Skip to content
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.

Sign up for GitHub

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

Jump to bottom

planner: support converting json_overlaps/contains to IndexMerge to access MVIndex #40195

Merged
merged 29 commits into from
Jan 3, 2023
Merged

planner: support converting json_overlaps/contains to IndexMerge to access MVIndex #40195

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
29 commits
Select commit Hold shift + click to select a range
adb0477
fixup
qw4990 Dec 26, 2022
473bf38
Merge remote-tracking branch 'upstream/master' into mvindex-1
qw4990 Dec 26, 2022
5f90e39
fixup
qw4990 Dec 26, 2022
f198d87
Merge branch 'master' into mvindex-1
qw4990 Dec 27, 2022
b5ad12f
fixup
qw4990 Dec 27, 2022
9f40ba6
fixup
qw4990 Dec 27, 2022
becc594
Merge remote-tracking branch 'upstream/master' into mvindex-1
qw4990 Dec 27, 2022
78e1f19
fixup
qw4990 Dec 27, 2022
c57b3f7
fixup
qw4990 Dec 27, 2022
575b43b
fixup
qw4990 Dec 27, 2022
196ef40
fixup
qw4990 Dec 27, 2022
7a177b9
fixup
qw4990 Dec 27, 2022
11d71c6
fixup
qw4990 Dec 27, 2022
9139768
fixup
qw4990 Dec 27, 2022
56edff6
Merge branch 'master' into mvindex-1
qw4990 Dec 27, 2022
558af9b
fixup
qw4990 Dec 28, 2022
4e7799e
fixup
qw4990 Dec 28, 2022
7227c45
Merge remote-tracking branch 'upstream/master' into mvindex-2
qw4990 Dec 28, 2022
26d4e5a
fixup
qw4990 Dec 28, 2022
a5004fd
fixup
qw4990 Dec 28, 2022
920e8c5
Merge branch 'master' into mvindex-2
qw4990 Dec 28, 2022
5706d09
fixup
qw4990 Dec 28, 2022
ab00124
fixup
qw4990 Dec 28, 2022
f557807
fixup
qw4990 Dec 28, 2022
7114797
fixup
qw4990 Dec 30, 2022
feeccdb
Merge branch 'master' into mvindex-2
qw4990 Dec 30, 2022
c293240
fixup
qw4990 Dec 30, 2022
27a8ee7
Merge branch 'mvindex-2' of https://github.com/qw4990/tidb into mvind…
qw4990 Dec 30, 2022
10a94b5
Merge branch 'master' into mvindex-2
qw4990 Jan 3, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
78 changes: 75 additions & 3 deletions planner/core/indexmerge_path.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,9 @@ import (
"github.com/pingcap/tidb/parser/model"
"github.com/pingcap/tidb/parser/mysql"
"github.com/pingcap/tidb/planner/util"
"github.com/pingcap/tidb/sessionctx"
"github.com/pingcap/tidb/types"
"github.com/pingcap/tidb/util/chunk"
"github.com/pingcap/tidb/util/logutil"
"github.com/pingcap/tidb/util/ranger"
"go.uber.org/zap"
Expand Down Expand Up @@ -552,6 +554,7 @@ func (ds *DataSource) generateIndexMergeJSONMVIndexPath(normalPathCnt int, filte

var jsonPath expression.Expression
var vals []expression.Expression
var indexMergeIsIntersection bool
switch sf.FuncName.L {
case ast.JSONMemberOf: // (1 member of a->'$.zip')
jsonPath = sf.GetArgs()[1]
Expand All @@ -560,10 +563,29 @@ func (ds *DataSource) generateIndexMergeJSONMVIndexPath(normalPathCnt int, filte
continue
}
vals = append(vals, v)
case ast.JSONOverlaps: // (json_overlaps(a->'$.zip', '[1, 2, 3]')
continue // TODO: support json_overlaps
case ast.JSONContains: // (json_contains(a->'$.zip', '[1, 2, 3]')
continue // TODO: support json_contains
indexMergeIsIntersection = true
jsonPath = sf.GetArgs()[0]
var ok bool
vals, ok = jsonArrayExpr2Exprs(ds.ctx, sf.GetArgs()[1])
if !ok {
continue
}
case ast.JSONOverlaps: // (json_overlaps(a->'$.zip', '[1, 2, 3]')
var jsonPathIdx int
if sf.GetArgs()[0].Equal(ds.ctx, targetJSONPath) {
jsonPathIdx = 0 // (json_overlaps(a->'$.zip', '[1, 2, 3]')
} else if sf.GetArgs()[1].Equal(ds.ctx, targetJSONPath) {
jsonPathIdx = 1 // (json_overlaps('[1, 2, 3]', a->'$.zip')
} else {
continue
}
jsonPath = sf.GetArgs()[jsonPathIdx]
var ok bool
vals, ok = jsonArrayExpr2Exprs(ds.ctx, sf.GetArgs()[1-jsonPathIdx])
if !ok {
continue
}
default:
continue
}
Expand Down Expand Up @@ -612,12 +634,62 @@ func (ds *DataSource) generateIndexMergeJSONMVIndexPath(normalPathCnt int, filte
partialPaths = append(partialPaths, partialPath)
}
indexMergePath := ds.buildIndexMergeOrPath(filters, partialPaths, filterIdx)
indexMergePath.IndexMergeIsIntersection = indexMergeIsIntersection
Copy link
Member

Choose a reason for hiding this comment

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

We are using buildIndexMergeOrPath to handle both OR and AND cases. Is that expected and correct?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Now indexMergeIsIntersection only works for ast.JSONContains. Is it any problem here?

Copy link
Member

Choose a reason for hiding this comment

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

From its name, buildIndexMergeOrPath is for the OR, not AND.

mvIndexPaths = append(mvIndexPaths, indexMergePath)
}
}
return
}

// jsonArrayExpr2Exprs converts a JsonArray expression to expression list: cast('[1, 2, 3]' as JSON) --> []expr{1, 2, 3}
func jsonArrayExpr2Exprs(sctx sessionctx.Context, jsonArrayExpr expression.Expression) ([]expression.Expression, bool) {
// only support cast(const as JSON)
arrayExpr, wrappedByJSONCast := unwrapJSONCast(jsonArrayExpr)
if !wrappedByJSONCast {
return nil, false
}
if _, isConst := arrayExpr.(*expression.Constant); !isConst {
return nil, false
}
if expression.IsMutableEffectsExpr(arrayExpr) {
return nil, false
}

jsonArray, isNull, err := jsonArrayExpr.EvalJSON(sctx, chunk.Row{})
Copy link
Member

Choose a reason for hiding this comment

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

I have two questions about this:

  1. Can we directly use the Constant from the last line to avoid the eval and transform logic here and below?
  2. It seems we need to consider it may be a parameter inside the Constant.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

If we don't use EvalJSON here, we need to convert the array string like "[1, 2, 3]" to the corresponding slice like [1, 2, 3] manually, which seems dangerous. So just using EvalJSON + JSON.ArrayGetElem seems safer and easier.

Copy link
Member

Choose a reason for hiding this comment

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

Got it. What about the second question?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

OK, I use IsMutableEffectsExpr to check it here.

if isNull || err != nil {
return nil, false
}
if jsonArray.TypeCode != types.JSONTypeCodeArray {
single, ok := jsonValue2Expr(jsonArray) // '1' -> []expr{1}
if ok {
return []expression.Expression{single}, true
}
return nil, false
}
var exprs []expression.Expression
for i := 0; i < jsonArray.GetElemCount(); i++ { // '[1, 2, 3]' -> []expr{1, 2, 3}
expr, ok := jsonValue2Expr(jsonArray.ArrayGetElem(i))
if !ok {
return nil, false
}
exprs = append(exprs, expr)
}
return exprs, true
}

func jsonValue2Expr(v types.BinaryJSON) (expression.Expression, bool) {
if v.TypeCode != types.JSONTypeCodeInt64 {
// only support INT now
// TODO: support more types
return nil, false
}
val := v.GetInt64()
return &expression.Constant{
Value: types.NewDatum(val),
RetType: types.NewFieldType(mysql.TypeLonglong),
}, true
}

func unwrapJSONCast(expr expression.Expression) (expression.Expression, bool) {
if expr == nil {
return nil, false
Expand Down
14 changes: 13 additions & 1 deletion planner/core/testdata/index_merge_suite_in.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,19 @@
"select /*+ use_index_merge(t, j0_1) */ * from t where (1 member of (j0->'$.path1')) and a<10",
"select /*+ use_index_merge(t, j0_1) */ * from t where (1 member of (j0->'$.XXX')) and a<10",
"select /*+ use_index_merge(t, j0_1) */ * from t where (1 member of (j0->'$.path1')) and (2 member of (j1)) and a<10",
"select /*+ use_index_merge(t, j1) */ * from t where (1 member of (j0->'$.path1')) and (2 member of (j1)) and a<10"
"select /*+ use_index_merge(t, j1) */ * from t where (1 member of (j0->'$.path1')) and (2 member of (j1)) and a<10",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_contains((j0->'$.path0'), '[1, 2, 3]')",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps((j0->'$.path0'), '[1, 2, 3]')",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps('[1, 2, 3]', (j0->'$.path0'))",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_contains((j0->'$.path0'), '[1, 2, 3]') and a<10",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps((j0->'$.path0'), '[1, 2, 3]') and a<10",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps('[1, 2, 3]', (j0->'$.path0')) and a<10",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_contains((j0->'$.path0'), '1')",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps((j0->'$.path0'), '1')",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps('1', (j0->'$.path0'))",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_contains((j0->'$.path0'), '1') and a<10",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps((j0->'$.path0'), '1') and a<10",
"select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps('1', (j0->'$.path0')) and a<10"
]
},
{
Expand Down
122 changes: 122 additions & 0 deletions planner/core/testdata/index_merge_suite_out.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,6 +49,128 @@
" └─Selection(Probe) 0.00 cop[tikv] lt(test.t.a, 10)",
" └─TableRowIDScan 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_contains((j0->'$.path0'), '[1, 2, 3]')",
"Plan": [
"IndexMerge 0.00 root type: intersection",
"├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
"├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[2,2], keep order:false, stats:pseudo",
"├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[3,3], keep order:false, stats:pseudo",
"└─TableRowIDScan(Probe) 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps((j0->'$.path0'), '[1, 2, 3]')",
"Plan": [
"Selection 0.00 root json_overlaps(json_extract(test.t.j0, \"$.path0\"), cast(\"[1, 2, 3]\", json BINARY))",
Copy link
Member

Choose a reason for hiding this comment

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

It looks like this Selection is redundant.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

The reason is ast.JSONOverlaps cannot be pushed down to TiKV (https://github.com/pingcap/tidb/blob/master/expression/expression.go#L1036), and this let the optimizer keeps an Selection upon the DataSrouce (https://github.com/pingcap/tidb/blob/master/planner/core/rule_predicate_push_down.go#L141), which is expected.

Copy link
Member

Choose a reason for hiding this comment

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

But the json_overlaps has already been handled by the IndexMerge.
And json_contains also doesn't have the Selection.

Copy link
Contributor Author

@qw4990 qw4990 Dec 30, 2022 •
edited
Loading

Choose a reason for hiding this comment

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

YES, but it's another problem and I'll submit another PR to fix it after. Otherwise, this PR will become hard to review.

"└─IndexMerge 0.00 root type: union",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[2,2], keep order:false, stats:pseudo",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[3,3], keep order:false, stats:pseudo",
" └─TableRowIDScan(Probe) 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps('[1, 2, 3]', (j0->'$.path0'))",
"Plan": [
"Selection 0.00 root json_overlaps(cast(\"[1, 2, 3]\", json BINARY), json_extract(test.t.j0, \"$.path0\"))",
"└─IndexMerge 0.00 root type: union",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[2,2], keep order:false, stats:pseudo",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[3,3], keep order:false, stats:pseudo",
" └─TableRowIDScan(Probe) 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_contains((j0->'$.path0'), '[1, 2, 3]') and a<10",
"Plan": [
"IndexMerge 0.00 root type: intersection",
"├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
"├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[2,2], keep order:false, stats:pseudo",
"├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[3,3], keep order:false, stats:pseudo",
"└─Selection(Probe) 0.00 cop[tikv] lt(test.t.a, 10)",
" └─TableRowIDScan 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps((j0->'$.path0'), '[1, 2, 3]') and a<10",
"Plan": [
"Selection 0.00 root json_overlaps(json_extract(test.t.j0, \"$.path0\"), cast(\"[1, 2, 3]\", json BINARY))",
"└─IndexMerge 0.00 root type: union",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[2,2], keep order:false, stats:pseudo",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[3,3], keep order:false, stats:pseudo",
" └─Selection(Probe) 0.00 cop[tikv] lt(test.t.a, 10)",
" └─TableRowIDScan 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps('[1, 2, 3]', (j0->'$.path0')) and a<10",
"Plan": [
"Selection 0.00 root json_overlaps(cast(\"[1, 2, 3]\", json BINARY), json_extract(test.t.j0, \"$.path0\"))",
"└─IndexMerge 0.00 root type: union",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[2,2], keep order:false, stats:pseudo",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[3,3], keep order:false, stats:pseudo",
" └─Selection(Probe) 0.00 cop[tikv] lt(test.t.a, 10)",
" └─TableRowIDScan 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_contains((j0->'$.path0'), '1')",
"Plan": [
"IndexMerge 0.00 root type: intersection",
"├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
"└─TableRowIDScan(Probe) 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps((j0->'$.path0'), '1')",
"Plan": [
"Selection 0.00 root json_overlaps(json_extract(test.t.j0, \"$.path0\"), cast(\"1\", json BINARY))",
"└─IndexMerge 0.00 root type: union",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
" └─TableRowIDScan(Probe) 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps('1', (j0->'$.path0'))",
"Plan": [
"Selection 0.00 root json_overlaps(cast(\"1\", json BINARY), json_extract(test.t.j0, \"$.path0\"))",
"└─IndexMerge 0.00 root type: union",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
" └─TableRowIDScan(Probe) 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_contains((j0->'$.path0'), '1') and a<10",
"Plan": [
"IndexMerge 0.00 root type: intersection",
"├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
"└─Selection(Probe) 0.00 cop[tikv] lt(test.t.a, 10)",
" └─TableRowIDScan 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps((j0->'$.path0'), '1') and a<10",
"Plan": [
"Selection 0.00 root json_overlaps(json_extract(test.t.j0, \"$.path0\"), cast(\"1\", json BINARY))",
"└─IndexMerge 0.00 root type: union",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
" └─Selection(Probe) 0.00 cop[tikv] lt(test.t.a, 10)",
" └─TableRowIDScan 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
},
{
"SQL": "select /*+ use_index_merge(t, j0_0) */ * from t where json_overlaps('1', (j0->'$.path0')) and a<10",
"Plan": [
"Selection 0.00 root json_overlaps(cast(\"1\", json BINARY), json_extract(test.t.j0, \"$.path0\"))",
"└─IndexMerge 0.00 root type: union",
" ├─IndexRangeScan(Build) 10.00 cop[tikv] table:t, index:j0_0(cast(json_extract(`j0`, _utf8mb4'$.path0') as signed array)) range:[1,1], keep order:false, stats:pseudo",
" └─Selection(Probe) 0.00 cop[tikv] lt(test.t.a, 10)",
" └─TableRowIDScan 0.00 cop[tikv] table:t keep order:false, stats:pseudo"
]
}
]
},
Expand Down