Difference between a Lookup and a Join in Spark - apache-spark

I am wondering if anyone is aware of any discussion of Joins vs Lookups in Spark? I have seen this page : Lookup in spark dataframes where everyone basically says that joins are far superior to lookups and I was unsuccessful in my google-fu attempt to find anything backing that up or even discussing the two topics.

Such thing as lookup in Spark DataFrame simply doesn't exist, therefore it is inferior to any other solution and join (hash or broadcast) or using local data structures is the only option.

Lookups and Joins are two different concepts in relational data systems. Therefore, it doesn't really make sense in a general context to say that one is superior to the other because they have different functions. A lookup is simply finding data, sometimes using a key or hash value to optimize query speed. A join is using common elements in two data sets to create a new data set.
E.g. (completely hypothetical and abstract)
Lookup query 1
= 'Hello'
Join query 1 , query 2
=
'Hello world'
if query 2 equals world

Related

Cassandra data model with multiple conditions

I'm new to Cassandra, so I read a dozen articles about it and thus I know the basics. All the tutorials show efficient data retrieval by 1 or 2 columns and a time range. What I could not find was how to correctly model your data if you have more conditions.
I have a big events normalised database, with quite a few columns, say:
Event type
time
email
User_age
user_country
user_language
and so on.
I would need to be able to query by all columns. So in RDBMS I would query:
SELECT email FROM table WHERE time > X AND user_age BETWEEN X AND X AND user_language = 'nl' etc..
I know I can make a separate table for each column, but then I would still need to combine the results. Maybe this is not a bad approach, but I doubt it since there are no subqueries.
My question is obviously, how can I model this kind of data correctly in Cassandra?
Thanks a lot!
I would need to be able to query by all columns.
Let me stop you right there. In Cassandra, you create your tables based on your anticipated query patterns, and usually a table supports a single query. In your case, you have "quite a few" columns and you will need to duplicate that data into a table designed to support each possible query. That is going to get big and ungainly, very quickly.
Could we just add the rest as secondary indexes? there could potentially still be millions of rows in the eventtype table + merchant_id + time selection.
Secondary indexes are intended to be used on middle-of-the-road cardinality columns. So both, extremely low and extremely high cardinality columns are bad for secondary indexes. The problem, is that Cassandra will have to pick one of your nodes as a coordinator, scan the index on each node (incurring lots of network time), and then build and return the result set. It's a prescription for poor performance, that flies in-the-face of the best practices for working with a distributed database.
In short, Cassandra is not a good solution for use cases like this. It sounds like you want to be able to do OLAP-type queries, and for that you should use a tool that is better-suited for that purpose.

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.

Cassandra version differences

I started reading Cassandra the definitive guide, which is based on Cassandra 0.7. Now, I'm trying to experiment with Cassandra 2.1.5 and it seems that there's a lot of differences which is really confusing.
For example, I see that in 0.7 version CQL did not exist. On the other hand, data model seems quite different. You can now define a schema with CQL, while in version 0.7 there was no schema.
Can anyone shortly explain the differences, especially about the data model?
I understand that in 0.7 version the idea was about different length rows, that is, rows that have different number of columns. But now I understand that each column is actually a field that contains a number of parameters, so you can have as much fields as you want within the same row (same key).
Can someone summarize the differences? Maybe I did not understand correctly.
An important point to consider, is that the underlying storage model remains the same. CQL is simply an abstraction layer on top of that model, to make it easier to work with and model your data. DataStax MVP John Berryman has a great article on this: Understanding How CQL3 Maps to Cassandra’s Internal Data Structure
In this article, Berryman observes that:
The value of the CQL primary key is used internally as the row key (which in the new CQL paradigm is being called a “partition key”).
The names of the non-primary key CQL fields are used internally as columns names. The values of the non-primary key CQL fields are then internally stored as the corresponding column values.
Additionally, he outlines the benefits of using the CQL-based approach:
It provides fast look-up by partition key and efficient scans and slices by cluster key.
It groups together related data as CQL rows. This means that you can do in one query what would otherwise take multiple queries into different column families.
It allows for individual fields to be added, modified, and deleted independently.
It is strictly better than the old Cassandra paradigm. Proof: you can coerce CQL Tables to behave exactly like old-style Cassandra ColumnFamilies. (See the examples here.)
It extends easily to implementation of sets lists and maps (which are super ugly if you’re working directly in old cassandra) — but that’s for another blog post.
The CQL protocol allows for asynchronous communication as compared with the synchronous, call-response communication required by Thrift. As a result, CQL is capable of being much faster and less resource intensive than Thrift – especially when using single threaded clients.
can have as much fields as you want within the same row (same key).
Actually, there is a hard limit of about 2 billion columns per partition (rowkey).

Using intensive update in Map type column in Cassandra is anti-pattern?

Friends,
I am modeling a table in Cassandra which contains a Map column. So this Map should contains dynamic values and will be update so much for that row (I will update by a Primary Key)
Is it an anti-patterns, which other options should I consider ?
What you're trying to do is possibly what I described here.
First big limitations that comes into my mind are the one given by the specification:
64KB is the max size of an item in a collection
65536 is the max number of queryable elements inside a collection
More there are the problems described in other post
you can not retrieve part of a collection: even if internally each entry of a map is stored as a column you can only retrieve the whole collection (this can lead to very slow performances)
you have to choose whether creating an index on keys or on values, both simultaneously are not supported.
Since maps are typed you can't put mixed values inside: you have to represent everything as a string or bytes and then transform your data client side
I personally consider this approach as an anti pattern for all these reasons -- this approach provide a schema less solution but reduce performances and introduce lots of limitations like the one secondary indexes and typing.
HTH, Carlo

How to perform intersection operation on two datasets in Key-Value store?

Let's say I have 2 datasets, one for rules, and the other for values.
I need to filter the values based on rules.
I am using a Key-Value store (couchbase, cassandra etc.). I can use multi-get to retrieve all the values from one table, and all rules for the other one, and perform validation in a loop.
However I find this is very inefficient. I move massive volume of data (values) over the network, and the client busy working on filtering.
What is the common pattern for finding the intersection between two tables with Key-Value store?
The idea behind the nosql data model is to write data in a denormalized way so that a table can answer to a precise query. To make an example imagine you have reviews made by customers on shops. You need to know the reviews made by a user on shops and also reviews received by a shop. This would be modeled using two tables
ShopReviews
UserReviews
In the first table you query by shop id in the second by user id but data are written twice and accessed directly using just a key access.
In the same way you should organize values by rules (can't be more precise without knowing what's the relation between them) and so on. One more consideration: newer versions of nosql db supports collections which might help to model 1 to many relations.
HTH, Carlo

Resources