Optimized version of `SortPreservingMerge` that doesn't actually compare sort keys of the key ranges are ordered

[alamb]

Is your feature request related to a problem or challenge?

When merging a large number of pre-sorted streams (e.g. in our case, a large number of pre-sorted parquet files) the actual work in SortPreservingMerge to keep them sorted is often substantial (as the sort key of each row in each stream must be compared the other potential candidates)

Here is the sort preserving merge

datafusion/datafusion/physical-plan/src/sorts/sort_preserving_merge.rs

Lines 39 to 67 in 4edbdd7

	/// Sort preserving merge execution plan
	///
	/// This takes an input execution plan and a list of sort expressions, and
	/// provided each partition of the input plan is sorted with respect to
	/// these sort expressions, this operator will yield a single partition
	/// that is also sorted with respect to them
	///
	/// ```text
	/// ┌─────────────────────────┐
	/// │ ┌───┬───┬───┬───┐ │
	/// │ │ A │ B │ C │ D │ ... │──┐
	/// │ └───┴───┴───┴───┘ │ │
	/// └─────────────────────────┘ │ ┌───────────────────┐ ┌───────────────────────────────┐
	/// Stream 1 │ │ │ │ ┌───┬───╦═══╦───┬───╦═══╗ │
	/// ├─▶│SortPreservingMerge│───▶│ │ A │ B ║ B ║ C │ D ║ E ║ ... │
	/// │ │ │ │ └───┴─▲─╩═══╩───┴───╩═══╝ │
	/// ┌─────────────────────────┐ │ └───────────────────┘ └─┬─────┴───────────────────────┘
	/// │ ╔═══╦═══╗ │ │
	/// │ ║ B ║ E ║ ... │──┘ │
	/// │ ╚═══╩═══╝ │ Note Stable Sort: the merged stream
	/// └─────────────────────────┘ places equal rows from stream 1
	/// Stream 2
	///
	///
	/// Input Streams Output stream
	/// (sorted) (sorted)
	/// ```
	#[derive(Debug)]
	pub struct SortPreservingMergeExec {

However, in some cases (such as @suremarc has identified in #6672) we can use information about how the values of the sort key columns are distributed to avoid needing a sort

For example, if we have three files that are each sorted by time and have the following ranges

• file1.paquet min(time) = 2024-01-01 and max(time) = 2024-01-31
• file2.paquet min(time) = 2024-02-01 and max(time) = 2024-02-28
• file3.paquet min(time) = 2024-03-01 and max(time) = 2024-03-31

We can produce the output sorted stream by first reading file1.parquet entirely then file2.parquet, then file3.parquet

Not only will this be faster than using SortPreservingMerge it will require less intermediate memory as we don't need to read a batch from each input stream to begin producing output. For cases where there may be 100s of files, this can minimize the amount of concurrently outstanding requests substantially

Also, for a query that will not read the entire dataset (e.g. only wants the most recent values) it can be especially beneficial:

SELECT * FROM data ORDER BY time limit 10

In this case our example above would only ever read file1.parquet (wouldn't even open the others) if it had more than 10 rows

Describe the solution you'd like

I would like an operator that does not actually merge if not needed

Describe alternatives you've considered

@NGA-TRAN implemented the following operator in InfluxDB IOx ProgressiveEval which we have found works pretty well and has offered to contribute it back upstream

We wrote about using this operator here: https://www.influxdata.com/blog/making-recent-value-queries-hundreds-times-faster/

/// ProgressiveEval return a stream of record batches in the order of its inputs.
/// It will stop when the number of output rows reach the given limit.
///
/// This takes an input execution plan and a number n, and provided each partition of
/// the input plan is in an expected order, this operator will return top record batches that covers the top n rows
/// in the order of the input plan.
///
/// ```text
/// ┌─────────────────────────┐
/// │ ┌───┬───┬───┬───┐       │
/// │ │ A │ B │ C │ D │       │──┐
/// │ └───┴───┴───┴───┘       │  │
/// └─────────────────────────┘  │  ┌───────────────────┐    ┌───────────────────────────────┐
///   Stream 1                   │  │                   │    │ ┌───┬───╦═══╦───┬───╦═══╗     │
///                              ├─▶│  ProgressiveEval  │───▶│ │ A │ B ║ C ║ D │ M ║ N ║ ... │
///                              │  │                   │    │ └───┴─▲─╩═══╩───┴───╩═══╝     │
/// ┌─────────────────────────┐  │  └───────────────────┘    └─┬─────┴───────────────────────┘
/// │ ╔═══╦═══╗               │  │
/// │ ║ M ║ N ║               │──┘                             │
/// │ ╚═══╩═══╝               │                Output only include top record batches that cover top N rows
/// └─────────────────────────┘                
///   Stream 2
///
///
///  Input Streams                                             Output stream
///  (in some order)                                           (in same order)
/// ```

Additional context

The original inspiration for this operator came from @pauldix (who I think mentioned it was inspired by ElasticSearch)

[alamb]

Here is the entire implementation in case anyone wants this: https://github.com/influxdata/influxdb3_core/blob/b546e7f86ee9adbff0dd3c5e687140848397604a/iox_query/src/provider/progressive_eval.rs

[pauldix]

Oh yeah, we're going to be hitting this all the time. Definitely want to see this! 😁

[alamb]

I probably should have mentioned that in order to use this operator, one needs to be able to determine if the input streams to SortPreservingMerge are non overlapping in value. Conveniently, that is basically the analysis that @suremarc has implemented in #9593

[alamb]

Oh yeah, we're going to be hitting this all the time. Definitely want to see this! 😁

Also to be clear to everyone else -- we have this code in InfluxDB already but it would be great to have other people be able to use it (and help maintain it)

[alamb]

FYI @ozankabak @mustafasrepo @matthewmturner

[NGA-TRAN]

Thanks @alamb. I am happy to port ProgressiveEval from InfluxDB to DB

[alamb]

I think we should both port ProgressiveEval as well as hook it up in the optimizer so it is is used (likely based on the analysis in #9593)

[NGA-TRAN]

If the hooking work is not that tricky (which I think the case), I am happy to do that, too

[suremarc]

@alamb Just to be absolutely clear, if the plan consists entirely of Parquet files from a single table, then the SortPreservingMerge will be eliminated. As you can see in this sqllogictest, a single file group is produced if none of the files overlap, and DataFusion correctly understands a SortPreservingMergeExec is not required. In particular, after #9593, #6672 should be fixed.

That said, it's worth pointing out that ProgressiveEval is more general (it can work for arbitrary plans e.g. heterogeneous unions).

[berkaysynnada]

What comes to my mind is that if we can successfully bring the statistics to SortPreservingMerge, it could handle this work without a separate operator (This will probably need a change in statistics API to return statistics for each partition).

[alamb]

What comes to my mind is that if we can successfully bring the statistics to SortPreservingMerge, it could handle this work without a separate operator (This will probably need a change in statistics API to return statistics for each partition).

I think this approach would work as well -- or maybe just a flag on the SortPreservingMerge and a different stream rather than an entirely ExecutionPlan 🤔

Per partition statistics is an interesting idea 🤔 -- it certainly makes sense for things like data sources. I wonder how generally useful it could be

[NGA-TRAN]

@alamb : I am starting porting ProgresssiveEval from IOx. Do you want me to use this ticket or #10433 for that work or you want me to open a new ticket for the porting?