Umang D.

right, got it

sorry i didnt get it, Can you clarify the local version please? what do you mean by that? Local HDFS path in EMR?

no, even though I pass it, still it is set to /tmp..., but with this change it started working

Hey Sonal G. to pass the checkpoint directory to spark, I made below change to the 0.6.0 Jar and then it worked. Attaching the code block which was modified. This is for the first time setup of EMR where we have a fresh context, and to pass a directory for checkpointing. In spark-submit we have to add --conf spark.checkpoint.dir=s3://OUR_BUCKET/checkpoints/ In spark/client/src/main/java/zingg/spark/client/SparkClient.java

public void checkAndSetCheckpoint(SparkSession session) {
    SparkContext sc = session.sparkContext();
    if (sc.getCheckpointDir().isEmpty()) {
        String checkpointDir = sc.getConf().get("spark.checkpoint.dir", DEFAULT_CHECKPOINT_DIR);
        sc.setCheckpointDir(checkpointDir);
    }
}

Let me know what do you think of it.

I corrected the config still i am not able to run any phases, neither training nor matching. I will start tomorrow with a clean environment and try it out again.

right

I also tried just now with --zinggDir s3://OUR_BUCKET/checkpoints/ in the scripts/zingg.sh, still the logs gave the same WARN SparkContext: Spark is not running in local mode, therefore the checkpoint directory must not be on the local filesystem. Directory '/tmp/checkpoint' appears to be on the local filesystem.

        $SPARK_HOME/bin/spark-submit --master $SPARK_MASTER $PROPERTIES --files "$ZINGG_HOME/log4j2.properties" --conf spark.hadoop.fs.s3.impl=com.amazon.ws.emr.hadoop.fs.EmrFileSystem --conf spark.hadoop.fs.s3a.impl=com.amazon.ws.emr.hadoop.fs.EmrFileSystem --conf spark.executor.extraJavaOptions="$log4j_executor -verbose:gc -XX:+PrintGCDetails -XX:+PrintGCTimeStamps -XX:+HeapDumpOnOutOfMemoryError -Xloggc:/tmp/memLog.txt -XX:+UseCompressedOops" --conf spark.driver.extraJavaOptions="$log4j_driver" $LOGGING --driver-class-path $ZINGG_JARS $EXECUTABLE $@ --email $EMAIL --license $LICENSE --zinggDir s3://OUR_BUCKET/checkpoints/

I am passing --zinggDir s3://OUR_BUCKET/checkpoints/ as well as the config.json has the same path.

We are on 0.6.0 Zingg Version and running it on EMR 7.13 which gives Hadoop 3.4.2 and Spark 3.5.6