If nondeterministic code runs on Spark, this can cause a problem when recovery from failure of a node is necessary, because the new output may not be exactly the same as the old output. My interpretation is that the entire job might need to be rerun in this case, because otherwise the output data could be inconsistent with itself (as different data was produced at different times). At the very least any nodes that are downstream from the recovered node would probably need to be restarted from scratch, because they have processed data that may now change. That's my understanding of the situation anyway, please correct me if I am wrong.
My question is whether Spark can somehow automatically detect if code is nondeterministic (for example by comparing the old output to the new output) and adjust the failure recovery accordingly. If this were possible it would relieve application developers of the requirement to write nondeterministic code, which might sometimes be challenging and in any case this requirement can easily be forgotten.
No. Spark will not be able to handle non deterministic code in case of failures. The fundamental data structure of Spark, RDD is not only immutable but it
should also be determinstic function of it's input. This is necessary otherwise Spark framework will not be able to recompute the partial RDD (partition) in case of
failure. If the recomputed partition is not deterministic then it had to re-run the transformation again on full RDDs in lineage. I don't think that Spark is a right
framework for non-deterministic code.
If Spark has to be used for such use case, application developer has to take care of keeping the output consistent by writing code carefully. It can be done by using RDD only (no datframe or dataset) and persisting output after every transformation executing non-determinstic code. If performance is the concern, then the intermediate RDDs can be persisted on Alluxio.
A long term approach would be to open a feature request in apache spark jira. But I am not too positive about the acceptance of feature. A little hint in syntax to know wether code is deterministic or not and framework can switch to recover RDD partially or fully.
Non-deterministic results are not detected and accounted for in failure recovery (at least in spark 2.4.1, which I'm using).
I have encountered issues with this a few times on spark. For example, let's say I use a window function:
first_value(field_1) over (partition by field_2 order by field_3)
If field_3 is not unique, the result is non-deterministic and can differ each time that function is run. If a spark executor dies and restarts while calculating this window function, you can actually end up with two different first_value results output for the same field_2 partition.
Related
We have a spark app that does some customized aggregation on our data. And we've created a Spark Aggregator to do it, which is basically extending the org.apache.spark.sql.expressions.Aggregator and calling it through a UDAF. These aggregators basically defines a pre-aggregate process so that executors could process the data at local first and then go to exchange phase, which is pretty sweet. More on them could be found here https://spark.apache.org/docs/latest/sql-ref-functions-udf-aggregate.html
Now my question is, I've noticed that for these aggregators, the second ObjectHashAggregate process is always much much slower than the first partial ObjectHashAggregate. For example in the screenshot I've attached, the second ObjectHashAggregate is significantly slower than the first one. SparkJobScreenShot
However in our code both of them should be running kinda the same aggregation methods on the data or on the intermediate results. So why is it the second one is much slower than the first one? Is it because the hashing and exchanging after the first ObjectHashAggregate? I'm not really sure what could be causing Spark with such behavior.
We have a large RDD with millions of rows. Each row needs to be processed with a third-party optimizer that is licensed (Gurobi). We have a limited number of licenses.
We have been calling the optimizer in the Spark .map() function. The problem is that Spark will run many more mappers than it needs and throw away the results. This causes a problem with license exhaustion.
We're looking at calling Gurobi inside the Spark .foreach() method. This works, but we have two problems:
Getting the data back from the optimizer into another RDD. Our tentative plan for this is to write the results into a database (e.g. MongoDB or DynamoDB).
What happens if the node on which the .foreach() method dies? Spark guarantees that each foreach only runs once. Does it detect that it dies and restart it elsewhere? Or does something else happen?
In general if task executed with foreachPartition dies a whole job dies.
This means that, if not additional steps are taken to ensure correctness, partial result might have been acknowledged by an external system, leading to inconsistent state.
Considering limited number of licenses map or foreachPartition shouldn't make any difference. Not going into discussion if using Spark in this case makes any sense, the best way to resolve it, is to limit number of executor cores, to the number of licenses you own.
If the goal here is to limit just X number of concurrent calls, I would repartition the RDD with x, and then run a partition level operation. I think that should keep you from exhausting the licenses.
I'm pretty new to spark streaming and I need some basic clarification that I couldn't fully understand reading the documentation.
The use case is that I have a set of files containing dumping EVENTS, and each events has already inside a field TIMESTAMP.
At the moment I'm loading this file and extracting all the events in a JavaRDD and I would like to pass them to Spark Streaming in order to collect some stats based on the TIMESTAMP (a sort of replay).
My question is if it is possible to process these event using the EVENT TIMESTAMP as temporal reference instead of the actual time of the machine (sorry for the silly question).
In case it is possible, will I need simply spark streaming or I need to switch to Structured Streaming?
I found a similar question here:
Aggregate data based on timestamp in JavaDStream of spark streaming
Thanks in advance
TL;DR
yes you could use either Spark Streaming or Structured Streaming, but I wouldn't if I were you.
Detailed answer
Sorry, no simple answer to this one. Spark Streaming might be better for the per-event processing if you need to individually examine each event. Structured Streaming will be a nicer way to perform aggregations and any processing where per-event work isn't necessary.
However, there is a whole bunch of complexity in your requirements, how much of the complexity you address depends on the cost of inaccuracy in the Streaming job output.
Spark Streaming makes no guarantee that events will be processed in any kind of order. To impose ordering, you will need to setup a window in which to do your processing that minimises the risk of out-of-order processing to an acceptable level. You will need to use a big enough window of data to accurately capture your temporal ordering.
You'll need to give these points some thought:
If a batch fails and is retried, how will that affect your counters?
If events arrive late, will you ignore them, re-process the whole affected window, or update the output? If the latter how can you guarantee the update is done safely?
Will you minimise risk of corruption by keeping hold of a large window of events, or accept any inaccuracies that may arise from a smaller window?
Will the partitioning of events cause complexity in the order that they are processed?
My opinion is that, unless you have relaxed constraints over accuracy, Spark is not the right tool for the job.
I hope that helps in some way.
It is easy to do aggregations based on event-time with Spark SQL (in either batch or structured streaming). You just need to group by a time window over your timestamp column. For example, the following will bucket you data into 1 minute intervals and give you the count for each bucket.
df.groupBy(window($"timestamp", "1 minute") as 'time)
.count()
I'm coming from a Hadoop background and have limited knowledge about Spark. BAsed on what I learn so far, Spark doesn't have mapper/reducer nodes and instead it has driver/worker nodes. The worker are similar to the mapper and driver is (somehow) similar to reducer. As there is only one driver program, there will be one reducer. If so, how simple programs like word count for very big data sets can get done in spark? Because driver can simply run out of memory.
The driver is more of a controller of the work, only pulling data back if the operator calls for it. If the operator you're working on returns an RDD/DataFrame/Unit, then the data remains distributed. If it returns a native type then it will indeed pull all of the data back.
Otherwise, the concept of map and reduce are a bit obsolete here (from a type of work persopective). The only thing that really matters is whether the operation requires a data shuffle or not. You can see the points of shuffle by the stage splits either in the UI or via a toDebugString (where each indentation level is a shuffle).
All that being said, for a vague understanding, you can equate anything that requires a shuffle to a reducer. Otherwise it's a mapper.
Last, to equate to your word count example:
sc.textFile(path)
.flatMap(_.split(" "))
.map((_, 1))
.reduceByKey(_+_)
In the above, this will be done in one stage as the data loading (textFile), splitting(flatMap), and mapping can all be done independent of the rest of the data. No shuffle is needed until the reduceByKey is called as it will need to combine all of the data to perform the operation...HOWEVER, this operation has to be associative for a reason. Each node will perform the operation defined in reduceByKey locally, only merging the final data set after. This reduces both memory and network overhead.
NOTE that reduceByKey returns an RDD and is thus a transformation, so the data is shuffled via a HashPartitioner. All of the data does NOT pull back to the driver, it merely moves to nodes that have the same keys so that it can have its final value merged.
Now, if you use an action such as reduce or worse yet, collect, then you will NOT get an RDD back which means the data pulls back to the driver and you will need room for it.
Here is my fuller explanation of reduceByKey if you want more. Or how this breaks down in something like combineByKey
I currently have a spark app that reads a couple of files and forms a data frame out of them and implements some logic on the data frames.
I can see the number and size of these files growing by a lot in the future and wanted to understand what goes on behind the scenes to be able to keep up with this growth.
Firstly, I just wanted to double check that since all machines on the cluster can access the files (which is a requirement by spark), the task of reading in data from these files is distributed and no one machine is burdened by it?
I was looking at the Spark UI for this app but since it only shows what actions were performed by which machines and since "sc.textFile(filePath)" is not an action I couldn't be sure what machines are performing this read.
Secondly, what advantages/disadvantages would I face if I were to read this data from a database like Cassandra instead of just reading in files?
Thirdly, in my app I have some code where I perform a collect (val treeArr = treeDF.collect()) on the dataframe to get an array and then I have some logic implemented on those arrays. But since these are not RDDs, how does Spark distribute this work? Or does it distribute them at all?
In other words, should I be doing maximum amount of my work transforming and performing actions on RDDs than converting them into arrays or some other data structure and then implementing the logic like I would in any programming language?
I am only about two weeks into Spark so I apologize if these are stupid questions!
Yes, sc.textFile is distributed. It even has an optional minPartitions argument.
This question is too broad. But the short answer is that you should benchmark it for yourself.
collect fetches all the data to the master. After that it's just a plain array. Indeed the idea is that you should not use collect if you want to perform distributed computations.