Sometimes some Spark job which runs in our cluster runs too long not because of bad optimization, but because of bad logic of the algorithm. In most cases this is a consequence of some unnecessary joins that produce too many rows. Normally we spot such jobs by looking at Spark execution plan where we can find such joins by looking at "number of output rows: xxx" in blue stage labels.
I want to understand - is it possible to optimize this procedure and somehow automatically notify the programmer that the job has too many rows in some dataset (after execution)?
Maybe we can print this in logs (without manually counting dataset's size in code)?
Maybe after running the job we can get the output of the execution plan somehow and save it for further investigations?
No, it's not an option. Spark will do its best to optimize the query plan, so manual interaction with lower execution level is pretty much limited. However, you can "control" the rows for each jobs/tasks by changing some configurations (like spark.sql.shuffle.partitions or spark.sql.files.maxPartitionBytes), or by repartitioning data, which will cause data to be shuffled and re-distributed nearly equally between executors.
Related
If I have an application that runs the same job on the same set of columns (not necessarily same row values) every day. Is there a way I can save the spark execution plan without having spark recompute it every time?
My application requires thousands of transformations and there is significant time involved in building the lineage graph and optimization plan.
Is there a way I can save the spark execution plan without having spark recompute it every time?
I have never came across such possibility, so with large dose of confidence I can say that it's not an option.
What instead you can do it to optimize the data that is the input to Spark - optimal partitioning, compression, a format that supports predicate pushdown is probably the places where you can look for some time savings.
We are having multiple joins involving a large table (about 500gb in size). The output of the joins is stored into multiple small files each of size 800kb-1.5mb. Because of this the job is split into multiple tasks and taking a long time to complete.
We have tried using spark tuning configurations like using broadcast join, changing partition size, changing max records per file etc., But there is no performance improvement with this methods and the issue is also not fixed. Using coalesce makes the job struck at that stage and there is no progress.
Please view this link for Spark UI metrics screenshot, https://i.stack.imgur.com/FfyYy.png
The spark UI confirms your report of too many small files. You will get a file for every spark partition, and you have 33,479 in your final stage where you're writing the output. 33k partitions was probably the right number of partitions for your join but not the right number for your write.
You need to add another stage in your job that comes after your join. That 2nd needs to reduce the number of spark partitions to a reasonable number (that outputs 32MB - ~128MB files)
Something like a coalesce, or repartition. Maybe even a sort :(
You want to target ~350 partitions.
This diagram shows what you want to do manually or automatically (with spark on Databricks)
If you're using Databricks then it's easy as with Delta Lake you can turn on Auto Optimize
1.) I understand that "Spark's operators spills data to disk if it does not fit memory allowing it to run well on any sized data".
If this is true, why do we ever get OOM (Out of Memory) errors?
2.) Increasing the no. of executor cores increases parallelism. Would that also increase the chances of OOM, because the same memory is now divided into smaller parts for each core?
3.) Spark is much more susceptible to OOM because it performs operations in memory as compared to Hive, which repeatedly reads, writes into disk. Is that correct?
There is one angle that you need to consider there. You may get memory leaks if the data is not properly distributed. That means that you need to distribute your data evenly (if possible) on the Tasks so that you reduce shuffling as much as possible and make those Tasks to manage their own data. So if you need to perform a join, if data is distributed randomly, every Task (and therefore executor) will have to:
See what data they have
Send data to other executors (and tasks) to provide the same keys they need
Request the data that is needed by that task to the others
All that data exchange may cause network bottlenecks if you have a large dataset and also will make every Task to hold their data in memory plus whatever has been sent and temporary objects. All of those will blow up memory.
So to prevent that situation you can:
Load the data already repartitioned. By that I mean, if you are loading from a DB, try Spark stride as defined here. Please refer to the partitionColumn, lowerBound, upperBound attributes. That way you will create a number of partitions on the dataframe that will set the data on different tasks based on the criteria you need. If you are going to use a join of two dataframes, try similar approach on them so that partitions are similar (for not to say same) and that will prevent shuffling over network.
When you define partitions, try to make those values as evenly distributed among tasks as possible
The size of each partition should fit on memory. Although there could be spill to disk, that would slow down performance
If you don't have a column that make the data evenly distributed, try to create one that would have n number of different values, depending on the n number of tasks that you have
If you are reading from a csv, that would make it harder to create partitions, but still it's possible. You can either split the data (csv) on multiple files and create multiple dataframes (performing a union after they are loaded) or you can read that big csv and apply a repartition on the column you need. That will create shuffling as well, but it will be done once if you cache the dataframe already repartitioned
Reading from parquet it's possible that you may have multiple files but if they are not evenly distributed (because the previous process that generated didn't do it well) you may end up on OOM errors. To prevent that situation, you can load and apply repartition on the dataframe too
Or another trick valid for csv, parquet files, orc, etc. is to create a Hive table on top of that and run a query from Spark running a distribute by clause on the data, so that you can make Hive to redistribute, instead of Spark
To your question about Hive and Spark, I think you are right up to some point. Depending on the execute engine that Hive uses in your case (map/reduce, Tez, Hive on Spark, LLAP) you can have different behaviours. With map/reduce, as they are mostly disk operations, the chance to have a OOM is much lower than on Spark. Actually from Memory point of view, map/reduce is not that affected because of a skewed data distribution. But (IMHO) your goal should be to find always the best data distribution for the Spark job you are running and that will prevent that problem
Another consideration is if you are testing in a dev environment that doesn't have same data as in a prod environment. I suppose the data distribution should be similar although volumes may differ a lot (I am talking from experience ;)). In that case, when you assign Spark tuning parameters on the spark-submit command, they may be different in prod. So you need to invest some time on finding the best approach on dev and fine tune in prod
Huge majority of OOM in Spark are on the driver, not executors. This is usually a result of running .collect or similar actions on a dataset that won't fit in the driver memory.
Spark does a lot of work under the hood to parallelize the work, when using structured APIs (in contrast to RDDs) the chances of causing OOM on executor are really slim. Some combinations of cluster configuration and jobs can cause memory pressure that will impact performance and cause lots of garbage collection to happen so you need to address it, however spark should be able to handle low memory without explicit exception.
Not really - as above, Spark should be able to recover from memory issues when using structured APIs, however it may need intervention if you see garbage collection and performance impact.
By view Spark UI timeline, I find my spark application's last task of a specific stage always cost too much time. It seem the task can't finish forever, I have even waited six times longer time than normal tasks.
I want to get more information about the lask task, but I don't know how to debug this specific task, is there anyone can give me some suggestions?
Thanks for your help!
The data has been partitioned well, so the lask task don't have too much data.
Check the explain plan of the resulting dataframe to understand what operations are happening. Are there any shuffles? Sometimes when operations are performed on a dataframe(such as joins) it can result in intermediate dataframes being mapped to a smaller number of partitions and this can cause slower performance because the data isnt as distributed as can be.
Check if there are a lot of shuffles and repeated calls to such dataframes and try to cache the dataframe that comes right after a shuffle.
Check in the Spark UI (address of the driver:4040 is default) and see what the data volume of cached dataframes is, what are the processes and if there are any other overheads such as gc or if it is pure processing time.
Hope that helps.
I'm running application that loads data (.csv) from s3 into DataFrames, and than register those Dataframes as temp tables. After that, I use SparkSQL to join those tables and finally write result into db. Issue that is currently bottleneck for me is that I feel tasks are not evenly split and i get no benefits or parallelization and multiple nodes inside cluster. More precisely, this is distribution of task duration in problematic stage
task duration distribution
Is there way for me to enforce more balanced distribution ? Maybe manually writing map/reduce functions ?
Unfortunately, this stage has 6 more tasks that are still running (1.7 hours atm), which will prove even greater deviation.
There are two likely possibilities: one is under your control and .. unfortunately one is likely not ..
Skewed data. Check that the partitions are of relatively similar size - say within a factor of three or four.
Inherent variability of Spark tasks runtime. I have seen behavior of large delays in stragglers on Spark Standalone, Yarn, and Mesos without an apparent reason. The symptoms are:
extended periods (minutes) where little or no cpu or disk activity were occurring on the nodes hosting the straggler tasks
no apparent correlation of data size to the stragglers
different nodes/workers may experience the delays on subsequent runs of the same job
One thing to check: do hdfs dfsadmin -report and hdfs fsck to see if hdfs were healthy.