Re: Apache Spark - Structured Streaming reading from Kafka some tasks take much longer
Hi Vijay: I am using spark-shell because I am still prototyping the steps involved. Regarding executors - I have 280 executors and UI only show a few straggler tasks on each trigger. The UI does not show too much time spend on GC. suspect the delay is because of getting data from kafka. The number of straggler is generally less than 5 out 240 but sometimes is higher. I will try to dig more into it and see if changing partitions etc helps but was wondering if anyone else has encountered similar stragglers holding up processing of a window trigger. Thanks On Friday, February 23, 2018 6:07 PM, vijay.bvp wrote: Instead of spark-shell have you tried running it as a job. how many executors and cores, can you share the RDD graph and event timeline on the UI and did you find which of the tasks taking more time was they are any GC please look at the UI if not already it can provide lot of information -- Sent from: http://apache-spark-user-list.1001560.n3.nabble.com/ - To unsubscribe e-mail: user-unsubscr...@spark.apache.org
Re: Apache Spark - Structured Streaming reading from Kafka some tasks take much longer
Instead of spark-shell have you tried running it as a job. how many executors and cores, can you share the RDD graph and event timeline on the UI and did you find which of the tasks taking more time was they are any GC please look at the UI if not already it can provide lot of information -- Sent from: http://apache-spark-user-list.1001560.n3.nabble.com/ - To unsubscribe e-mail: user-unsubscr...@spark.apache.org
Apache Spark - Structured Streaming reading from Kafka some tasks take much longer
Hi: I am working with spark structured streaming (2.2.1) reading data from Kafka (0.11). I need to aggregate data ingested every minute and I am using spark-shell at the moment. The message rate ingestion rate is approx 500k/second. During some trigger intervals (1 minute) especially when the streaming process is started, all tasks finish in 20seconds but during some triggers, it takes 90 seconds. I have tried to reduce the number of partitions approx (100 from 300) to reduce the consumers for Kafka, but that has not helped. I also tried the kafkaConsumer.pollTimeoutMs to 30 seconds but then I see a lot of java.util.concurrent.TimeoutException: Cannot fetch record for offset. So I wanted to see if anyone has any thoughts/recommendations. Thanks