I am being new on Spark. I am facing performance issue when the number of worker nodes are increased. So to investigate that, I have tried some sample code on spark-shell.
I have created a Amazon AWS EMR with 2 worker nodes (m3.xlarge). I have used the following code on spark-shell on the master node.
var df = sqlContext.range(0,6000000000L).withColumn("col1",rand(10)).withColumn("col2",rand(20))
df.selectExpr("id","col1","col2","if(id%2=0,1,0) as key").groupBy("key").agg(avg("col1"),avg("col2")).show()
This code executed without any issues and took around 8 mins. But when I have added 2 more worker nodes (m3.xlarge) and executed the same code using spark-shell on master node, the time increased to 10 mins.
Here is the issue, I think the time should be decreased, not by half, but I should decrease. I have no idea why on increasing worker node same spark job is taking more time. Any idea why this is happening? Am I missing any thing?
This should not happen, but it is possible for an algorithm to run slower when distributed.
Basically, if the synchronization part is a heavy one, doing that with 2 nodes will take more time then with one.
I would start by comparing some simpler transformations, running a more asynchronous code, as without any sync points (such as group by key), and see if you get the same issue.
#z-star, yes an algorithm might b slow when distributed. I found the solution by using Spark Dynamic Allocation. This enable spark to use only required executors. While the static allocation runs a job on all executors, which was increasing the execution time with more nodes.
Related
I am running a query using Hive on Spark which is exhibiting some strange behavior. I've run it multiple times and observed the same behavior. The query:
reads from a large Hive external table
Spark creates about ~990,000 tasks
runs in a YARN queue with > 2900 CPUs available
uses 700 executors with 4 CPUs per executor
All is well at the start of the job. After ~1.5 hours of 2800 CPUs cranking, the job is ~80% complete (800k/990k tasks). From there, things start to nosedive: Spark stops using all of the CPUs available to it to work on tasks. With ~190k tasks to go, Spark will gradually drop from using 2800 CPUs to double digits (usually bottoming out around 20 total CPUs). This makes the last 190k tasks take significantly longer to finish than the previous 800k.
I could see as the job got very close to completing that Spark would be unable to parallelize a small amount of remaining tasks across a large number of CPUs. But with 190k tasks left to be started, it seems way too early for that.
Things I've checked:
No other job is pre-empting its resources in YARN. (In addition, if this were the case, I would expect the job to randomly lose/regain resources, instead of predictably losing steam at the 80% mark).
This occurs whether dynamic allocation is enabled or disabled. If disabled, Spark has all 2800 CPUs available for the entire run time of the job - it just doesn't use them. If enabled, Spark does spin down executors as it decides it no longer needs them.
If data skew were the issue, I could see some tasks taking longer than others to finish. But it doesn't explain why Spark wouldn't be using idle CPUs to start on the backlog of tasks still to go.
Does anyone have any advice?
For posterity, this answer from Travis Hegner contained the answer.
Setting spark.locality.wait=0s fixes this issue. I'm also not sure why a 3 second wait causes such a pile up in Spark's ability to schedule tasks, but setting to 0 makes the job run extremely well.
Using spark 2.4.4 running in YARN cluster mode with the spark FIFO scheduler.
I'm submitting multiple spark dataframe operations (i.e. writing data to S3) using a thread pool executor with a variable number of threads. This works fine if I have ~10 threads, but if I use hundreds of threads, there appears to be a deadlock, with no jobs being scheduled according to the Spark UI.
What factors control how many jobs can be scheduled concurrently? Driver resources (e.g. memory/cores)? Some other spark configuration settings?
EDIT:
Here's a brief synopsis of my code
ExecutorService pool = Executors.newFixedThreadPool(nThreads);
ExecutorCompletionService<Void> ecs = new ExecutorCompletionService<>(pool);
Dataset<Row> aHugeDf = spark.read.json(hundredsOfPaths);
List<Future<Void>> futures = listOfSeveralHundredThings
.stream()
.map(aThing -> ecs.submit(() -> {
df
.filter(col("some_column").equalTo(aThing))
.write()
.format("org.apache.hudi")
.options(writeOptions)
.save(outputPathFor(aThing));
return null;
}))
.collect(Collectors.toList());
IntStream.range(0, futures.size()).forEach(i -> ecs.poll(30, TimeUnit.MINUTES));
exec.shutdownNow();
At some point, as nThreads increases, spark no longer seems to be scheduling any jobs as evidenced by:
ecs.poll(...) timing out eventually
The Spark UI jobs tab showing no active jobs
The Spark UI executors tab showing no active tasks for any executor
The Spark UI SQL tab showing nThreads running queries with no running job ID's
My execution environment is
AWS EMR 5.28.1
Spark 2.4.4
Master node = m5.4xlarge
Core nodes = 3x rd5.24xlarge
spark.driver.cores=24
spark.driver.memory=32g
spark.executor.memory=21g
spark.scheduler.mode=FIFO
If possible write the output of the jobs to AWS Elastic MapReduce hdfs (to leverage on the almost instantaneous renames and better file IO of local hdfs) and add a dstcp step to move the files to S3, to save yourself all the troubles of handling the innards of an object store trying to be a filesystem. Also writing to local hdfs will allow you to enable speculation to control runaway tasks without falling into the deadlock traps associated with DirectOutputCommiter.
If you must use S3 as the output directory ensure that the following Spark configurations are set
spark.hadoop.mapreduce.fileoutputcommitter.algorithm.version 2
spark.speculation false
Note: DirectParquetOutputCommitter is removed from Spark 2.0 due to the chance of data loss. Unfortunately until we have improved consistency from S3a we have to work with the workarounds. Things are improving with Hadoop 2.8
Avoid keynames in lexicographic order. One could use hashing/random prefixes or reverse date-time to get around.The trick is to name your keys hierarchically, putting the most common things you filter by on the left side of your key. And never have underscores in bucket names due to DNS issues.
Enabling fs.s3a.fast.upload upload parts of a single file to Amazon S3 in parallel
Refer these articles for more detail-
Setting spark.speculation in Spark 2.1.0 while writing to s3
https://medium.com/#subhojit20_27731/apache-spark-and-amazon-s3-gotchas-and-best-practices-a767242f3d98
IMO you're likely approaching this problem wrong. Unless you can guarantee that the number of tasks per job is very low, you're likely not going to get much performance improvement by parallelizing 100s of jobs at once. Your cluster can only support 300 tasks at once, assuming you're using the default parallelism of 200 thats only 1.5 jobs. I'd suggest rewriting your code to cap max concurrent queries at 10. I highly suspect that you have 300 queries with only a single task of several hundred actually running. Most OLTP data processing system intentionally have a fairly low level of concurrent queries compared to more traditional RDS systems for this reason.
also
Apache Hudi has a default parallelism of several hundred FYI.
Why don't you just partition based on your filter column?
I would start by eliminating possible causes. Are you sure its spark that is not able to submit many jobs? Is it spark or is it YARN? If it is the later, you might need to play with the YARN scheduler settings. Could it be something to do with ExecutorService implementation that may have some limitation for the scale you are trying to achieve? Could it be hudi? With the snippet thats hard to determine.
How does the problem manifest itself other than no jobs starting up? Do you see any metrics / monitoring on the cluster or any logs that point to the problem as you say it?
If it is to do with scaling, is is possible for you to autoscale with EMR flex and see if that works for you?
How many executor cores?
Looking into these might help you narrow down or perhaps confirm the issue - unless you have already looked into these things.
(I meant to add this as comment rather than answer but text too long for comment)
Using threads or thread pools are always problematic and error prone.
I had similar problem in processing spark jobs in one of Internet of things application. I resolved using fair scheduling.
Suggestions :
Use fair scheduling (fairscheduler.xml) instead of yarn capacity scheduler
how to ? see this by using dedicated resource pools one per module. when used it will look like below spark ui
See that unit of parllelism (number of partitions ) are correct for data frames you use by seeing spark admin ui. This is spark native way of using parllelism.
I am using some AWS glue to perform some ETL operations. My program writes a computed dataframe to S3. When I look at the metrics, i find that not all my executors are being used, infact just one is being used.
How do I make sure all my allocated executors are being busy ?
Thanks.
I do not use gluecontext in my program just native sparkcontext
Not using gluecontext could be one of the reason for one executor being used.
https://docs.aws.amazon.com/glue/latest/dg/monitor-profile-debug-straggler.html
Especially read Memory Profile section:
After the first two stages, only executor number 3 is actively
consuming memory to process the data. The remaining executors are
simply idle or have been relinquished shortly after the completion of
the first two stages.
I found that my job was not using all the executors, despite having a lot of data to process. The problem was in the set up on my SparkContext. I was using SparkContext.setMaster("local[*]"), which I believe makes the job run on only one executor (driver). If that helps your problem or anyone else facing the same issue.
We are benchmarking spark with alluxio and presto with alluxio. For evaluating the performance we took 5 different queries (with some joins, group by and sort) and ran this on a dataset 650GB in orc.
Spark execution environment is setup in such a way that we have a ever running spark context and we are submitting queries using REST api (Jetty server). We are not considering first batch execution time for this load test as its taking little more time because of task deserialization and all.
What we observed while evaluating is that when we ran individual queries or even all these 5 queries executed concurrently, spark is performing very well compared to presto and is finishing all the execution in half the time than of presto.
But for actual load test, we executed 10 batches (one batch is this 5 queries submitted at the same time) with a batch interval of 60 sec. At this point presto is performing a lot better than spark. Presto finished all job in ~11 mins and spark is taking ~20 mins to complete all the task.
We tried different configurations to improve spark concurrency like
Using 20 pools with equal resource allocation and submitting jobs in a round robin fashion.
Tried using one FAIR pool and submitted all jobs to this default pool and let spark decide on resource allocations
Tuning some spark properties like spark.locality.wait and some other memory related spark properties.
All tasks are NODE_LOCAL (we replicated data in alluxio to make this happen)
Also tried playing arround with executor memory allocation, like tried with 35 small executors (5 cores, 30G) and also tried with (60core, 200G) executors.
But all are resulting in same execution time.
We used dstat on all the workers to see what was happening when spark was executing task and we could see no or minimal IO or network activity . And CPU was alway at 95%+ (Looks like its bounded on CPU) . (Saw almost similar dstat out with presto)
Can someone suggest me something which we can try to achieve similar or better results than presto?
And any explanation why presto is performing well with concurrency than spark ? We observed that presto's 1st batch is taking more time than the succeeding batches . Is presto cacheing some data in memory which spark is missing ? Or presto's resource management/ execution plan is better than spark ?
Note: Both clusters are running with same hardware configuration
I'm currently developing a Spark (v 2.2.0) Streaming application and am running into issues with the way Spark seems to be allocating work across the cluster. This application is submitted to AWS EMR using client mode, so there is a driver node and a couple of worker nodes. Here is a screenshot of Ganglia that shows memory usage in the last hour:
The left-most node is the "master" or "driver" node, and the other two are worker nodes. There are spikes in the memory usage for all three nodes that correspond to workloads coming in through the stream, but the spikes are not equal (even when scaled to % memory usage). When a large workload comes in, the driver node appears to be overworked, and the job will crash with an error regarding memory:
OpenJDK 64-Bit Server VM warning: INFO: os::commit_memory(0x000000053e980000, 674234368, 0) failed; error='Cannot allocate memory' (errno=12)
I've also run into this:
Exception in thread "streaming-job-executor-10" java.lang.OutOfMemoryError: Java heap space when the master runs out of memory, which is equally confusing, as my understanding is that "client" mode would not use the driver / master node as an executor.
Pertinent details:
As mentioned earlier, this application is submitted in client mode: spark-submit --deploy-mode client --master yarn ....
Nowhere in the program am I running collect or coalesce
Any work that I suspect of being run on a single node (jdbc reads mainly) is repartition'd after completion.
There are a couple of very, very small datasets persist into memory.
1 x Driver specs: 4 cores, 16GB RAM (m4.xlarge instance)
2 x Worker specs: 4 cores, 30.5GB RAM (r3.xlarge instance)
I have tried both allowing Spark to choose executor size / cores and specifying them manually. Both cases behave the same. (I manually specified 6 executors, 1 core, 9GB RAM)
I'm certainly at a loss here. I'm not sure what is going on in the code to be triggering the driver to hog the workload like this.
The only suspect I can think of is a code snippet similar to the following:
val scoringAlgorithm = HelperFunctions.scoring(_: Row, batchTime)
val rawScored = dataToScore.map(scoringAlgorithm)
Here, a function is being loaded from a static object, and used to map over the Dataset. It is my understanding that Spark will serialize this function across the cluster (re: http://spark.apache.org/docs/2.2.0/rdd-programming-guide.html#passing-functions-to-spark). However perhaps I am mistaken and it is simply running this transformation on the driver.
If anyone has any insight to this issue, I would love to hear it!
I ended up solving this issue. Here's how I addressed it:
I made an incorrect assertion in stating the problem: there was a collect statement at the beginning of the Spark program.
I had a transaction that required collect() to run as it was designed. My assumption was that calling repartition(n) on the resulting data would split the data back amongst the executors in the cluster. From what I can tell, this strategy does not work. Once I re-wrote this line, Spark started behaving as I expected and farming jobs out to worker nodes.
My advice to any lost soul who stumbles across this issue: don't collect unless it's the end of your Spark program. You can not recover from it. Find another way to perform your task. (I ended up switching a SQL transaction from where col in (,,,) syntax to a join on the database.)