Multi-schema data pipeline - apache-spark

We want to create a Spark-based streaming data pipeline that consumes from a source (e.g. Kinesis), apply some basic transformations, and write the data to a file-based sink (e.g. s3). We have thousands of different event types coming in and the transformations would take place on a set of common fields. Once the events are transformed, they need to be split by writing them to different output locations according to the event type. This pipeline is described in the figure below:
Goals:
To infer schema safely in order to apply transformations based on the merged schema. The assumption is that the event types are compatible with each other (i.e. without overlapping schema structure) but the schema of any of them can change at unpredictable times. The pipeline should handle it dynamically.
To split the output after the transformations while keeping the original individual schema.
What we considered:
Schema inference seems to work fine on sample data. But is it safe for production usecases and for a large number of different event types?
Simply using partitionBy("type") while writing out is not enough because it would use the merged schema.

Doing the same here, casting everything to string, using marshamallow to validate, and then using from_json in a foreach like in https://www.waitingforcode.com/apache-spark-structured-streaming/two-topics-two-schemas-one-subscription-apache-spark-structured-streaming/read
seems the more reasonable approach

Related

Export spark feature transformation pipeline to a file

PMML, Mleap, PFA currently only support row based transformations. None of them support frame based transformations like aggregates or groupby or join. What is the recommended way to export a spark pipeline consisting of these operations.
I see 2 options wrt Mleap:
1) implement dataframe based transformers and the SQLTransformer-Mleap equivalent. This solution seems to be conceptually the best (since you can always encapsule such transformations in a pipeline element) but also alot of work tbh. See https://github.com/combust/mleap/issues/126
2) extend the DefaultMleapFrame with the respective operations, you want to perform and then actually apply the required actions to the data handed to the restserver within a modified MleapServing subproject.
I actually went with 2) and added implode, explode and join as methods to the DefaultMleapFrame and also a HashIndexedMleapFrame that allows for fast joins. I did not implement groupby and agg, but in Scala this is relatively easy to accomplish.
PMML and PFA are standards for representing machine learning models, not data processing pipelines. A machine learning model takes in a data record, performs some computation on it, and emits an output data record. So by definition, you are working with a single isolated data record, not a collection/frame/matrix of data records.
If you need to represent complete data processing pipelines (where the ML model is just part of the workflow) then you need to look for other/combined standards. Perhaps SQL paired with PMML would be a good choice. The idea is that you want to perform data aggregation outside of the ML model, not inside it (eg. a SQL database will be much better at it than any PMML or PFA runtime).

How to use from_json to allow for messages to have different fields?

I am trying to process data from Kafka using Spark Structured Streaming. The code for ingesting the data is as follows:
val enriched = df.select($"value" cast "string" as "json")
.select(from_json($"json", schema) as "data")
.select("data.*")
ds is a DataFrame with the data consumed from Kafka.
The problem comes when I try to read is as JSON in order to do faster queries. the function that comes from org.apache.spark.sql.functions from_json() is asking obligatory for a schema. What if the messages have some different fields?
As #zero323 and the answer he or she referenced suggest, you are asking a contradictory question: essentially how does one impose a schema when one doesn't know the schema? One can't of course. I think the idea to use open-ended collection types is your best option.
Ultimately though, it is almost certainly true that you can represent your data with a case class even if it means using a lot of Options, strings you need to parse, and maps you need to interrogate. Invest in the effort to define that case class. Otherwise, your Spark jobs will essentially a lot of ad hoc, time-consuming busywork.

Complex json log data transformation using?

I am new to data science tools and have a use case to transform json logs into a flattened columnar data maybe considered as normal csv, I was looking into a lot of alternatives (tools) to approach this problem and found that I can easily solve this using Apache Spark Sql but the problem is my json log can be a complex data structure with hierarchical arrays i.e. I would have to explode the dataset multiple times to transform it.
The problem is I don't want to hard code the logic for data transformation as I wish to reuse the same chunk of code with different transformation logic, or to put it in a better way I want my transformation to be driven by configurations rather than code.
For the same reason I was looking into Apache Avro which provides me with liberty to define my own schema for the input, but here the problem is I am unaware if I can also define the output schema as well ? If not then it will be same as reading and filtering the avro data structure (generated) into my code logic.
One probable solution which I can think of is to define my schema along with the array fields and some flags to notify my parser to explode on them, which might be recursive as well till I transform the input schema into output i.e. generating the transformation logic based on my input and output schemas.
Is there any better approach which I am unaware of or not being able to think about ?

Implementing a Spark SQL UserDefinedAggregateFunction that performs multiple passes over a column

I've been experimenting with the UserDefinedAggregateFunction class to write aggregate functions for use in Spark SQL.
It works well for implementing single pass operations like sum(), avg() etc., but is there a trick you can use to perform multiple passes over a column?
For example, Calculating variance using the naive approach. i.e. With a first pass calculating the column mean and then a second pass that uses this value to calculate the variance. I know that there are single pass algorithms for doing this that give good approximations (as in fact implemented by Spark). I was just using this as an example of a two-pass operation.
It would be nice to be able to do the following,
spark.sql("SELECT product, MultiPassAgg(price) FROM products GROUP BY product")
I appreciate that I can do this kind of thing using Dataset / DataFrame operations in stages etc., but I was just looking clean approach as illustrated in the SQL above.
Any ideas or suggestions?
This should be possible, though the following suggestion could potentially use a large amount of memory if a large number of rows are involved in any given partition.
In the implementation of your UserDefinedAggregateFunction, set up the bufferSchema having a StructField that includes a DataType that is a collection (such as ArrayType) to act as an internal collection of inputs provided via update.
Then, in update you append each input to your collection, and in merge you combine all of the collections into a single collection. This allows you to have the full partition available for use in evaluate.
Finally, during evaluate you can operate across the entire collection of rows in any way you see fit.

Avoid the use of Java data structures in Apache Spark to avoid copying the data

I have a MySQL database with a single table containing about 100 million records (~25GB, ~5 columns). Using Apache Spark, I extract this data via a JDBC connector and store it in a DataFrame.
From here, I do some pre-processing of the data (e.g. replacing the NULL values), so I absolutely need to go through each record.
Then I would like to perform dimensionality reduction and feature selection (e.g. using PCA), perform clustering (e.g. K-Means) and later on do the testing of the model on new data.
I have implemented this in Spark's Java API, but it is too slow (for my purposes) since I do a lot of copying of the data from a DataFrame to a java.util.Vector and java.util.List (to be able to iterate over all records and do the pre-processing), and later back to a DataFrame (since PCA in Spark expects a DataFrame as input).
I have tried extracting information from the database into a org.apache.spark.sql.Column but cannot find a way to iterate over it.
I also tried avoiding the use of Java data structures (such as List and Vector) by using the org.apache.spark.mllib.linalg.{DenseVector, SparseVector}, but cannot get that to work either.
Finally, I also considered using JavaRDD (by creating it from a DataFrame and a custom schema), but couldn't work it out entirely.
After a lengthy description, my question is: is there a way to do all steps mentioned in the first paragraph, without copying all the data into a Java data structure?
Maybe one of the options I tried could actually work, but I just can't seem to find out how, as the docs and literature on Spark are a bit scarce.
From the wording of your question, it seems there is some confusion about the stages of Spark processing.
First, we tell Spark what to do by specifying inputs and transformations. At this point, the only things that are known are (a) the number of partitions at various stages of processing and (b) the schema of the data. org.apache.spark.sql.Column is used at this stage to identify the metadata associated with a column. However, it doesn't contain any of the data. In fact, there is no data at all at this stage.
Second, we tell Spark to execute an action on a dataframe/dataset. This is what kicks off processing. The input is read and flows through the various transformations and into the final action operation, be it collect or save or something else.
So, that explains why you cannot "extract information from the database into" a Column.
As for the core of your question, it's hard to comment without seeing your code and knowing exactly what it is you are trying to accomplish but it is safe to say that much migrating between types is a bad idea.
Here are a couple of questions that might help guide you to a better outcome:
Why can't you perform the data transformations you need by operating directly on the Row instances?
Would it be convenient to wrap some of your transformation code into a UDF or UDAF?
Hope this helps.

Resources