Author Archives: Saeed Barghi

About Saeed Barghi

Big Data/DW/ETL Consultant based in Melbourne, Australia

Airflow & Celery on Redis: when Airflow picks up old task instances

This is going to be a quick post on Airflow. We realized that in one of our environments, Airflow scheduler picks up old task instances that were already a success (whether marked as success or completed successfully). You can verify this is actually your issue by ssh into your Airflow workers, and run:

ps -ef | grep airflow

And check the DAG Run IDs: most of them are for old runs.

This happens when Celery’s Backend, in our case Redis, has old keys (or duplicate keys) of task runs. So the solution would be to clear Celery queue. And here are the steps to do it when Celery runs on Redis:

1- Stop Airflow Scheduler:

sudo initctl status airflow-scheduler

sudo initctl stop airflow-scheduler

2- Stop webserver:

sudo initctl status airflow-webserver

sudo initctl stop airflow-webserver

3- Stop Celery Flower:

cd /var/lib/airflow/bin

./ flower status

./ flower stop

4- Stop workers:

cd /var/lib/airflow/bin

./ worker status

./ worker stop

Now ssh into the server where Redis is running and type “redis-cli” and press enter to get into Redis CLI. Follow steps below to flush Redis DB:

  1. INFO keyspace — List keyspaces

    a. You should get only 1 result back

  2. SELECT 0 — Select Database
  3. config get dir —  Get database file location to take backup
  4. Copy file “xxxx.db” from above location to your home directory
  5. FLUSHDB — Flush database

Now you can start all Airflow services:

1- Scheduler commands

sudo initctl start airflow-scheduler

sudo initctl status airflow-scheduler

2- Webserver commands

sudo initctl start airflow-webserver

sudo initctl status airflow-webserver

3- Flower commands

cd /var/lib/airflow/prd/bin

nohup ./ flower start &

./ flower status

4- Worker commands

cd /var/lib/airflow/prd/bin

nohup ./ worker start &

./ worker status


Go back to Airflow and validate all DAGs are starting and completing successfully.

And he happy ever after! 🙂


How to import spark.implicits._ in Spark 2.2: error “value toDS is not a member of org.apache.spark.rdd.RDD”

I wrote about how to import implicits in spark 1.6 more than 2 years ago. But things have changed in Spark 2.2: the first thing you need to do when coding in Spark 2.2 is to set up an SparkSession object. SparkSession is the entry point to programming Spark with DataSet and DataFrame.

Like Spark 1.6, spark.implicits are required to be able to use Spark’s API for DataSets and DataFrames in version 2.2. And like version 1.6, an instance of SparkContext is needed in Spark 2.2 before being able to import spark.implicits. Since each instance of SparkSession comes with and an instance of SparkContext associated with it, all you have to do is to create an object of SparkSession and you’re set.

I have seen other posts that mention bits and pieces of how to do it. Here I give you the full code that works just fine and you can tweek it based on your requirements:

import org.apache.spark.sql._

import org.apache.log4j._

object sparkSQLWithCaseClass {

case class Person (ID: Int, name: String)

def mapper(l: String): Person = {

val fields = l.split(‘,’)

val person: Person = Person(fields(0).toInt, fields(1))

return person



def main(args: Array[String]){


val spark = SparkSession.builder.appName(“Spark SQL”).getOrCreate()


val lines = spark.sparkContext.textFile(“../../people.csv”)

val people =


import spark.implicits._

val schemaPeople = people.toDS()




val t = spark.sql(“select * from people where age >= 13”)

val res = t.collect()





Spark Error “java.lang.IllegalArgumentException: Size exceeds Integer.MAX_VALUE” in Spark 1.6

RDDs are the building blocks of Spark and what make it so powerful: they are stored in memory for fast processing. RDDs are broken down into partitions (blocks) of data, a logical piece of distributed dataset.

The underlying abstraction for blocks in Spark is a ByteBuffer, which limits the size of the block to 2 GB.

In brief, this error means that the block size for the resulting RDD is larger than 2GB:

One way to work around this issue is to increase application’s parallelism. We can define the default number of partitions in RDDs returned by join and reduceByKey, by adjusting


What this configuration parameter does is basically to define how many blocks of data our dataset, in this case RDD, is going to be divided into.

As you have probably realized by now, we would need to set spark.default.parallelism to a higher value when processing large datasets. This way we can make sure the size of data blocks do not exceed 2GB limitations.


Spark Error CoarseGrainedExecutorBackend Driver disassociated! Shutting down: Spark Memory & memoryOverhead

Another common error we saw in yarn application logs was this:

17/08/31 15:58:07 WARN CoarseGrainedExecutorBackend: An unknown (datanode-022:43969) driver disconnected.

17/08/31 15:58:07 ERROR CoarseGrainedExecutorBackend: Driver disassociated! Shutting down.

Googling this error suggests increasing spark.yarn.driver.memoryOverhead or spark.yarn.executor.memoryOverhead or both. That has apparently worked for a lot of people. Or at least those who were smart enough to understand how these properties work.

What you need to consider here is that memoryOverhead is allocated out of the total amount of memory available to driver or executor, which is controlled by spark.driver.memory & spark.executor.memory.

What this means is that if you’re increasing executor’s or driver’s memoryOverhead, double check if there is enough memory allocated to driver and executor or not. In our case, the user was allocating all the memory available to driver as memoryOverhead, which meant there was none left for other other driver operations:

spark-submit \
–queue default \
–verbose \
–master yarn-cluster \
–conf spark.shuffle.service.enabled=true \
–conf spark.shuffle.manager=sort \
–conf spark.executor.memory=8g \
–conf spark.dynamicAllocation.enabled=true \
–conf spark.dynamicAllocation.minExecutors=10 \
–conf spark.executor.cores=2 \
–conf spark.driver.memory=8g \
–conf \
–conf spark.scheduler.executorTaskBlacklistTime=3600000 \
–conf spark.yarn.driver.memoryOverhead=8192 \
–conf spark.yarn.executor.memoryOverhead=8192 \

You can clearly see what I meant in above paragraph. Instead of doing this, user should have increased executor and driver memory according to increase in executor memory overhead:

spark-submit \
–queue default \
–verbose \
–master yarn-cluster \
–conf spark.shuffle.service.enabled=true \
–conf spark.shuffle.manager=sort \
–conf spark.executor.memory=16g \
–conf spark.dynamicAllocation.enabled=true \
–conf spark.dynamicAllocation.minExecutors=10 \
–conf spark.executor.cores=2 \
–conf spark.driver.memory=16g \
–conf \
–conf spark.scheduler.executorTaskBlacklistTime=3600000 \
–conf spark.yarn.driver.memoryOverhead=8192 \
–conf spark.yarn.executor.memoryOverhead=8192 \



Spark Error: Failed to Send RPC to Datanode

This past week we had quite few issues with users not being able to run Spark jobs running in YARN Cluster mode. Particularly a team that was on tight schedule used to get errors like this all the time: Failed to send RPC 8277242275361198650 to datanode-055: java.nio.channels.ClosedChannelException

Mostly accompanied by error messages like:

org.apache.spark.SparkException: Error sending message [message = Heartbeat(9,[Lscala.Tuple2;@e47ba81,BlockManagerId(9, datanode-50 , 43381))]

ERROR Executor: Exit as unable to send heartbeats to driver more than 60 times

These errors basically mean the connection between Spark driver and executors are broken, mainly because executor is killed. This could happen because of a number of reasons:

1- We realized this happens a lot more often when our cluster is too busy and has hit maximum usage. What it means is that executors are accepted to DataNodes, but they fail to acquire enough memory on the datanode and therefore get killed.

2- Metaspace attempts to grow beyond the executor(JVM) memory limits, resulting in loss of executors.The best way to stop this error from appearing is to set below properties when launching Spark-Shell or submitting application using spark-submit:

spark.driver.extraJavaOptions = -XX:ReservedCodeCacheSize=100M-XX:MaxMetaspaceSize=256m


spark.executor.extraJavaOptions = -XX:ReservedCodeCacheSize=100M



Please note that depending on your project and code, you may need to increase the values mentioned above.

3- Network is slow for whatever reason. In our case, this was caused by a change in DNS which resulted in turning off caching.This case could be fixed by adjusting spark.executor.heartbeatInterval and Default values for these 2 parameters are 10s and 120s. You can adjust these 2 values based on how your network, the only point to consider here is that the later property,, should be greater than the first one.

If none of what mentioned above helps your situation, then it is something you need to take to your cluster’s administrator. There could be something wrong with the datanodes where executors are sent to that admins are not aware of.

Happy coding!


YARN Capacity Scheduler: Queue Priority

Capacity Scheduler is designed to run Hadoop jobs in a shared, multi-tenant cluster in a friendly manner. Its main strength is that it guarantees specific capacity for a certain group of users by supporting multiple queues and allowing users to submit their queries into their dedicated queues. Each queue is given a fraction of total cluster capacity (RAM and CPU) and all jobs submitted to a queue will have access to the capacity dedicated to that queue.

Queue priority in Capacity Scheduler is implemented by assigning higher/lower capacity to the queues which should have higher/lower priority. Another way of making sure of this arrangement is by setting the maximum percentage of cluster resources each queue can use. Therefore to assign lower priority to a queue we should limit the amount of resource it can use.

Doing so on default queue is a bit tricky, as all the jobs submitted to the platform go through the default queue and get their Application Master Container created in there. It is a very small container that controls application execution and requests resources for YARN job submiited to the cluster. Having said that, we can use another setting in our platform that allows assigning higher priority to more important applications by setting mapred.capacity-scheduler.queue.<queue-name>.supports-priority.

To see the settings of each queue in the cluster, you should navigate to resource manager’s web UI and click on Scheduler from left menu. Then click the arrow on the left hand side of each queue to expand the settings. The 2 most important settings to check are Absolute Capacity (Queue capacity in percentage) and Absolute Max Capacity (Maximum queue capacity in percentage (%) as a float. This limits the elasticity for applications in the queue):

C Scheduler


Hive Performance Tuning

If you have been working in Big Data, you have definitely heard of Hive. Apache Hive is the data warehouse infrastructure build on top of Hadoop. I did a presentation on how to best use Apache Hive and few tips on how to best use it for one of our clients last week that I would like to share with you here. This is designed to help developers and analysts writing better queries and get result faster from Hive.

To best understand how Hive works, we need to picture it as a file system. Data in each table is divided into partitions based on partitioning strategy and each partition is stored as a physical file in hdfs. Files are replicated as many times as hdfs’s replication factor dictates, which is usually 3.


Use Hive Partitions

As you have probably guessed, partitioning can be used to limit the files our query scans to complete its job, instead of having to go through each and every file. Using partitions is more efficient than creating and then using indexes as well, as it physically limits the set of files each query scans.

The question for a developer or analyst would then be “How can I know what partitions exist for a table?” SHOW PARTITIONS are the keywords to achieve this. It can be used in Hue, Aginity, Hive CLI, or any other way you would use to query Hive tables. Partitions could be used in a query’s where clause, the same way we filter the query on any other column. you can see that as well as the difference using a partition in a query makes below:


Check Query Plan

Like any other database, hive provides a way to see the query plan it’ll use to execute the query. A query plan is the set of steps and commands the DB engine takes to execute the query and produce the result. You can use EXPLAIN  in the beginning of your query to see the query plan, or use Explain button in Hue to do the same:


I used the query we looked at previously to explain a couple of points on what to look for when checking query plans:

  1. TableScan: As its name implies, this step reads (scans) the tables for a purpose. In this example, we can clearly see that our filter on src_date is applied to the table scan. And since this column is our partition column, Hive knows clearly which files it should open and read to get the result it is looking for
  2. Check the step at which the filter on partition column is being applied. The sooner this filter is applied, the less records will be passed to the downstream steps of query execution and the more efficient it is.


Hive and Joins

When joining 2 tables, apply as much filter as possible on the bigger table in join itself, instead of where clause. This limits the number of records being joined to the smaller table and therefore, less records to work on or filter later.


Using all these techniques will help getting faster results from Hive, but nothing is more important than writing a good query. I was contacted by one of the analysts working for my client a few weeks ago, complaining that his query takes more than 12 hours to run. At the very first glance I realized he is scanning one of the biggest tables we have 5 times, with no filter on partitioned column. After spending a good hour on it and re-writing the query, we managed to get the same job done in about 25 minutes.

So, please write good queries. And always filter on partition column. Good luck!