Skip to main content



SPARK running explained - 2

SPARK running explained - 1

Set Speculative execution by setting –
1.       “spark.speculation” to true, default is false.
2.       “spark.speculation.interval” Spark checks with given interval to see if any task needs to be restarted
3.       “spark.speculation.quantile” percentage of tasks that need to complete before speculation is started for a stage
4.       “spark.speculation.multiplier” how many times a task needs to run before it needs to be restarted

Data locality means Spark tries to run tasks as close to the data location as possible. Five levels of data locality –
1.       PROCESS_LOCAL - Execute a task on the executor that cached the partition
2.       NODE_LOCAL - Execute a task on the node where the partition is available
3.       RACK_LOCAL - Execute the task on the same rack as the partition, Rack information is available from YARN cluster.
4.       NO_PREF - No preferred locations are associated with the task
5.       ANY - Default if everything else fails
Note –
1.       “spark.locality.wait” – Determines the time the scheduler waits for each locality level before moving to next. Default is 30 seconds.
2.       Set wait time specific to locality level –
a.       “spark.locality.wait.process”
b.      “spark.locality.wait.node”
c.       “spark.locality.wait.rack”

Spark memory scheduling
Spark manages the JVM heap memory allocated by the cluster manager by separating it into several segments
1.       Set “spark.executor.memory” for memory you want allocated for your executors. This memory is managed by cluster manager.
2.       Spark reserves parts of that memory for-
a.       Cached data storage – set “spark.storage.memoryFraction”, default 0.6
b.      Temporary shuffle data – set “spark.shuffle.memoryFraction”, default 0.2
c.       As above 2 parts of heap can grow before Spark can measure and limit them. There are 2 safety parameters –
                                                               i.      “spark.storage.safetyFraction”, default 0.9
                                                             ii.      “spark.shuffle.safetyFraction”, default 0.8
d.      Safety parameters lowers the memory fraction by safety fraction. Thus, defaults too-
                                                               i.      0.6*0.9 = 54%
                                                             ii.      0.2*0.8 = 16%
e.      And rest heap is for other Java objects and resources needed to run tasks.
3.       To set driver memory –
a.       Set “spark.driver.memory”
b.      If you start Spark application programmatically then that application contains your driver. Therefore, to increase the memory available to your driver, use the -Xmx Java option to set the maximum size of the Java heap of the containing process.




Running Spark on the local machine
1.       Local mode – This mode runs the entire cluster in a single JVM and is useful for testing purposes. To run Spark in local mode, set the master parameter to one of the following values –
a.       local[<n>] - Run a single executor using <n> integer threads
b.      local - Run a single executor using one thread
c.       local[*] - Run a single executor using a number of threads equal to the number of CPU cores available on the local machine.
d.      local[<n>,<f>] - Run a single executor using <n> integer threads and allow maximum <f> failures per task.
Note - If you use --master local, there will be one thread, and you may notice that log lines are missing from the driver’s output. Because, In Spark Streaming that single thread is occupied by receiver and the driver wouldn’t have any threads left to print out results. Thus, specify at least 2 threads.

2.       Local cluster mode – The difference between local cluster mode and full standalone cluster is that the master isn’t a separate process but runs in the client JVM. To run Spark in this mode, set the master –
a.       local-cluster[<n>,<c>,<m>] - <n> integer executors, and each using <c> threads, and <m> megabytes of memory. Each executor in local cluster mode runs in a separate JVM.

3.       Spark Standalone cluster mode – This is built specifically for Spark and can’t execute any other type of applications. The standalone cluster consists of –
a.       Master process
                                                               i.      Acts as the cluster manager
                                                             ii.      Accepts application to run
                                                            iii.      Schedules worker resources (CPU cores)
b.      Worker (also called slave) processes
                                                               i.      Launch application executors (also, driver in cluster-deploy mode) for task execution
Note - Spark has to be installed on all nodes in the cluster in order for them to be usable as slaves.

Spark standalone cluster running on two nodes with two workers (Cluster-deploy mode) –
1.       Client process submit application to master
2.       Master instructs one of worker to start driver
3.       Worker spawns driver JVM
4.       Master instructs workers to launch executor JVMs
5.       Workers spawns executor JVMs
6.       Driver & executors communicate with each other, independently.



Note –
1.       In a Spark standalone cluster, for an application, there can be only one executor per worker process. If you need more executors per machine, you can start multiple worker processes.
2.       If there are more applications in cluster then, each would have its own set of executors and a separate drivers
Viewing Spark processes- You can use the JVM Process Status Tool (jps command) to view them.
1.       Master and worker processes appear as “Master” and “Worker”
2.       A driver running in the cluster appears as “DriverWrapper”
3.       A driver running in client mode as “SparkSubmit”
4.       Executor processes appear as “CoarseGrainedExecutorBackend”

Specify Number of executors –
1.       To control how many executors are allocated for your application. Set “spark.cores.max” to the total number of cores you wish to use.
2.       set “spark.executor.cores” to the number of cores per executor
3.       Equivalent properties for above points --executor-cores and --total-executor-cores

Killing applications –
spark-class org.apache.spark.deploy.Client kill <master_URL> <driver_ID>

Application automatic restart –
When submitting an application in cluster-deploy mode, a special command-line option (--supervise) tells Spark to restart the driver process if it fails.


Comments

Post a Comment

Popular posts

Spring MongoDB Rename field with derived Value of another field

Input Collection -  [ { 'k' : 'Troubleshooting' , 'hour' : '2024-10-10T16' , 'v' : [ 'WebPage, Login' ] }, { 'k' : 'TroubleshootingMe' , 'hour' : '2024-10-07T01' , 'v' : [ 'Accounts, etc' ] }  ] Expected Output -  [ { 'hour' : '2024-10-10T16' , 'Troubleshooting' : [ 'WebPage, Login' ] }, { 'hour' : '2024-10-07T01' , 'TroubleshootingMe' : [ 'Accounts, etc' ] }  ]   Above Can be achieved by  $replaceRoot / $replaceWith as follows - { $replaceWith : { $mergeObjects : [ { hour : "$hour" }, { "$arrayToObject" : [ [ { k : "$k" , v : "$v" } ] ] } ] } } or { $replaceRoo...
Post a Comment
Read more




Spark MongoDB Connector Not leading to correct count or data while reading

  We are using Scala 2.11 , Spark 2.4 and Spark MongoDB Connector 2.4.4 Use Case 1 - We wanted to read a Shareded Mongo Collection and copy its data to another Mongo Collection. We noticed that after Spark Job successful completion. Output MongoDB did not had many records. Use Case 2 -  We read a MongoDB collection and doing count on dataframe lead to different count on each execution. Analysis,  We realized that MongoDB Spark Connector is missing data on bulk read as a dataframe. We tried various partitioner, listed on page -  https://www.mongodb.com/docs/spark-connector/v2.4/configuration/  But, none of them worked for us. Finally, we tried  MongoShardedPartitioner  this lead to constant count on each execution. But, it was greater than the actual count of records on the collection. This seems to be limitation with MongoDB Spark Connector. But,  MongoShardedPartitioner  seemed closest possible solution to this kind of situation. But, it per...
Post a Comment
Read more




Experience with MongoDB and Optimizations

  Experience with MongoDB and Optimizations Before reading below. I would like to point out that this  experience  is related to version  6.0.14-ent, having 6 shards, each shard having 3 machines, each machine is VM with 140 GB RAM and 2TB SSD. And, we had been hosting almost 36 TB of data. MongoDB is not good with Big Data Joins and/ or Big Data OLAP processing. It is mainly meant for OLTP purposes.  Instead of joining millions of keys between 2 collections. It is better to lookup data of one key from one collection then lookup it in other collection. Thus, merging data from 2 collection for same key. Its better to keep De-normalized data in one document.  Updating a document later is cumbersome.  MongoDB crash if data is overloaded. And, it has long downtime if crashed unlike other databases which fails write to database if disk space achieves certain limit. Thus, keeping database active and running for read traffic. MongoDB needs indexes for fast qu...
Post a Comment
Read more




Spring MongoDB Log Connection Pool Details - Active, Used, Waiting

  We couldn't find any direct way to log Mongo Connection pool Size. So, we did implement an indirect way as below.  This may be incorrect at times when dealing with Sharded MongoDB having Primaty & Secondary nodes. Because, connection may be used based on read prefrence - Primary, primaryPreferred, Secondary, etc. But, this gives an understanding if connections are used efficiently and there is no wait to acquire connections from pool. This can be further enhanced to log correct connection pool statistics.  1) Implement  MyConnectionPoolListener  as below -  import java.util.concurrent.atomic.AtomicInteger; import org.slf4j.Logger; import org.slf4j.LoggerFactory; import com.mongodb.event.ConnectionCheckOutFailedEvent; import com.mongodb.event.ConnectionCheckOutStartedEvent; import com.mongodb.event.ConnectionCheckedInEvent; import com.mongodb.event.ConnectionCheckedOutEvent; import com.mongodb.event.ConnectionClosedEvent; import com.mongodb.event.Conne...
Post a Comment
Read more




Spark Streaming with Kafka Leading to increase in Open File Descriptors ( Kafka )

  Open File Descriptors w.r.t Kafka brokers relates with following -  number of file descriptors to just track log segment files. Additional file descriptors to communicate via network sockets with external parties (such as clients, other brokers, Zookeeper, and Kerberos). For # 1 this is formula -  (number of partitions)*(partition size / segment size) Reference -  https://docs.cloudera.com/cdp-private-cloud-base/7.1.6/kafka-performance-tuning/topics/kafka-tune-broker-syslevel-file-descriptors.html For #2, every connection made my consumer or producer or zookeeper or  Kerberos  opens file descriptors. Note that each TCP connection creates 2 file descriptors. These connections can be for internal communication of heartbeat, or  security handshake , or data transfer to or from client (producer or consumer) When we run a Spark application integrating it with  Kafka . And, if it is not stable, meaning -  Streaming window for micro batches is les...
Post a Comment
Read more