There are two lessons you could learn from this episode:

1. Use shallow trees and the clever workaround presented in the article.

2. Don't use Spark for tasks that require complex logic.

People should trace out the line of reasoning that leads them to use tools like Spark. It is convoluted and contingent - it goes back to work done at Google in the early 2000s, when the key to getting good price / performance was using a large number of commodity machines. Because they were cheap, these machines would break often, so you needed some really smart fault tolerance technology like Hadoop/HDFS, which was followed by Spark.

The current era is completely different. Now the key to good price / performance is to light up machines on-demand and then shut them down, only paying for what you use - and perhaps using the spot market. You don't need to worry about storage - that's taken care of by the cloud provider, and you can't "bring the computation to the data" like in the old days, removing one of the big advantages of Hadoop/HDFS. Because they are doing mostly IO and networking, and because computers are just more resilient nowadays, jobs rarely fail because of hardware errors. So almost the entire rationale that led to Hadoop/HDFS/Spark is gone. But people still use Spark - and put up with "accidentally exponential behavior" - because the tech industry is do dominated by groupthink and marketing dollars.

I've hit almost the exact same issue with Hive, with a somewhat temporary workaround (like this post) to build a balanced tree out of this by reading it into a list [1] and rebuilding a binary balanced tree out of it.

But we ended up implementing a single level Multi-AND [2] so that this no longer a tree for just AND expressions & can be vectorized neater than the nested structure with a function call for each (this looks more like a tail-call rather than a recursive function).

The ORC CNF conversion has a similar massively exponential item inside which is protected by a check for 256 items or less[3].

[1] - https://github.com/t3rmin4t0r/captain-hook/blob/master/src/m...

[2] - https://issues.apache.org/jira/browse/HIVE-11398

[3] - https://github.com/apache/hive/blob/master/storage-api/src/j...

Why not just...

  val transformedLeftTemp = transform(tree.left)

  val transformedLeft = if (transformedLeftTemp.isDefined) {
    transformedLeftTemp
  } else None

Post author here. Let me know if you have any questions!

Good read - fwiw if this is your blog some of your links are broken and think they are local - https://heap.io/blog/%E2%80%9Dhttps://github.com/apache/spar...

Spark is this weird ecosystem of people who take absolutely trivial concepts in SQL, bury their heads in the sand and ignore the past 50 years of RDBMS evolution, and then write extremely complicated (or broken) and expensive to run code. But whatever it takes to get Databricks to IPO! Afterwards the hype will die down and everyone will collectively abandon it just like MongoDB except for the unfortunate companies with so much technical debt they can't extricate themselves from it.

It seems to me that the basic problem is that binary trees are the wrong data type. For instance, you can transform the tree to balance it:

    p1 AND (p2 AND (p3 AND notp4)) -> (p1 AND p2) AND (p3 AND notp4)

But the abstraction is specifying the order of operations unnecessarily. Using general trees, I think you avoid the need to transform the order of operations and "NOT" doesn't have to be a special case.

    ALL (p1 p2 p3 NOT(p4))

Is there any reason to choose binary trees for this? (Other than inertia).

Good question, the simplified example doesn't make this clear.

The real implementation has a mutable `builder` argument used to gradually build the converted filter. If we perform the `transform().isDefined` call directly on the "main" builder, but the subtree turns out to not be convertible, we can mess up the state of the builder.

The second example from the post would look roughly like this:

  val transformedLeft = if (transform(tree.left, new Builder()).isDefined) {
    transform(tree.left, mainBuilder)
  } else None

Since the two `transform` invocations are different, we can't cache the result this way.

There's a more detailed explanation in the old comment to the method: https://github.com/apache/spark/pull/24068/files#diff-5de773... .

It looks like transform(tree.left) returns an Option[Tree] already (otherwise the code would not type check) so the entire if-else in the original code seems redundant and could be replaced with:

    val transformedLeft = transform(tree.left)

It boggles my mind that the author wrote an entire long article based on this.

The rhetorical question saying that surely that weird refactor of two different functions into one, followed by calling that new, non-trivial function twice for no reason surely shouldn't affect performance.. He already lost me during the premise of the article.