perf: benchmark and sample

benchmarks/ycsb/Dockerfile

@@ -1,21 +1,22 @@
 # BUILD
 FROM maven:3.8.4-eclipse-temurin-17-alpine AS build
-# Copy over build files to docker image.
-COPY LICENSE ./
-COPY CONTRIBUTING.md ./
-COPY README.md ./
-COPY NOTIFICATIONS.md ./
-COPY logging.properties ./
-COPY src src/
-COPY pom.xml ./
-COPY license-checks.xml ./
-COPY java.header ./
-# Download dependencies
-RUN mvn dependency:go-offline
-
-# Build from source.
-RUN mvn package -Passembly -DskipTests
-
+## Copy over build files to docker image.
+#COPY LICENSE ./
+#COPY CONTRIBUTING.md ./
+#COPY README.md ./
+#COPY NOTIFICATIONS.md ./
+#COPY logging.properties ./
+#COPY src src/
+#COPY pom.xml ./
+#COPY license-checks.xml ./
+#COPY java.header ./
+## Download dependencies
+#RUN mvn dependency:go-offline
+#
+## Build from source.
+#RUN mvn package -Passembly -DskipTests
+
+COPY target/pgadapter target/pgadapter
 
 # Docker image for the YCSB runner.
 FROM gcr.io/google.com/cloudsdktool/google-cloud-cli:slim
@@ -27,7 +28,7 @@ RUN apt -y install python
 
 COPY --from=build target/pgadapter /
 
-ADD https://github.com/brianfrankcooper/YCSB/releases/download/0.17.0/ycsb-0.17.0.tar.gz /ycsb-0.17.0.tar.gz
+COPY benchmarks/ycsb/ycsb-0.17.0.tar.gz ./
 RUN tar xfvz ycsb-0.17.0.tar.gz
 RUN mv ycsb-0.17.0 ycsb
 

benchmarks/ycsb/create-and-run-ycsb-job.sh

@@ -1,16 +1,19 @@
 #!/bin/bash
 set -euo pipefail
 
-PGADAPTER_YCSB_RUNNER=pgadapter-ycsb-runner
-PGADAPTER_YCSB_JOB=pgadapter-ycsb-job
+PGADAPTER_YCSB_RUNNER=pgadapter-channels-ycsb-runner
+PGADAPTER_YCSB_JOB=pgadapter-channels-ycsb-job
 PGADAPTER_YCSB_REGION=europe-north1
+NUMBER_OF_TASKS=4
 
 SPANNER_INSTANCE=pgadapter-ycsb-regional-test
-SPANNER_DATABASE=pgadapter-ycsb-test
+SPANNER_DATABASE=pgadapter-channels-ycsb-test
 
 gcloud config set run/region $PGADAPTER_YCSB_REGION
 gcloud config set builds/region $PGADAPTER_YCSB_REGION
 
+mvn clean package -Passembly -DskipTests
+
 gcloud auth configure-docker gcr.io --quiet
 docker build --platform=linux/amd64 . -f benchmarks/ycsb/Dockerfile \
   -t gcr.io/$(gcloud config get project --quiet)/$PGADAPTER_YCSB_RUNNER
@@ -20,7 +23,7 @@ gcloud beta run jobs delete $PGADAPTER_YCSB_JOB --quiet || true
 sleep 3
 gcloud beta run jobs create $PGADAPTER_YCSB_JOB \
     --image gcr.io/$(gcloud config get project --quiet)/$PGADAPTER_YCSB_RUNNER \
-    --tasks 1 \
+    --tasks $NUMBER_OF_TASKS \
     --set-env-vars SPANNER_INSTANCE=$SPANNER_INSTANCE \
     --set-env-vars SPANNER_DATABASE=$SPANNER_DATABASE \
     --max-retries 0 \

benchmarks/ycsb/run.sh

@@ -3,13 +3,15 @@ set -euox pipefail
 
 echo "Starting Task #${CLOUD_RUN_TASK_INDEX}, Attempt #${CLOUD_RUN_TASK_ATTEMPT}..."
 EXECUTED_AT=`date +"%Y-%m-%dT%T"`
+EXECUTED_AT="${EXECUTED_AT} ${CLOUD_RUN_TASK_INDEX}"
 
 DATABASES=$(gcloud spanner databases list --instance $SPANNER_INSTANCE --filter="name:${SPANNER_DATABASE}")
 if [[ "$DATABASES" != *"$SPANNER_DATABASE"* ]]; then
   gcloud spanner databases create $SPANNER_DATABASE --instance=$SPANNER_INSTANCE --database-dialect='POSTGRESQL'
 fi
 
-java -jar pgadapter.jar -p $(gcloud --quiet config get project) -i $SPANNER_INSTANCE -r="minSessions=1000;maxSessions=1000;numChannels=20" &
+PARAMETERS="minSessions=400;maxSessions=400;numChannels=100"
+java -jar pgadapter.jar -p $(gcloud --quiet config get project) -i $SPANNER_INSTANCE -r="${PARAMETERS}" &
 sleep 6
 export PGDATABASE=$SPANNER_DATABASE
 psql -h localhost -c "CREATE TABLE IF NOT EXISTS usertable (
@@ -65,33 +67,33 @@ db.user=
 db.passwd=
 EOT
 
-psql -h localhost -c "set spanner.autocommit_dml_mode='partitioned_non_atomic'; delete from usertable;"
+#psql -h localhost -c "set spanner.autocommit_dml_mode='partitioned_non_atomic'; delete from usertable;"
 
 # Load workloada
-./bin/ycsb load jdbc -P workloads/workloada \
-  -threads 20 \
-  -p recordcount=100000 \
-  -p db.batchsize=200 \
-  -p jdbc.batchupdateapi=true \
-  -P tcp.properties \
-  -cp "jdbc-binding/lib/*" \
-  > ycsb.log
+#./bin/ycsb load jdbc -P workloads/workloada \
+#  -threads 40 \
+#  -p recordcount=5000000 \
+#  -p db.batchsize=200 \
+#  -p jdbc.batchupdateapi=true \
+#  -P tcp.properties \
+#  -cp "jdbc-binding/lib/*" \
+#  > ycsb.log
 
 # Run workloads a, b, c, f, and d.
-for WORKLOAD in a b c f d
+for WORKLOAD in a
 do
-  for DEPLOYMENT in java_tcp java_uds
+  for DEPLOYMENT in java_uds
   do
     if [ $DEPLOYMENT == 'java_tcp' ]
     then
       CONN=tcp.properties
     else
       CONN=uds.properties
     fi
-    for THREADS in 1 5 20 50 200
+    for THREADS in 100
     do
-      OPERATION_COUNT=`expr $THREADS \* 100`
-      for BATCH_SIZE in 1 10 50 200
+      OPERATION_COUNT=`expr $THREADS \* 15000`
+      for BATCH_SIZE in 10
       do
         if [ $BATCH_SIZE == 1 ]
         then
@@ -135,11 +137,11 @@ do
         INSERT_P99=$(grep '\[INSERT\], 99thPercentileLatency(us), ' ycsb.log | sed 's/^.*, //' || echo null)
 
         psql -h localhost \
-          -c "insert into run (executed_at, deployment, workload, threads, batch_size, operation_count, run_time, throughput,
+          -c "insert into run (description, executed_at, deployment, workload, threads, batch_size, operation_count, run_time, throughput,
                                read_min, read_max, read_avg, read_p50, read_p95, read_p99,
                                update_min, update_max, update_avg, update_p50, update_p95, update_p99,
                                insert_min, insert_max, insert_avg, insert_p50, insert_p95, insert_p99) values
-                               ('$EXECUTED_AT', '$DEPLOYMENT', '$WORKLOAD', $THREADS, $BATCH_SIZE, $OPERATION_COUNT, $OVERALL_RUNTIME, $OVERALL_THROUGHPUT,
+                               ('$PARAMETERS', '$EXECUTED_AT', '$DEPLOYMENT', '$WORKLOAD', $THREADS, $BATCH_SIZE, $OPERATION_COUNT, $OVERALL_RUNTIME, $OVERALL_THROUGHPUT,
                                 $READ_MIN, $READ_MAX, $READ_AVG, $READ_P50, $READ_P95, $READ_P99,
                                 $UPDATE_MIN, $UPDATE_MAX, $UPDATE_AVG, $UPDATE_P50, $UPDATE_P95, $UPDATE_P99,
                                 $INSERT_MIN, $INSERT_MAX, $INSERT_AVG, $INSERT_P50, $INSERT_P95, $INSERT_P99)"

pom.xml

@@ -34,7 +34,7 @@
       com.google.cloud.spanner.pgadapter.nodejs.NodeJSTest
     </excludedTests>
 
-    <spanner.version>6.35.2</spanner.version>
+    <spanner.version>6.38.3-SNAPSHOT</spanner.version>
     <junixsocket.version>2.6.2</junixsocket.version>
   </properties>
 

samples/ycsb/run-ycsb.sh

@@ -0,0 +1,100 @@
+
+# Make sure you have psql, Java, wget and python installed on the system.
+
+#sudo apt update && apt -y install postgresql-client
+#sudo apt -y install default-jre
+#sudo apt -y install wget
+#sudo apt -y install python
+
+# Change these variables to your database name and service account.
+SPANNER_DATABASE="projects/my-project/instances/my-instance/databases/my-database"
+CREDENTIALS=/path/to/service-account.json
+
+echo "Downloading YCSB"
+wget https://github.com/brianfrankcooper/YCSB/releases/download/0.17.0/ycsb-0.17.0.tar.gz
+
+echo "Extracting YCSB"
+tar xfvz ycsb-0.17.0.tar.gz
+mv ycsb-0.17.0 ycsb
+
+echo "Download the PostgreSQL JDBC driver and the additional libraries that are needed to use Unix Domain Sockets"
+wget -P ycsb/jdbc-binding/lib/ https://repo1.maven.org/maven2/org/postgresql/postgresql/42.5.0/postgresql-42.5.0.jar
+wget -P ycsb/jdbc-binding/lib/ https://repo1.maven.org/maven2/com/kohlschutter/junixsocket/junixsocket-common/2.6.2/junixsocket-common-2.6.2.jar
+wget -P ycsb/jdbc-binding/lib/ https://repo1.maven.org/maven2/com/kohlschutter/junixsocket/junixsocket-native-common/2.6.2/junixsocket-native-common-2.6.2.jar
+
+echo "Pull and start the PGAdapter Docker container"
+docker pull gcr.io/cloud-spanner-pg-adapter/pgadapter
+docker run \
+  --name pgadapter \
+  --rm \
+  -d -p 5432:5432 \
+  -v /tmp:/tmp \
+  -v $CREDENTIALS:/credentials.json:ro \
+  gcr.io/cloud-spanner-pg-adapter/pgadapter \
+  -c /credentials.json -x
+sleep 6
+
+
+echo "Create the 'usertable' table that is used by YCSB"
+psql -h /tmp -d $SPANNER_DATABASE \
+     -c "CREATE TABLE IF NOT EXISTS usertable (
+           YCSB_KEY VARCHAR(255) PRIMARY KEY,
+           FIELD0 TEXT, FIELD1 TEXT,
+           FIELD2 TEXT, FIELD3 TEXT,
+           FIELD4 TEXT, FIELD5 TEXT,
+           FIELD6 TEXT, FIELD7 TEXT,
+           FIELD8 TEXT, FIELD9 TEXT
+        );"
+
+cd ycsb
+
+# This replaces '/' in the database name with '%2f' to make it URL safe.
+# This is necessary when using a fully qualified database name in a JDBC connection URL. 
+URL_ENCODED_SPANNER_DATABASE=$(echo $SPANNER_DATABASE | sed -e 's/\//%2f/g')
+
+# Create a properties file that can be used with YCSB.
+# This tells YCSB which database it should connect to.
+cat <<EOT >> uds.properties
+db.driver=org.postgresql.Driver
+db.url=jdbc:postgresql://localhost/$URL_ENCODED_SPANNER_DATABASE?socketFactory=org.newsclub.net.unix.AFUNIXSocketFactory\$FactoryArg&socketFactoryArg=/tmp/.s.PGSQL.5432
+db.user=
+db.passwd=
+EOT
+
+# Delete any data currently in the `usertable` table.
+# Skip this if you want to reuse existing data.
+psql -h /tmp -d $SPANNER_DATABASE \
+     -c "set spanner.autocommit_dml_mode='partitioned_non_atomic'; delete from usertable;"
+
+# Load 'workload a' into `usertable`.
+# Skip this if you want to reuse existing data.
+# See also https://github.com/brianfrankcooper/YCSB/wiki/Running-a-Workload
+echo "Load workload a"
+echo ""
+./bin/ycsb load jdbc -P workloads/workloada \
+  -threads 20 \
+  -p recordcount=10000 \
+  -p db.batchsize=200 \
+  -p jdbc.batchupdateapi=true \
+  -P uds.properties \
+  -cp "jdbc-binding/lib/*"
+
+# Run 'workload a'.
+# Repeat this step without any of the other steps to just repeatedly run a workload.
+echo ""
+echo "---------------"
+echo "Run workload a"
+echo ""
+./bin/ycsb run jdbc -P workloads/workloada \
+  -threads 20 \
+  -p hdrhistogram.percentiles=50,95,99 \
+  -p operationcount=1000 \
+  -p recordcount=10000 \
+  -p db.batchsize=20 \
+  -p jdbc.batchupdateapi=true \
+  -P uds.properties \
+  -cp "jdbc-binding/lib/*"
+
+
+# Stop PGAdapter.
+docker stop pgadapter

src/main/java/com/google/cloud/spanner/pgadapter/ConnectionHandler.java

@@ -250,6 +250,7 @@ private String appendPropertiesToUrl(String url, Properties info) {
       return url;
     }
     StringBuilder result = new StringBuilder(url);
+    result.append(";numChannels=4");
     for (Entry<Object, Object> entry : info.entrySet()) {
       if (entry.getValue() != null && !"".equals(entry.getValue())) {
         result.append(";").append(entry.getKey()).append("=").append(entry.getValue());

src/test/java/com/google/cloud/spanner/pgadapter/JdbcMockServerTest.java

@@ -81,6 +81,9 @@
 import java.time.ZoneOffset;
 import java.time.temporal.ChronoUnit;
 import java.util.*;
+import java.util.concurrent.Callable;
+import java.util.concurrent.ExecutorService;
+import java.util.concurrent.Executors;
 import java.util.concurrent.TimeUnit;
 import java.util.stream.Collectors;
 import java.util.stream.IntStream;
@@ -337,16 +340,46 @@ private String getExpectedInitialApplicationName() {
   }
 
   @Test
-  public void testQuery() throws SQLException {
+  public void testQuery() throws SQLException, InterruptedException {
     String sql = "SELECT 1";
 
-    try (Connection connection = DriverManager.getConnection(createUrl())) {
-      try (ResultSet resultSet = connection.createStatement().executeQuery(sql)) {
-        assertTrue(resultSet.next());
-        assertEquals(1L, resultSet.getLong(1));
-        assertFalse(resultSet.next());
-      }
-    }
+    int numThreads = 128;
+    ExecutorService service = Executors.newFixedThreadPool(128);
+    for (int i = 0; i < numThreads; i++) {
+      service.submit(
+          (Callable<Void>)
+              () -> {
+                try (Connection connection = DriverManager.getConnection(createUrl())) {
+                  try (ResultSet resultSet = connection.createStatement().executeQuery(sql)) {
+                    assertTrue(resultSet.next());
+                    assertEquals(1L, resultSet.getLong(1));
+                    assertFalse(resultSet.next());
+                  }
+                }
+                return null;
+              });
+    }
+    service.shutdown();
+    service.awaitTermination(60L, TimeUnit.SECONDS);
+    System.out.println("Finished");
+    // 4 channels:
+    // 16 threads for Transport-Channel
+    // 4 grpc default executor
+    // 8 grpc default worker
+    // 5 grpc transport
+    //
+    // 8 channels:
+    // 16 threads for Transport-Channel
+    // 8 grpc default executor
+    // 8 grpc default worker
+    // 8 grpc transport
+    //
+    // 12 channels:
+    // 16 threads for Transport-Channel
+    // 5 grpc default executor
+    // 8 grpc default worker
+    // 8 grpc transport
+    //
 
     // The statement is only sent once to the mock server. The DescribePortal message will trigger
     // the execution of the query, and the result from that execution will be used for the Execute