Prevent Early Ordering Pushdown to Enable Aggregation Pushdown to MySQL by systay · Pull Request #16278 · vitessio/vitess

systay · 2024-06-27T08:01:40Z

Description

This PR introduces a crucial check in the pushOrderingUnderAggr method to prevent the premature pushdown of the ORDER BY clause during the query planning process. By ensuring that the ordering is not pushed down too early, we allow the aggregation to be effectively pushed down to MySQL, optimizing query execution.

Details:

Added a Check for Planning Phase:

When attempting to push Ordering under Aggregation, we first check which phase we are in.
This check ensures that the ordering is only pushed down after reaching the appropriate phase in the query planning process.
If the planning phase has not yet reached delegateAggregation, the method returns early, leaving the ordering intact.
Rationale:

Optimization:

Pushing down the ORDER BY clause too early can interfere with the ability to push down the aggregation to MySQL.
Aggregation pushdown to MySQL is crucial for optimizing query performance, as it allows MySQL to handle the aggregation, reducing the data that needs to be processed by the application.
By preventing early ordering pushdown, we maintain the potential for aggregation pushdown, resulting in more efficient query execution.

Impact:

This change primarily affects the query planning phase, ensuring that the ORDER BY clause is only pushed down at the correct time.
It improves the overall performance of queries involving aggregation by leveraging MySQL's capabilities more effectively.

Related Issue(s)

Issue: #16279

Checklist

"Backport to:" labels have been added if this change should be back-ported to release branches
If this change is to be back-ported to previous releases, a justification is included in the PR description
Tests were added or are not required
Did the new or modified tests pass consistently locally and on CI?
Documentation was added or is not required

vitess-bot · 2024-06-27T08:01:42Z

codecov · 2024-06-27T08:19:00Z

Codecov Report

Attention: Patch coverage is 86.79245% with 14 lines in your changes missing coverage. Please review.

Project coverage is 68.71%. Comparing base (8685b9e) to head (102889a).
Report is 2 commits behind head on main.

Files	Patch %	Lines
go/vt/vtgate/planbuilder/operators/apply_join.go	69.23%	8 Missing ⚠️
...vt/vtgate/planbuilder/operators/offset_planning.go	93.33%	2 Missing ⚠️
...vtgate/planbuilder/plancontext/planning_context.go	91.30%	2 Missing ⚠️
.../vt/vtgate/planbuilder/operators/query_planning.go	66.66%	1 Missing ⚠️
...vt/vtgate/planbuilder/operators/queryprojection.go	87.50%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #16278      +/-   ##
==========================================
- Coverage   68.71%   68.71%   -0.01%     
==========================================
  Files        1547     1547              
  Lines      198290   198286       -4     
==========================================
- Hits       136248   136244       -4     
  Misses      62042    62042

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

Signed-off-by: Andres Taylor <andres@planetscale.com>

systay · 2024-07-02T06:18:17Z

go/vt/vtgate/planbuilder/operators/aggregation_pushing.go


 	columns := &applyJoinColumns{}
 	output, err := splitAggrColumnsToLeftAndRight(ctx, rootAggr, join, !join.JoinType.IsInner(), columns, lhs, rhs)
-	join.JoinColumns = columns


This bug was not related to the rest of the PR, but it caused tests to fail, so it had to go.

systay · 2024-07-02T06:20:49Z

go/vt/vtgate/planbuilder/operators/apply_join.go

+		panic(vterrors.VT13001(fmt.Sprintf("expression depends on tables outside this join: %s", sqlparser.String(e))))
 	}

+	col.GroupBy = addToGroupBy


This is another bug that was not related to the issue being found, but it caused some plans to be missing group by expressions

GuptaManan100

LGTM

GuptaManan100 · 2024-07-02T08:50:01Z

go/vt/vtgate/planbuilder/testdata/postprocess_cases.json

                      "Name": "user",
                      "Sharded": true
                    },
-                    "FieldQuery": "select min(a.id), a.tcol1, weight_string(a.tcol1), weight_string(a.id) from `user` as a where 1 != 1 group by a.tcol1, weight_string(a.tcol1), weight_string(a.id)",


Not just an optimization, the previous plan also looks wrong. It should have been selecting weight_string(min(a.id)) and not weight_string(a.id)

harshit-gangal

could you publish some benchmarks as the PR description claims the impact is of improved performance?

harshit-gangal · 2024-07-02T10:43:04Z

go/vt/vtgate/planbuilder/testdata/postprocess_cases.json

            },
-            "FieldQuery": "select foo, col, weight_string(foo) from `user` where 1 != 1 group by col, foo, weight_string(foo)",
+            "FieldQuery": "select foo, col, weight_string(foo) from `user` where 1 != 1 group by foo, col, weight_string(foo)",
            "OrderBy": "1 ASC, (0|2) ASC",
-            "Query": "select foo, col, weight_string(foo) from `user` where id between :vtg1 and :vtg2 group by col, foo, weight_string(foo) order by `user`.col asc, foo asc",
+            "Query": "select foo, col, weight_string(foo) from `user` where id between :vtg1 and :vtg2 group by foo, col, weight_string(foo) order by `user`.col asc, foo asc",
            "Table": "`user`"


there is no impact on the output by this change.

go/vt/vtgate/planbuilder/testdata/postprocess_cases.json

harshit-gangal

We should add some e2e tests for the fix and any new cases added in the plan tests.

go/vt/vtgate/planbuilder/testdata/tpch_cases.json

harshit-gangal

I could not find how we improved the planning unless the improvements are in tpch cases.

systay · 2024-07-02T13:59:10Z

I could not find how we improved the planning unless the improvements are in tpch cases.

I added a plan to the bottom of aggr_cases.json.

https://github.com/vitessio/vitess/pull/16278/files#diff-991c94d18be04f57568bac5c003b73009599842b38ee859c489fa286325b9b30R7150

That query produces a very different plan on main:

{
  "QueryType": "SELECT",
  "Original": "select sum(user.type) from user join user_extra on user.team_id = user_extra.id group by user_extra.col order by user_extra.col",
  "Instructions": {
    "OperatorType": "Aggregate",
    "Variant": "Ordered",
    "Aggregates": "sum(0) AS sum(`user`.type)",
    "GroupBy": "1",
    "ResultColumns": 1,
    "Inputs": [
      {
        "OperatorType": "Sort",
        "Variant": "Memory",
        "OrderBy": "1 ASC",
        "Inputs": [
          {
            "OperatorType": "Join",
            "Variant": "Join",
            "JoinColumnIndexes": "L:0,R:0",
            "JoinVars": {
              "user_team_id": 1
            },
            "TableName": "`user`_user_extra",
            "Inputs": [
              {
                "OperatorType": "Route",
                "Variant": "Scatter",
                "Keyspace": {
                  "Name": "user",
                  "Sharded": true
                },
                "FieldQuery": "select `user`.type, `user`.team_id from `user` where 1 != 1",
                "Query": "select `user`.type, `user`.team_id from `user`",
                "Table": "`user`"
              },
              {
                "OperatorType": "Route",
                "Variant": "Scatter",
                "Keyspace": {
                  "Name": "user",
                  "Sharded": true
                },
                "FieldQuery": "select user_extra.col from user_extra where 1 != 1",
                "Query": "select user_extra.col from user_extra where user_extra.id = :user_team_id",
                "Table": "user_extra"
              }
            ]
          }
        ]
      }
    ]
  },
  "TablesUsed": [
    "user.user",
    "user.user_extra"
  ]
}

As you can see here, MySQL is not doing any aggregation at all.

Signed-off-by: Andres Taylor <andres@planetscale.com>

systay · 2024-07-02T15:47:47Z

                                                 │    old.txt    │               new.txt                │
                                                 │    sec/op     │    sec/op     vs base                │
ShardedAggrPushDown/user-100-user_extra-100-10      15.41m ±  6%   16.71m ± 12%   +8.41% (p=0.029 n=10)
ShardedAggrPushDown/user-100-user_extra-500-10      15.21m ±  3%   15.52m ±  1%   +1.99% (p=0.015 n=10)
ShardedAggrPushDown/user-100-user_extra-1000-10     15.23m ±  2%   15.59m ±  1%   +2.38% (p=0.005 n=10)
ShardedAggrPushDown/user-500-user_extra-100-10      73.57m ±  1%   46.18m ±  3%  -37.23% (p=0.000 n=10)
ShardedAggrPushDown/user-500-user_extra-500-10      74.65m ±  1%   76.05m ±  1%   +1.87% (p=0.002 n=10)
ShardedAggrPushDown/user-500-user_extra-1000-10     77.46m ± 12%   76.23m ±  2%        ~ (p=0.796 n=10)
ShardedAggrPushDown/user-1000-user_extra-100-10    143.79m ±  2%   58.23m ±  1%  -59.51% (p=0.000 n=10)
ShardedAggrPushDown/user-1000-user_extra-500-10     146.1m ±  1%   131.8m ±  1%   -9.78% (p=0.000 n=10)
ShardedAggrPushDown/user-1000-user_extra-1000-10    146.3m ±  2%   151.8m ±  1%   +3.80% (p=0.000 n=10)
geomean                                             55.08m         47.63m        -13.53%

                                                 │    old.txt    │                new.txt                │
                                                 │     B/op      │     B/op       vs base                │
ShardedAggrPushDown/user-100-user_extra-100-10     11.39Ki ±  1%   11.46Ki ±  2%        ~ (p=0.796 n=10)
ShardedAggrPushDown/user-100-user_extra-500-10     11.48Ki ±  4%   11.46Ki ±  4%        ~ (p=0.780 n=10)
ShardedAggrPushDown/user-100-user_extra-1000-10    11.39Ki ±  1%   11.37Ki ±  1%        ~ (p=0.837 n=10)
ShardedAggrPushDown/user-500-user_extra-100-10     13.34Ki ±  8%   12.09Ki ±  8%   -9.40% (p=0.019 n=10)
ShardedAggrPushDown/user-500-user_extra-500-10     49.11Ki ±  2%   49.13Ki ±  2%        ~ (p=0.962 n=10)
ShardedAggrPushDown/user-500-user_extra-1000-10    50.10Ki ±  2%   49.62Ki ±  1%        ~ (p=0.267 n=10)
ShardedAggrPushDown/user-1000-user_extra-100-10    15.34Ki ± 13%   12.46Ki ± 10%  -18.79% (p=0.001 n=10)
ShardedAggrPushDown/user-1000-user_extra-500-10    50.43Ki ±  5%   50.11Ki ±  4%   -0.65% (p=0.048 n=10)
ShardedAggrPushDown/user-1000-user_extra-1000-10   93.17Ki ±  2%   93.16Ki ±  2%        ~ (p=0.122 n=10)
geomean                                            24.78Ki         23.92Ki         -3.48%

                                                 │   old.txt   │              new.txt               │
                                                 │  allocs/op  │  allocs/op   vs base               │
ShardedAggrPushDown/user-100-user_extra-100-10      220.0 ± 0%    216.0 ± 0%  -1.82% (p=0.000 n=10)
ShardedAggrPushDown/user-100-user_extra-500-10      220.0 ± 0%    216.0 ± 0%  -1.82% (p=0.000 n=10)
ShardedAggrPushDown/user-100-user_extra-1000-10     220.0 ± 0%    216.0 ± 0%  -1.82% (p=0.000 n=10)
ShardedAggrPushDown/user-500-user_extra-100-10      220.0 ± 0%    216.0 ± 0%  -1.82% (p=0.000 n=10)
ShardedAggrPushDown/user-500-user_extra-500-10     1.022k ± 0%   1.018k ± 0%  -0.39% (p=0.000 n=10)
ShardedAggrPushDown/user-500-user_extra-1000-10    1.022k ± 0%   1.018k ± 0%  -0.39% (p=0.000 n=10)
ShardedAggrPushDown/user-1000-user_extra-100-10     220.0 ± 0%    216.0 ± 0%  -1.82% (p=0.000 n=10)
ShardedAggrPushDown/user-1000-user_extra-500-10    1.022k ± 0%   1.018k ± 0%  -0.39% (p=0.000 n=10)
ShardedAggrPushDown/user-1000-user_extra-1000-10   2.023k ± 0%   2.019k ± 0%  -0.20% (p=0.000 n=10)

arthurschreiber · 2024-07-02T15:48:00Z

go/vt/vtgate/planbuilder/testdata/aggr_cases.json

                        },
-                        "FieldQuery": "select count(*), :user_id + user_extra.id, weight_string(:user_id + user_extra.id) from user_extra where 1 != 1 group by :user_id + user_extra.id",
-                        "Query": "select count(*), :user_id + user_extra.id, weight_string(:user_id + user_extra.id) from user_extra group by :user_id + user_extra.id",
+                        "FieldQuery": "select count(*), :user_id + user_extra.id, weight_string(:user_id + user_extra.id) from user_extra where 1 != 1 group by :user_id + user_extra.id, weight_string(:user_id + user_extra.id)",


Does the additional grouping by weight string make sense? 🤔 If so, can you explain why?

no, they don't really make sense

They are needed for some specific cases, we have CI tests which are failing due to removal of weight_string changes.

go/test/endtoend/vtgate/queries/benchmark/benchmark_test.go

frouioui · 2024-07-02T16:38:11Z

go/vt/vtgate/planbuilder/testdata/tpch_cases.json

+                              {
+                                "OperatorType": "Projection",
+                                "Expressions": [
+                                  "sum(volume) * count(*) as revenue",


can you explain why we are not passing volume as an index, like we do for the other columns?

we are, the printing of the plan is just not showing it

Signed-off-by: Andres Taylor <andres@planetscale.com>

This reverts commit f7c0ef7. Turns out we do need them. Signed-off-by: Andres Taylor <andres@planetscale.com>

Signed-off-by: Andres Taylor <andres@planetscale.com>

go/vt/vtgate/planbuilder/operators/offset_planning.go

Signed-off-by: Andres Taylor <andres@planetscale.com>

systay requested review from GuptaManan100, arthurschreiber, frouioui and harshit-gangal as code owners June 27, 2024 08:01

systay added Type: Enhancement Logical improvement (somewhere between a bug and feature) Component: Query Serving and removed NeedsWebsiteDocsUpdate What it says NeedsBackportReason If backport labels have been applied to a PR, a justification is required labels Jun 27, 2024

github-actions bot added this to the v21.0.0 milestone Jun 27, 2024

systay changed the title ~~feat: optimise aggregation with ORDER BY~~ Prevent Early Ordering Pushdown to Enable Aggregation Pushdown to MySQL Jun 27, 2024

systay removed NeedsDescriptionUpdate The description is not clear or comprehensive enough, and needs work NeedsIssue A linked issue is missing for this Pull Request labels Jun 27, 2024

systay marked this pull request as draft June 27, 2024 11:07

feat: optimise aggregation with ORDER BY

8fcc835

Signed-off-by: Andres Taylor <andres@planetscale.com>

systay force-pushed the push-down-aggr branch from 378f621 to 5adf8d5 Compare July 1, 2024 05:44

systay added 2 commits July 1, 2024 17:06

refactor: clean up code

fa00ad6

Signed-off-by: Andres Taylor <andres@planetscale.com>

bug: must check error before cleaning out the join columns

f9a4da5

Signed-off-by: Andres Taylor <andres@planetscale.com>

systay force-pushed the push-down-aggr branch from 5adf8d5 to f9a4da5 Compare July 1, 2024 15:13

bugfix: make sure to include all push down expressions

7277e3a

Signed-off-by: Andres Taylor <andres@planetscale.com>

systay force-pushed the push-down-aggr branch from f6ec905 to 7277e3a Compare July 2, 2024 05:40

refactor: clean up code

f3bb968

Signed-off-by: Andres Taylor <andres@planetscale.com>

systay marked this pull request as ready for review July 2, 2024 05:50

systay commented Jul 2, 2024

View reviewed changes

GuptaManan100 approved these changes Jul 2, 2024

View reviewed changes

harshit-gangal reviewed Jul 2, 2024

View reviewed changes

go/vt/vtgate/planbuilder/testdata/postprocess_cases.json Show resolved Hide resolved

harshit-gangal reviewed Jul 2, 2024

View reviewed changes

go/vt/vtgate/planbuilder/testdata/tpch_cases.json Show resolved Hide resolved

harshit-gangal reviewed Jul 2, 2024

View reviewed changes

added benchmark

455e1ee

Signed-off-by: Andres Taylor <andres@planetscale.com>

arthurschreiber reviewed Jul 2, 2024

View reviewed changes

frouioui reviewed Jul 2, 2024

View reviewed changes

systay added 2 commits July 3, 2024 10:22

remove weightstrings from group by sent to mysql

f7c0ef7

Signed-off-by: Andres Taylor <andres@planetscale.com>

undo accidental edit

28a6732

Signed-off-by: Andres Taylor <andres@planetscale.com>

systay marked this pull request as draft July 3, 2024 10:45

Merge remote-tracking branch 'upstream/main' into push-down-aggr

7579f3e

Signed-off-by: Andres Taylor <andres@planetscale.com>

systay force-pushed the push-down-aggr branch from 1d50a8d to 7579f3e Compare July 3, 2024 10:53

systay added 2 commits July 3, 2024 12:54

move UDF checking to operator transformer

a6566b8

Signed-off-by: Andres Taylor <andres@planetscale.com>

bugfix: add HAVING aggregations in a cleaner way

e6531e6

Signed-off-by: Andres Taylor <andres@planetscale.com>

systay marked this pull request as ready for review July 3, 2024 11:21

systay added 2 commits July 3, 2024 13:58

Revert "remove weightstrings from group by sent to mysql

dc42b54

This reverts commit f7c0ef7. Turns out we do need them. Signed-off-by: Andres Taylor <andres@planetscale.com>

Merge remote-tracking branch 'upstream/main' into push-down-aggr

1a43e29

Signed-off-by: Andres Taylor <andres@planetscale.com>

harshit-gangal reviewed Jul 4, 2024

View reviewed changes

go/vt/vtgate/planbuilder/operators/offset_planning.go Outdated Show resolved Hide resolved

harshit-gangal approved these changes Jul 4, 2024

View reviewed changes

change from unsupported to internal error

102889a

Signed-off-by: Andres Taylor <andres@planetscale.com>

systay merged commit 694a0cf into vitessio:main Jul 4, 2024

systay deleted the push-down-aggr branch July 4, 2024 12:30

Conversation

systay commented Jun 27, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Details:

Added a Check for Planning Phase:

Optimization:

Impact:

Related Issue(s)

Checklist

Uh oh!

vitess-bot bot commented Jun 27, 2024

Review Checklist

General

Tests

Documentation

New flags

If a workflow is added or modified:

Backward compatibility

Uh oh!

codecov bot commented Jun 27, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

GuptaManan100 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

harshit-gangal left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

harshit-gangal left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

harshit-gangal left a comment

Choose a reason for hiding this comment

Uh oh!

systay commented Jul 2, 2024

Uh oh!

systay commented Jul 2, 2024

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

systay commented Jun 27, 2024 •

edited

Loading

codecov bot commented Jun 27, 2024 •

edited

Loading