fix(interactive): Unified Query Cache for Gremlin and Cypher Queries (#…

…3890)

1. Use a unified query cache for both Gremlin and Cypher queries.
2. Use the pre-optimization `LogicalPlan` as the query cache key to
avoid applying optimizations when computing the cache key.
3. Directly use the string from the plan explanation as the hash code,
instead of computing it from all attributes within each operator of the
plan. This is mainly due to the following reasons:
1. Avoid modifying the hash code each time when a new attribute is added
to an operator.
2. For some static attributes, Java does not allow overloading, and it
does not guarantee these attributes will remain unchanged after a
service restart, such as
[enum](https://stackoverflow.com/questions/19496751/enum-hash-code-calculation-are-not-the-same).

interactive_engine/compiler/src/main/java/com/alibaba/graphscope/GraphServer.java

@@ -82,15 +82,16 @@ public GraphServer(
     public void start() throws Exception {
         ExecutionClient executionClient = ExecutionClient.Factory.create(configs, channelFetcher);
         QueryIdGenerator idGenerator = new QueryIdGenerator(configs);
+        QueryCache queryCache = new QueryCache(configs);
         if (!FrontendConfig.GREMLIN_SERVER_DISABLED.get(configs)) {
             GraphPlanner graphPlanner =
                     new GraphPlanner(configs, new LogicalPlanFactory.Gremlin(), optimizer);
-            QueryCache queryCache = new QueryCache(configs, graphPlanner);
             this.gremlinServer =
                     new IrGremlinServer(
                             configs,
                             idGenerator,
                             queryCache,
+                            graphPlanner,
                             executionClient,
                             channelFetcher,
                             metaQueryCallback,
@@ -100,10 +101,14 @@ public void start() throws Exception {
         if (!FrontendConfig.NEO4J_BOLT_SERVER_DISABLED.get(configs)) {
             GraphPlanner graphPlanner =
                     new GraphPlanner(configs, new LogicalPlanFactory.Cypher(), optimizer);
-            QueryCache queryCache = new QueryCache(configs, graphPlanner);
             this.cypherBootstrapper =
                     new CypherBootstrapper(
-                            configs, idGenerator, metaQueryCallback, executionClient, queryCache);
+                            configs,
+                            idGenerator,
+                            metaQueryCallback,
+                            executionClient,
+                            queryCache,
+                            graphPlanner);
             Path neo4jHomePath = getNeo4jHomePath();
             this.cypherBootstrapper.start(
                     neo4jHomePath,

interactive_engine/compiler/src/main/java/com/alibaba/graphscope/common/ir/tools/GraphPlanner.java

@@ -93,20 +93,34 @@ public PlannerInstance instance(String query, IrMeta irMeta) {
         if (mq != null) {
             optCluster.setMetadataQuerySupplier(() -> mq);
         }
-        return new PlannerInstance(query, optCluster, irMeta);
+        // build logical plan from parsed query
+        IrGraphSchema schema = irMeta.getSchema();
+        GraphBuilder graphBuilder =
+                GraphBuilder.create(
+                        graphConfig, optCluster, new GraphOptSchema(optCluster, schema));
+
+        LogicalPlan logicalPlan = logicalPlanFactory.create(graphBuilder, irMeta, query);
+        return new PlannerInstance(query, logicalPlan, graphBuilder, irMeta);
     }
 
     public class PlannerInstance {
         private final String query;
-        private final GraphOptCluster optCluster;
+        private final LogicalPlan parsedPlan;
+        private final GraphBuilder graphBuilder;
         private final IrMeta irMeta;
 
-        public PlannerInstance(String query, GraphOptCluster optCluster, IrMeta irMeta) {
+        public PlannerInstance(
+                String query, LogicalPlan parsedPlan, GraphBuilder graphBuilder, IrMeta irMeta) {
             this.query = query;
-            this.optCluster = optCluster;
+            this.parsedPlan = parsedPlan;
+            this.graphBuilder = graphBuilder;
             this.irMeta = irMeta;
         }
 
+        public LogicalPlan getParsedPlan() {
+            return parsedPlan;
+        }
+
         public Summary plan() {
             LogicalPlan logicalPlan =
                     ClassUtils.callException(() -> planLogical(), Code.LOGICAL_PLAN_BUILD_FAILED);
@@ -117,16 +131,7 @@ public Summary plan() {
         }
 
         public LogicalPlan planLogical() {
-            // build logical plan from parsed query
-            IrGraphSchema schema = irMeta.getSchema();
-            GraphBuilder graphBuilder =
-                    GraphBuilder.create(
-                            graphConfig,
-                            this.optCluster,
-                            new GraphOptSchema(this.optCluster, schema));
-
-            LogicalPlan logicalPlan = logicalPlanFactory.create(graphBuilder, irMeta, query);
-
+            LogicalPlan logicalPlan = parsedPlan;
             // apply optimizations
             if (logicalPlan.getRegularQuery() != null && !logicalPlan.isReturnEmpty()) {
                 RelNode before = logicalPlan.getRegularQuery();

interactive_engine/compiler/src/main/java/com/alibaba/graphscope/common/ir/tools/LogicalPlan.java

@@ -22,7 +22,6 @@
 import com.google.common.collect.ImmutableList;
 import com.google.common.collect.Lists;
 
-import org.apache.calcite.plan.RelDigest;
 import org.apache.calcite.rel.RelNode;
 import org.apache.calcite.rel.logical.LogicalValues;
 import org.apache.calcite.rel.type.RelDataType;
@@ -136,7 +135,7 @@ public int hashCode() {
         return Objects.hash(getDigest(regularQuery), procedureCall, returnEmpty, dynamicParams);
     }
 
-    private RelDigest getDigest(RelNode rel) {
-        return rel == null ? null : rel.getRelDigest();
+    private String getDigest(RelNode rel) {
+        return rel == null ? null : rel.explain();
     }
 }

interactive_engine/compiler/src/main/java/com/alibaba/graphscope/common/ir/tools/QueryCache.java

@@ -18,8 +18,6 @@
 
 import com.alibaba.graphscope.common.config.Configs;
 import com.alibaba.graphscope.common.config.FrontendConfig;
-import com.alibaba.graphscope.common.ir.meta.IrMeta;
-import com.alibaba.graphscope.common.ir.runtime.PhysicalPlan;
 import com.google.common.cache.CacheBuilder;
 import com.google.common.cache.CacheLoader;
 import com.google.common.cache.LoadingCache;
@@ -32,34 +30,22 @@
 
 public class QueryCache {
     private final LoadingCache<Key, Value> cache;
-    private final GraphPlanner graphPlanner;
 
-    public QueryCache(Configs configs, GraphPlanner graphPlanner) {
-        this.graphPlanner = graphPlanner;
+    public QueryCache(Configs configs) {
         int cacheSize = FrontendConfig.QUERY_CACHE_SIZE.get(configs);
         this.cache =
                 CacheBuilder.newBuilder()
                         .maximumSize(cacheSize)
-                        .build(
-                                CacheLoader.from(
-                                        key -> {
-                                            PhysicalPlan physicalPlan =
-                                                    key.plannerInstance.planPhysical(
-                                                            key.logicalPlan);
-                                            GraphPlanner.Summary summary =
-                                                    new GraphPlanner.Summary(
-                                                            key.logicalPlan, physicalPlan);
-                                            return new Value(summary, null);
-                                        }));
+                        .build(CacheLoader.from(key -> new Value(key.instance.plan(), null)));
     }
 
     public class Key {
-        public final GraphPlanner.PlannerInstance plannerInstance;
+        public final GraphPlanner.PlannerInstance instance;
         public final LogicalPlan logicalPlan;
 
-        public Key(String query, IrMeta irMeta) {
-            this.plannerInstance = Objects.requireNonNull(graphPlanner.instance(query, irMeta));
-            this.logicalPlan = Objects.requireNonNull(this.plannerInstance.planLogical());
+        public Key(GraphPlanner.PlannerInstance instance) {
+            this.instance = instance;
+            this.logicalPlan = instance.getParsedPlan();
         }
 
         @Override
@@ -76,8 +62,8 @@ public int hashCode() {
         }
     }
 
-    public Key createKey(String query, IrMeta irMeta) {
-        return new Key(query, irMeta);
+    public Key createKey(GraphPlanner.PlannerInstance instance) {
+        return new Key(instance);
     }
 
     public static class Value {

interactive_engine/compiler/src/main/java/com/alibaba/graphscope/cypher/executor/GraphQueryExecutor.java

@@ -64,6 +64,7 @@ public class GraphQueryExecutor extends FabricExecutor {
     private final QueryIdGenerator idGenerator;
     private final FabricConfig fabricConfig;
     private final QueryCache queryCache;
+    private final GraphPlanner graphPlanner;
 
     public GraphQueryExecutor(
             FabricConfig config,
@@ -77,7 +78,8 @@ public GraphQueryExecutor(
             QueryIdGenerator idGenerator,
             IrMetaQueryCallback metaQueryCallback,
             ExecutionClient client,
-            QueryCache queryCache) {
+            QueryCache queryCache,
+            GraphPlanner graphPlanner) {
         super(
                 config,
                 planner,
@@ -92,6 +94,7 @@ public GraphQueryExecutor(
         this.metaQueryCallback = metaQueryCallback;
         this.client = client;
         this.queryCache = queryCache;
+        this.graphPlanner = graphPlanner;
     }
 
     /**
@@ -124,7 +127,8 @@ public StatementResult run(
                 return super.run(fabricTransaction, statement, parameters);
             }
             irMeta = metaQueryCallback.beforeExec();
-            QueryCache.Key cacheKey = queryCache.createKey(statement, irMeta);
+            QueryCache.Key cacheKey =
+                    queryCache.createKey(graphPlanner.instance(statement, irMeta));
             QueryCache.Value cacheValue = queryCache.get(cacheKey);
             Preconditions.checkArgument(
                     cacheValue != null,

interactive_engine/compiler/src/main/java/com/alibaba/graphscope/cypher/service/CypherBootstrapper.java

@@ -18,6 +18,7 @@
 
 import com.alibaba.graphscope.common.client.ExecutionClient;
 import com.alibaba.graphscope.common.config.Configs;
+import com.alibaba.graphscope.common.ir.tools.GraphPlanner;
 import com.alibaba.graphscope.common.ir.tools.QueryCache;
 import com.alibaba.graphscope.common.ir.tools.QueryIdGenerator;
 import com.alibaba.graphscope.common.manager.IrMetaQueryCallback;
@@ -55,18 +56,20 @@ public CypherBootstrapper(
             QueryIdGenerator idGenerator,
             IrMetaQueryCallback queryCallback,
             ExecutionClient client,
-            QueryCache queryCache) {
+            QueryCache queryCache,
+            GraphPlanner graphPlanner) {
         this.client = client;
         this.externalDependencies =
                 createExternalDependencies(
-                        graphConfig, idGenerator, queryCallback, client, queryCache);
+                        graphConfig, idGenerator, queryCallback, client, queryCache, graphPlanner);
         this.externalClassTypes =
                 Arrays.asList(
                         Configs.class,
                         QueryIdGenerator.class,
                         IrMetaQueryCallback.class,
                         ExecutionClient.class,
-                        QueryCache.class);
+                        QueryCache.class,
+                        GraphPlanner.class);
     }
 
     @Override
@@ -94,9 +97,11 @@ private Dependencies createExternalDependencies(
             QueryIdGenerator idGenerator,
             IrMetaQueryCallback queryCallback,
             ExecutionClient client,
-            QueryCache queryCache) {
+            QueryCache queryCache,
+            GraphPlanner graphPlanner) {
         Dependencies dependencies = new Dependencies();
-        dependencies.satisfyDependencies(configs, idGenerator, queryCallback, client, queryCache);
+        dependencies.satisfyDependencies(
+                configs, idGenerator, queryCallback, client, queryCache, graphPlanner);
         return dependencies;
     }
 

interactive_engine/compiler/src/main/java/com/alibaba/graphscope/cypher/service/CypherQueryServiceBootstrap.java

@@ -22,6 +22,7 @@
 
 import com.alibaba.graphscope.common.client.ExecutionClient;
 import com.alibaba.graphscope.common.config.Configs;
+import com.alibaba.graphscope.common.ir.tools.GraphPlanner;
 import com.alibaba.graphscope.common.ir.tools.QueryCache;
 import com.alibaba.graphscope.common.ir.tools.QueryIdGenerator;
 import com.alibaba.graphscope.common.manager.IrMetaQueryCallback;
@@ -164,6 +165,7 @@ public void bootstrapServices() {
         var metaQueryCallback = (IrMetaQueryCallback) resolve(IrMetaQueryCallback.class);
         var executionClient = (ExecutionClient) resolve(ExecutionClient.class);
         var queryCache = (QueryCache) resolve(QueryCache.class);
+        var graphPlanner = (GraphPlanner) resolve(GraphPlanner.class);
         var fabricExecutor =
                 new GraphQueryExecutor(
                         fabricConfig,
@@ -177,7 +179,8 @@ public void bootstrapServices() {
                         idGenerator,
                         metaQueryCallback,
                         executionClient,
-                        queryCache);
+                        queryCache,
+                        graphPlanner);
         register(fabricExecutor, FabricExecutor.class);
 
         register(

interactive_engine/compiler/src/main/java/com/alibaba/graphscope/gremlin/integration/processor/IrTestOpProcessor.java

@@ -74,6 +74,7 @@ public IrTestOpProcessor(
             Configs configs,
             QueryIdGenerator idGenerator,
             QueryCache queryCache,
+            GraphPlanner graphPlanner,
             ExecutionClient executionClient,
             ChannelFetcher fetcher,
             IrMetaQueryCallback metaQueryCallback,
@@ -84,6 +85,7 @@ public IrTestOpProcessor(
                 configs,
                 idGenerator,
                 queryCache,
+                graphPlanner,
                 executionClient,
                 fetcher,
                 metaQueryCallback,
@@ -157,7 +159,9 @@ public ThrowingConsumer<Context> select(Context ctx) {
                                     break;
                                 case GremlinCalciteScriptEngineFactory.LANGUAGE_NAME:
                                     QueryCache.Value value =
-                                            queryCache.get(queryCache.createKey(script, irMeta));
+                                            queryCache.get(
+                                                    queryCache.createKey(
+                                                            graphPlanner.instance(script, irMeta)));
                                     GraphPlanner.Summary summary = value.summary;
                                     ResultSchema resultSchema =
                                             new ResultSchema(summary.getLogicalPlan());

interactive_engine/compiler/src/main/java/com/alibaba/graphscope/gremlin/plugin/processor/IrStandardOpProcessor.java

@@ -35,6 +35,7 @@
 import com.alibaba.graphscope.common.exception.FrontendException;
 import com.alibaba.graphscope.common.intermediate.InterOpCollection;
 import com.alibaba.graphscope.common.ir.meta.IrMeta;
+import com.alibaba.graphscope.common.ir.tools.GraphPlanner;
 import com.alibaba.graphscope.common.ir.tools.QueryCache;
 import com.alibaba.graphscope.common.ir.tools.QueryIdGenerator;
 import com.alibaba.graphscope.common.manager.IrMetaQueryCallback;
@@ -109,6 +110,7 @@ public class IrStandardOpProcessor extends StandardOpProcessor {
     protected final IrMetaQueryCallback metaQueryCallback;
     protected final QueryIdGenerator idGenerator;
     protected final QueryCache queryCache;
+    protected final GraphPlanner graphPlanner;
     protected final ExecutionClient executionClient;
     protected Tracer tracer;
     protected LongHistogram queryHistogram;
@@ -124,6 +126,7 @@ public IrStandardOpProcessor(
             Configs configs,
             QueryIdGenerator idGenerator,
             QueryCache queryCache,
+            GraphPlanner graphPlanner,
             ExecutionClient executionClient,
             ChannelFetcher fetcher,
             IrMetaQueryCallback metaQueryCallback,
@@ -143,6 +146,7 @@ public IrStandardOpProcessor(
         this.metaQueryCallback = metaQueryCallback;
         this.idGenerator = idGenerator;
         this.queryCache = queryCache;
+        this.graphPlanner = graphPlanner;
         this.executionClient = executionClient;
         this.printThreshold = FrontendConfig.QUERY_PRINT_THRESHOLD_MS.get(configs);
         this.opentelemetryIdGenerator = IdGenerator.random();
@@ -215,6 +219,7 @@ protected void evalOpInternal(
                                         configs,
                                         ctx,
                                         queryCache,
+                                        graphPlanner,
                                         executionClient,
                                         jobId,
                                         jobName,

interactive_engine/compiler/src/main/java/com/alibaba/graphscope/gremlin/plugin/processor/LifeCycleSupplier.java

@@ -40,6 +40,7 @@
 public class LifeCycleSupplier implements Supplier<GremlinExecutor.LifeCycle> {
     private final Configs configs;
     private final QueryCache queryCache;
+    private final GraphPlanner graphPlanner;
     private final ExecutionClient client;
     private final Context ctx;
     private final BigInteger queryId;
@@ -52,6 +53,7 @@ public LifeCycleSupplier(
             Configs configs,
             Context ctx,
             QueryCache queryCache,
+            GraphPlanner graphPlanner,
             ExecutionClient client,
             BigInteger queryId,
             String queryName,
@@ -61,6 +63,7 @@ public LifeCycleSupplier(
         this.configs = configs;
         this.ctx = ctx;
         this.queryCache = queryCache;
+        this.graphPlanner = graphPlanner;
         this.client = client;
         this.queryId = queryId;
         this.queryName = queryName;
@@ -76,6 +79,7 @@ public GremlinExecutor.LifeCycle get() {
                 .beforeEval(
                         b -> {
                             b.put("graph.query.cache", queryCache);
+                            b.put("graph.planner", graphPlanner);
                             b.put("graph.meta", meta);
                         })
                 .withResult(

interactive_engine/compiler/src/main/java/com/alibaba/graphscope/gremlin/plugin/script/GremlinCalciteScriptEngineFactory.java

@@ -18,6 +18,7 @@
 
 import com.alibaba.graphscope.common.exception.FrontendException;
 import com.alibaba.graphscope.common.ir.meta.IrMeta;
+import com.alibaba.graphscope.common.ir.tools.GraphPlanner;
 import com.alibaba.graphscope.common.ir.tools.QueryCache;
 
 import org.apache.tinkerpop.gremlin.jsr223.AbstractGremlinScriptEngineFactory;
@@ -69,8 +70,10 @@ public Object eval(String script, ScriptContext ctx) throws ScriptException {
             try {
                 Bindings globalBindings = ctx.getBindings(ScriptContext.ENGINE_SCOPE);
                 QueryCache queryCache = (QueryCache) globalBindings.get("graph.query.cache");
+                GraphPlanner graphPlanner = (GraphPlanner) globalBindings.get("graph.planner");
                 IrMeta irMeta = (IrMeta) globalBindings.get("graph.meta");
-                QueryCache.Key cacheKey = queryCache.createKey(script, irMeta);
+                QueryCache.Key cacheKey =
+                        queryCache.createKey(graphPlanner.instance(script, irMeta));
                 return queryCache.get(cacheKey);
             } catch (FrontendException e) {
                 throw e;