fix(cdc): fix debezium engine config (#15536) (#15553)

Co-authored-by: StrikeW <[email protected]>

java/connector-node/risingwave-connector-service/src/main/java/com/risingwave/connector/ConnectorService.java

@@ -14,7 +14,6 @@
 
 package com.risingwave.connector;
 
-import com.risingwave.connector.source.common.DbzConnectorConfig;
 import com.risingwave.metrics.ConnectorNodeMetrics;
 import io.grpc.Server;
 import io.grpc.ServerBuilder;
@@ -37,16 +36,6 @@ public static void main(String[] args) throws Exception {
         CommandLineParser parser = new DefaultParser();
         CommandLine cmd = parser.parse(options, args);
 
-        // Quoted from the debezium document:
-        // > Your application should always properly stop the engine to ensure graceful and complete
-        // > shutdown and that each source record is sent to the application exactly one time.
-        // However, in RisingWave we assume the upstream changelog may contain duplicate events and
-        // handle conflicts in the mview operator, thus we don't need to obey the above
-        // instructions.
-        // So we decrease the wait time to 1 second here to reclaim grpc resources faster when the
-        // grpc channel is broken.
-        System.setProperty(DbzConnectorConfig.WAIT_FOR_CONNECTOR_EXIT_BEFORE_INTERRUPT_MS, "1000");
-
         int port = DEFAULT_PORT;
         String prometheusHttpHost = DEFAULT_PROMETHEUS_HOST;
         if (cmd.hasOption("p")) {

java/connector-node/risingwave-connector-service/src/main/java/com/risingwave/connector/source/common/DbzConnectorConfig.java

@@ -30,10 +30,6 @@
 public class DbzConnectorConfig {
     private static final Logger LOG = LoggerFactory.getLogger(DbzConnectorConfig.class);
 
-    /* Debezium private configs */
-    public static final String WAIT_FOR_CONNECTOR_EXIT_BEFORE_INTERRUPT_MS =
-            "debezium.embedded.shutdown.pause.before.interrupt.ms";
-
     public static final String WAIT_FOR_STREAMING_START_BEFORE_EXIT_SECS =
             "cdc.source.wait.streaming.before.exit.seconds";
 

java/connector-node/risingwave-connector-service/src/main/resources/debezium.properties

@@ -12,3 +12,10 @@ interval.handling.mode=string
 max.batch.size=${debezium.max.batch.size:-1024}
 max.queue.size=${debezium.max.queue.size:-8192}
 time.precision.mode=adaptive_time_microseconds
+# Quoted from the debezium document:
+# > Your application should always properly stop the engine to ensure graceful and complete
+# > shutdown and that each source record is sent to the application exactly one time.
+# In RisingWave we assume the upstream changelog may contain duplicate events and
+# handle conflicts in the mview operator, thus we don't need to obey the above
+# instructions. So we decrease the wait time here to reclaim jvm thread faster.
+debezium.embedded.shutdown.pause.before.interrupt.ms=1

src/jni_core/src/jvm_runtime.rs

@@ -103,13 +103,6 @@ impl JavaVmWrapper {
             .option(format!("-Djava.class.path={}", class_vec.join(":")))
             .option("-Xms16m")
             .option(format!("-Xmx{}", jvm_heap_size))
-            // Quoted from the debezium document:
-            // > Your application should always properly stop the engine to ensure graceful and complete
-            // > shutdown and that each source record is sent to the application exactly one time.
-            // In RisingWave we assume the upstream changelog may contain duplicate events and
-            // handle conflicts in the mview operator, thus we don't need to obey the above
-            // instructions. So we decrease the wait time here to reclaim jvm thread faster.
-            .option("-Ddebezium.embedded.shutdown.pause.before.interrupt.ms=1")
             .option("-Dcdc.source.wait.streaming.before.exit.seconds=30");
 
         tracing::info!("JVM args: {:?}", args_builder);