationList List to collect edge mutations.
-   * @param obsMutationList List to collect observation mutations.
    */
   private static void processImport(
       Pipeline pipeline,
       SpannerClient spannerClient,
       String importName,
       String graphPath,
-      boolean skipDelete,
-      List<PCollection<Void>> deleteOpsList,
-      List<PCollection<Mutation>> nodeMutationList,
-      List<PCollection<Mutation>> edgeMutationList,
-      List<PCollection<Mutation>> obsMutationList) {
+      boolean skipDelete) {
     LOGGER.info("Import: {} Graph path: {}", importName, graphPath);
 
     String provenance = "dc/base/" + importName;
 
     // 1. Prepare Deletes:
     // Generate mutations to delete existing data for this import/provenance.
+    // Create a dummy signal if deletes are skipped, so downstream dependencies are satisfied
+    // immediately.
+    PCollection<Void> deleteObsWait;
+    PCollection<Void> deleteEdgesWait;
     if (!skipDelete) {
-      List<PCollection<Void>> deleteOps =
-          GraphReader.deleteExistingDataForImport(importName, provenance, pipeline, spannerClient);
-      deleteOpsList.addAll(deleteOps);
+      deleteObsWait =
+          spannerClient.deleteDataForImport(
+              pipeline, importName, spannerClient.getObservationTableName(), "import_name");
+      deleteEdgesWait =
+          spannerClient.deleteDataForImport(
+              pipeline, provenance, spannerClient.getEdgeTableName(), "provenance");
+    } else {
+      deleteObsWait =
+          pipeline.apply(
+              "CreateEmptyObsWait-" + importName, Create.empty(TypeDescriptor.of(Void.class)));
+      deleteEdgesWait =
+          pipeline.apply(
+              "CreateEmptyEdgesWait-" + importName, Create.empty(TypeDescriptor.of(Void.class)));
     }
 
     // 2. Read and Split Graph:
     // Read the graph data (TFRecord or MCF files) and split into schema and observation nodes.
     PCollection<McfGraph> graph =
         graphPath.contains("tfrecord")
-            ? PipelineUtils.readMcfGraph(graphPath, pipeline)
-            : PipelineUtils.readMcfFiles(graphPath, pipeline);
-    PCollectionTuple graphNodes = PipelineUtils.splitGraph(graph);
+            ? PipelineUtils.readMcfGraph(importName, graphPath, pipeline)
+            : PipelineUtils.readMcfFiles(importName, graphPath, pipeline);
+    PCollectionTuple graphNodes = PipelineUtils.splitGraph(importName, graph);
     PCollection<McfGraph> observationNodes = graphNodes.get(PipelineUtils.OBSERVATION_NODES_TAG);
     PCollection<McfGraph> schemaNodes = graphNodes.get(PipelineUtils.SCHEMA_NODES_TAG);
 
     // 3. Process Schema Nodes:
-    // Combine schema nodes if required, then convert to Node and Edge mutations.
+    // Combine nodes if required.
     PCollection<McfGraph> combinedGraph = schemaNodes;
     if (IMPORTS_TO_COMBINE.contains(importName)) {
-      combinedGraph = PipelineUtils.combineGraphNodes(schemaNodes);
+      combinedGraph = PipelineUtils.combineGraphNodes(importName, schemaNodes);
     }
+
+    // Convert all nodes to mutations
     PCollection<Mutation> nodeMutations =
         GraphReader.graphToNodes(
-                importName, combinedGraph, spannerClient, nodeCounter, nodeInvalidTypeCounter)
+                "NodeMutations-" + importName,
+                combinedGraph,
+                spannerClient,
+                nodeCounter,
+                nodeInvalidTypeCounter)
             .apply("ExtractNodeMutations-" + importName, Values.create());
     PCollection<Mutation> edgeMutations =
-        GraphReader.graphToEdges(importName, combinedGraph, provenance, spannerClient, edgeCounter)
+        GraphReader.graphToEdges(
+                "EdgeMutations-" + importName,
+                combinedGraph,
+                provenance,
+                spannerClient,
+                edgeCounter)
             .apply("ExtractEdgeMutations-" + importName, Values.create());
 
-    nodeMutationList.add(nodeMutations);
-    edgeMutationList.add(edgeMutations);
+    // Write Nodes
+    SpannerWriteResult writtenNodes =
+        spannerClient.writeMutations(pipeline, "WriteNodesToSpanner-" + importName, nodeMutations);
+
+    // Write Edges (wait for Nodes write and Edges delete)
+    edgeMutations.apply(
+        "EdgesWaitOn-" + importName, Wait.on(List.of(writtenNodes.getOutput(), deleteEdgesWait)));
+    spannerClient.writeMutations(pipeline, "WriteEdgesToSpanner-" + importName, edgeMutations);
 
     // 4. Process Observation Nodes:
     // Build an optimized graph from observation nodes and convert to Observation mutations.
     PCollection<McfOptimizedGraph> optimizedGraph =
-        PipelineUtils.buildOptimizedMcfGraph(observationNodes);
+        PipelineUtils.buildOptimizedMcfGraph(importName, observationNodes);
     PCollection<Mutation> observationMutations =
         GraphReader.graphToObservations(optimizedGraph, importName, spannerClient, obsCounter)
-            .apply("ExtractObsMutations", Values.create());
-    obsMutationList.add(observationMutations);
+            .apply("ExtractObsMutations-" + importName, Values.create());
+    // Write Observations (wait for Obs delete)
+    observationMutations.apply("ObsWaitOn-" + importName, Wait.on(deleteObsWait));
+
+    spannerClient.writeMutations(
+        pipeline, "WriteObservationsToSpanner-" + importName, observationMutations);
   }
 }

pipeline/ingestion/src/test/java/org/datacommons/ingestion/pipeline/GraphIngestionPipelineIntegrationTest.java

@@ -23,7 +23,6 @@
 import java.io.IOException;
 import java.nio.charset.StandardCharsets;
 import java.nio.file.Files;
-import java.util.UUID;
 import org.apache.beam.runners.dataflow.DataflowRunner;
 import org.apache.beam.runners.dataflow.options.DataflowPipelineOptions;
 import org.apache.beam.sdk.Pipeline;
@@ -86,7 +85,7 @@ public class GraphIngestionPipelineIntegrationTest {
   private String region;
   private String emulatorHost;
   private boolean isLocal;
-  private String importName = "TestImport-" + UUID.randomUUID().toString();
+  private String importName = "TestImport";
   private String nodeNameValue = "Test Node Name";
   private SpannerClient spannerClient;
 

pipeline/ingestion/src/test/java/org/datacommons/ingestion/pipeline/GraphIngestionPipelineTest.java

@@ -19,10 +19,8 @@
 import org.apache.beam.sdk.testing.TestPipeline;
 import org.apache.beam.sdk.transforms.Create;
 import org.apache.beam.sdk.transforms.DoFn;
-import org.apache.beam.sdk.transforms.Flatten;
 import org.apache.beam.sdk.transforms.ParDo;
 import org.apache.beam.sdk.values.PCollection;
-import org.apache.beam.sdk.values.PCollectionList;
 import org.apache.beam.sdk.values.TypeDescriptor;
 import org.datacommons.ingestion.spanner.SpannerClient;
 import org.junit.Before;
@@ -121,33 +119,24 @@ public MockSpannerClient(PCollection<Void> deleteSignal, SpannerWriteResult mock
     }
 
     @Override
-    public PCollection<Void> deleteObservationsForImport(String importName, Pipeline pipeline) {
-      return deleteSignal;
-    }
-
-    @Override
-    public PCollection<Void> deleteEdgesForImport(String provenance, Pipeline pipeline) {
+    public PCollection<Void> deleteDataForImport(
+        Pipeline pipeline, String importName, String tableName, String columnName) {
       return deleteSignal;
     }
 
     @Override
     public SpannerWriteResult writeMutations(
-        Pipeline pipeline,
-        String name,
-        List<PCollection<Mutation>> mutationList,
-        PCollection<?> waitSignal) {
-
-      PCollectionList.of(mutationList)
-          .apply("Flatten" + name + "Mutations-Test", Flatten.pCollections())
-          .apply(
-              "Capture" + name + "Mutations",
-              ParDo.of(
-                  new DoFn<Mutation, Void>() {
-                    @ProcessElement
-                    public void processElement(@Element Mutation m) {
-                      capturedMutations.add(m);
-                    }
-                  }));
+        Pipeline pipeline, String name, PCollection<Mutation> mutations) {
+
+      mutations.apply(
+          "Capture" + name + "Mutations",
+          ParDo.of(
+              new DoFn<Mutation, Void>() {
+                @ProcessElement
+                public void processElement(@Element Mutation m) {
+                  capturedMutations.add(m);
+                }
+              }));
       return mockWriteResult;
     }
   }