Skip to content

feat: Support u32 indices for HashJoinExec #16434

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 5 commits into
base: main
Choose a base branch
from

Conversation

jonathanc-n
Copy link
Contributor

Which issue does this PR close?

Rationale for this change

We can use u32 indices instead of u64 indices when there are less than u32::MAX rows when building the hashmap. This acts as a memory optimization

What changes are included in this PR?

During HashJoinExec we construct the JoinLeftData with a Box<dyn JoinHashMapType> choosing between a u32 indice or u64 JoinHashMap.

I changed the JoinHashMapType to hold the update_from_iter, get_matched_indice, and get_matched_indices_with_limit_offset and split the JoinHashMap into JoinHashMapU32 and JoinHashMapU64.

I deliberately did not try to expose a generic in the trait, nor did I try to create a generic on the JoinHashMap struct as doing so would lead to many problems with having to call preceding functions with a generic; doing so is not possible since we are determining the JoinHashMapType during runtime.

Are these changes tested?

Yes I added a test for checking the hashmap created using u32 indices.

@github-actions github-actions bot added the physical-plan Changes to the physical-plan crate label Jun 18, 2025
@jonathanc-n
Copy link
Contributor Author

cc @Dandandan

@alamb
Copy link
Contributor

alamb commented Jun 18, 2025

🤖 ./gh_compare_branch.sh Benchmark Script Running
Linux aal-dev 6.11.0-1015-gcp #15~24.04.1-Ubuntu SMP Thu Apr 24 20:41:05 UTC 2025 x86_64 x86_64 x86_64 GNU/Linux
Comparing support-u32-hashmap (732fa21) to 056f546 diff
Benchmarks: tpch_mem clickbench_partitioned clickbench_extended
Results will be posted here when complete

@alamb
Copy link
Contributor

alamb commented Jun 18, 2025

🤖: Benchmark completed

Details

Comparing HEAD and support-u32-hashmap
--------------------
Benchmark clickbench_extended.json
--------------------
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query        ┃        HEAD ┃ support-u32-hashmap ┃        Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 0     │  1891.38 ms │          1886.58 ms │     no change │
│ QQuery 1     │   697.56 ms │           704.74 ms │     no change │
│ QQuery 2     │  1381.91 ms │          1361.41 ms │     no change │
│ QQuery 3     │   669.87 ms │           634.92 ms │ +1.06x faster │
│ QQuery 4     │  1327.15 ms │          1339.32 ms │     no change │
│ QQuery 5     │ 14912.55 ms │         14887.37 ms │     no change │
│ QQuery 6     │  2044.07 ms │          2059.57 ms │     no change │
│ QQuery 7     │  1784.47 ms │          1873.16 ms │     no change │
│ QQuery 8     │   797.99 ms │           802.46 ms │     no change │
└──────────────┴─────────────┴─────────────────────┴───────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                  ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                  │ 25506.95ms │
│ Total Time (support-u32-hashmap)   │ 25549.53ms │
│ Average Time (HEAD)                │  2834.11ms │
│ Average Time (support-u32-hashmap) │  2838.84ms │
│ Queries Faster                     │          1 │
│ Queries Slower                     │          0 │
│ Queries with No Change             │          8 │
│ Queries with Failure               │          0 │
└────────────────────────────────────┴────────────┘
--------------------
Benchmark clickbench_partitioned.json
--------------------
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┓
┃ Query        ┃        HEAD ┃ support-u32-hashmap ┃       Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━┩
│ QQuery 0     │    15.40 ms │            15.56 ms │    no change │
│ QQuery 1     │    33.85 ms │            33.67 ms │    no change │
│ QQuery 2     │    81.80 ms │            80.27 ms │    no change │
│ QQuery 3     │    94.51 ms │            98.91 ms │    no change │
│ QQuery 4     │   580.34 ms │           587.97 ms │    no change │
│ QQuery 5     │   818.51 ms │           852.27 ms │    no change │
│ QQuery 6     │    22.18 ms │            23.65 ms │ 1.07x slower │
│ QQuery 7     │    36.35 ms │            36.60 ms │    no change │
│ QQuery 8     │   840.07 ms │           846.70 ms │    no change │
│ QQuery 9     │  1163.06 ms │          1126.59 ms │    no change │
│ QQuery 10    │   256.44 ms │           253.13 ms │    no change │
│ QQuery 11    │   284.27 ms │           275.93 ms │    no change │
│ QQuery 12    │   855.55 ms │           892.86 ms │    no change │
│ QQuery 13    │  1240.05 ms │          1256.85 ms │    no change │
│ QQuery 14    │   788.52 ms │           799.55 ms │    no change │
│ QQuery 15    │   756.94 ms │           759.38 ms │    no change │
│ QQuery 16    │  1588.12 ms │          1573.61 ms │    no change │
│ QQuery 17    │  1596.56 ms │          1595.53 ms │    no change │
│ QQuery 18    │  2844.62 ms │          2904.45 ms │    no change │
│ QQuery 19    │    82.55 ms │            85.96 ms │    no change │
│ QQuery 20    │  1133.70 ms │          1187.01 ms │    no change │
│ QQuery 21    │  1273.35 ms │          1310.45 ms │    no change │
│ QQuery 22    │  2112.94 ms │          2171.74 ms │    no change │
│ QQuery 23    │  7356.55 ms │          7416.88 ms │    no change │
│ QQuery 24    │   427.69 ms │           444.19 ms │    no change │
│ QQuery 25    │   299.39 ms │           302.95 ms │    no change │
│ QQuery 26    │   433.22 ms │           449.51 ms │    no change │
│ QQuery 27    │  1555.11 ms │          1553.18 ms │    no change │
│ QQuery 28    │ 11668.95 ms │         11887.97 ms │    no change │
│ QQuery 29    │   519.95 ms │           508.95 ms │    no change │
│ QQuery 30    │   768.44 ms │           778.51 ms │    no change │
│ QQuery 31    │   799.04 ms │           818.27 ms │    no change │
│ QQuery 32    │  2409.37 ms │          2374.14 ms │    no change │
│ QQuery 33    │  3138.93 ms │          3153.14 ms │    no change │
│ QQuery 34    │  3162.98 ms │          3160.19 ms │    no change │
│ QQuery 35    │  1212.81 ms │          1231.29 ms │    no change │
│ QQuery 36    │   123.25 ms │           125.52 ms │    no change │
│ QQuery 37    │    57.19 ms │            57.98 ms │    no change │
│ QQuery 38    │   127.02 ms │           124.89 ms │    no change │
│ QQuery 39    │   195.28 ms │           197.94 ms │    no change │
│ QQuery 40    │    47.18 ms │            48.68 ms │    no change │
│ QQuery 41    │    45.15 ms │            42.97 ms │    no change │
│ QQuery 42    │    38.79 ms │            39.72 ms │    no change │
└──────────────┴─────────────┴─────────────────────┴──────────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Benchmark Summary                  ┃            ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ Total Time (HEAD)                  │ 52886.01ms │
│ Total Time (support-u32-hashmap)   │ 53485.49ms │
│ Average Time (HEAD)                │  1229.91ms │
│ Average Time (support-u32-hashmap) │  1243.85ms │
│ Queries Faster                     │          0 │
│ Queries Slower                     │          1 │
│ Queries with No Change             │         42 │
│ Queries with Failure               │          0 │
└────────────────────────────────────┴────────────┘
--------------------
Benchmark tpch_mem_sf1.json
--------------------
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Query        ┃      HEAD ┃ support-u32-hashmap ┃    Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ QQuery 1     │ 100.68 ms │            99.40 ms │ no change │
│ QQuery 2     │  21.39 ms │            21.69 ms │ no change │
│ QQuery 3     │  32.64 ms │            32.28 ms │ no change │
│ QQuery 4     │  18.13 ms │            18.58 ms │ no change │
│ QQuery 5     │  48.89 ms │            49.42 ms │ no change │
│ QQuery 6     │  11.87 ms │            11.86 ms │ no change │
│ QQuery 7     │  87.37 ms │            83.45 ms │ no change │
│ QQuery 8     │  23.84 ms │            23.90 ms │ no change │
│ QQuery 9     │  53.64 ms │            53.53 ms │ no change │
│ QQuery 10    │  42.80 ms │            42.86 ms │ no change │
│ QQuery 11    │  11.21 ms │            11.22 ms │ no change │
│ QQuery 12    │  34.76 ms │            34.72 ms │ no change │
│ QQuery 13    │  26.15 ms │            26.24 ms │ no change │
│ QQuery 14    │   9.82 ms │             9.88 ms │ no change │
│ QQuery 15    │  19.41 ms │            19.59 ms │ no change │
│ QQuery 16    │  18.81 ms │            18.85 ms │ no change │
│ QQuery 17    │  96.08 ms │            94.52 ms │ no change │
│ QQuery 18    │ 193.90 ms │           189.03 ms │ no change │
│ QQuery 19    │  25.54 ms │            25.12 ms │ no change │
│ QQuery 20    │  33.31 ms │            31.69 ms │ no change │
│ QQuery 21    │ 148.25 ms │           146.30 ms │ no change │
│ QQuery 22    │  15.18 ms │            15.31 ms │ no change │
└──────────────┴───────────┴─────────────────────┴───────────┘
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━┓
┃ Benchmark Summary                  ┃           ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━┩
│ Total Time (HEAD)                  │ 1073.66ms │
│ Total Time (support-u32-hashmap)   │ 1059.45ms │
│ Average Time (HEAD)                │   48.80ms │
│ Average Time (support-u32-hashmap) │   48.16ms │
│ Queries Faster                     │         0 │
│ Queries Slower                     │         0 │
│ Queries with No Change             │        22 │
│ Queries with Failure               │         0 │
└────────────────────────────────────┴───────────┘

@jonathanc-n
Copy link
Contributor Author

Those benchmarks make sense, just saves memory.

@alamb alamb requested review from Dandandan and Copilot June 19, 2025 11:32
Copy link
Contributor

@Copilot Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull Request Overview

Adds support for using 32-bit row indices in the hash join executor when the build side has fewer than u32::MAX rows to save memory.

  • Introduce JoinHashMapType trait with JoinHashMapU32 and JoinHashMapU64 implementations
  • Change all callers and constructors of the old JoinHashMap to use Box<dyn JoinHashMapType>
  • Update memory estimation to pick u32 or u64 variant based on row count and add corresponding tests

Reviewed Changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file
File Description
joins/utils.rs Remove obsolete JoinHashMap export, keep only JoinHashMapType
joins/symmetric_hash_join.rs Update get_matched_indices calls to accept boxed iterator
joins/stream_join_utils.rs Implement JoinHashMapType for PruningJoinHashMap using new helpers
joins/join_hash_map.rs Define JoinHashMapType, JoinHashMapU32, JoinHashMapU64, and generic helpers
joins/hash_join.rs Construct boxed hash map variant, update memory estimation, and adjust tests
Comments suppressed due to low confidence (3)

datafusion/physical-plan/src/joins/join_hash_map.rs:38

  • [nitpick] Clarify this comment to indicate that the chained list uses either Vec<u32> or Vec<u64> rather than both. E.g., “stored as either Vec<u32> or Vec<u64> based on size requirements.”
/// The indices (values) are stored in a separate chained list stored as `Vec<u32>` `Vec<u64>`.

datafusion/physical-plan/src/joins/hash_join.rs:3581

  • [nitpick] The test name now differs from the u32 variant (which uses collisions plural). For consistency, consider renaming both tests to match a common pattern, e.g., join_with_hash_collision_u32 and join_with_hash_collision_u64.
    fn join_with_hash_collisions_u64() -> Result<()> {

datafusion/physical-plan/src/joins/join_hash_map.rs:93

  • [nitpick] Add a doc comment to JoinHashMapType explaining its purpose and when each method should be used; this will help maintainers understand the runtime‐selected index strategy.
pub trait JoinHashMapType: Send + Sync {

Comment on lines +73 to +75
// Flatten the deque
let next: Vec<u64> = self.next.iter().copied().collect();
get_matched_indices::<u64>(&self.map, &next, iter, deleted_offset)
Copy link
Preview

Copilot AI Jun 19, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[nitpick] Collecting the VecDeque into a new Vec<u64> on every call can be expensive. Consider reusing a buffer or accessing the deque’s contiguous slice when possible to reduce allocations.

Suggested change
// Flatten the deque
let next: Vec<u64> = self.next.iter().copied().collect();
get_matched_indices::<u64>(&self.map, &next, iter, deleted_offset)
// Access the deque's slices directly
let (head, tail) = self.next.as_slices();
get_matched_indices::<u64>(&self.map, head, tail, iter, deleted_offset)

Copilot uses AI. Check for mistakes.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This seems to be alittle bit difficult to do without changing function parameters

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
physical-plan Changes to the physical-plan crate
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Support u32 indices in HashJoinExec
2 participants