Skip to content

[CBO] Make hints process many trees. #8554

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

Merged
pashandor789 merged 2 commits into from
Sep 2, 2024
Merged

[CBO] Make hints process many trees. #8554

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
665a9e4
[CBO] Make hints process many trees.
pashandor789 Aug 30, 2024
daf07df
[CBO] some fixies.
pashandor789 Sep 1, 2024
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
6 changes: 5 additions & 1 deletion ydb/core/kqp/ut/join/data/queries/join_order_hints_complex.sql
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,10 @@
PRAGMA TablePathPrefix='/Root';

PRAGMA ydb.OptJoinOrderHints='[ ["R", "S"], ["T", "U"] ]';
PRAGMA ydb.OptJoinOrderHints=
'[
[["R", "S"], ["T", "U"]]
]';

PRAGMA ydb.OptCardinalityHints =
'[
{"labels":["R"], "op":"#", "value":10e8},
Expand Down
22 changes: 22 additions & 0 deletions ydb/core/kqp/ut/join/data/queries/join_order_hints_many_hint_trees.sql
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,22 @@
PRAGMA TablePathPrefix='/Root';

PRAGMA ydb.OptJoinOrderHints=
'[
["R", "S"],
["T", "U"]
]';
PRAGMA ydb.OptCardinalityHints =
'[
{"labels":["R"], "op":"#", "value":10e8},
{"labels":["T"], "op":"#", "value":1},
{"labels":["R", "T"], "op":"#", "value":1},
{"labels":["R", "S"], "op":"#", "value":10e8},
{"labels":["T", "U"], "op":"#", "value":10e8},
{"labels":["V"], "op":"#", "value":1}
]';

SELECT * FROM
R INNER JOIN S on R.id = S.id
INNER JOIN T on R.id = T.id
INNER JOIN U on T.id = U.id
INNER JOIN V on U.id = V.id;
5 changes: 4 additions & 1 deletion ydb/core/kqp/ut/join/data/queries/join_order_hints_simple.sql
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,10 @@ PRAGMA ydb.OptCardinalityHints =
{"labels":["R", "T"], "op":"#", "value":1},
{"labels":["R", "S"], "op":"#", "value":10e8}
]';
PRAGMA ydb.OptJoinOrderHints='[ "T", ["R", "S"] ]';
PRAGMA ydb.OptJoinOrderHints=
'[
["T", ["R", "S"]]
]';

SELECT * FROM
R INNER JOIN S on R.id = S.id
Expand Down
33 changes: 33 additions & 0 deletions ydb/core/kqp/ut/join/kqp_join_order_ut.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -373,6 +373,39 @@ Y_UNIT_TEST_SUITE(KqpJoinOrder) {
UNIT_ASSERT_VALUES_EQUAL(GetJoinOrder(plan).GetStringRobust(), R"([[["R","S"],["T","U"]],"V"])") ;
}

void TraverseJoinTree(
const NJson::TJsonValue& tree,
std::function<void(const NJson::TJsonValue&, const NJson::TJsonValue&)> joinVisitFunc
) {
if (tree.IsArray()) {
auto join = tree.GetArray();
Y_ENSURE(join.size() == 2, Sprintf("Incorrect join tree piece, expected 2 children for join, got : %s", tree.GetStringRobust().c_str()));
joinVisitFunc(join[0], join[1]);
TraverseJoinTree(join[0], joinVisitFunc);
TraverseJoinTree(join[1], joinVisitFunc);
} else if (tree.IsString()) {
} else {
Y_ENSURE(false, Sprintf("Incorrect join tree piece: %s", tree.GetStringRobust().c_str()));
}
}

Y_UNIT_TEST_XOR_OR_BOTH_FALSE(TestJoinOrderHintsManyHintTrees, StreamLookupJoin, ColumnStore) {
auto plan = ExecuteJoinOrderTestDataQueryWithStats("queries/join_order_hints_many_hint_trees.sql", "stats/basic.json", StreamLookupJoin, ColumnStore);
auto joinOrder = GetJoinOrder(plan);

THashSet<TString> expectedJoins = { "R, S", "T, U" };
TraverseJoinTree(
joinOrder,
[&expectedJoins](const NJson::TJsonValue& joinLhs, const NJson::TJsonValue& joinRhs){
if (joinLhs.IsString() && joinRhs.IsString()) {
expectedJoins.erase(Sprintf("%s, %s", joinLhs.GetString().c_str(), joinRhs.GetString().c_str()));
}
}
);

UNIT_ASSERT_C(expectedJoins.empty(), Sprintf("%s wasn't found in '%s' join order!", JoinSeq(", ", expectedJoins).c_str(), joinOrder.GetStringRobust().c_str()));
}

void JoinOrderTestWithOverridenStats(const TString& queryPath, const TString& statsPath, TString correctJoinOrderPath, bool useStreamLookupJoin, bool useColumnStore
) {
auto kikimr = GetKikimrWithJoinSettings(useStreamLookupJoin, GetStatic(statsPath));
Expand Down
9 changes: 8 additions & 1 deletion ydb/library/yql/core/cbo/cbo_optimizer_new.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -330,9 +330,16 @@ std::shared_ptr<IBaseOptimizerNode> MakeJoinTreeFromJson(const NJson::TJsonValue
}

TJoinOrderHints::TJoinOrderHints(const TString& json) {
const static TString PARSING_FORMAT_ERROR =
R"(Join order hints parsing failed. The example of the format: [ ["A", "B"], ["B", "C"] ])";

NJson::TJsonValue jsonTree;
NJson::ReadJsonTree(json, &jsonTree, true);
HintsTree = MakeJoinTreeFromJson(jsonTree);

Y_ENSURE(jsonTree.IsArray(), PARSING_FORMAT_ERROR);
for (const auto& hintTreeJson: jsonTree.GetArray()) {
HintTrees.push_back(MakeJoinTreeFromJson(hintTreeJson));
}
}

TJoinAlgoHints::TJoinAlgoHints(const TString& json) {
Expand Down
2 changes: 1 addition & 1 deletion ydb/library/yql/core/cbo/cbo_optimizer_new.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -109,7 +109,7 @@ struct TJoinAlgoHints {
};

struct TJoinOrderHints {
std::shared_ptr<IBaseOptimizerNode> HintsTree;
TVector<std::shared_ptr<IBaseOptimizerNode>> HintTrees;

TJoinOrderHints() {}
TJoinOrderHints(const TString& json);
Expand Down
37 changes: 17 additions & 20 deletions ydb/library/yql/dq/opt/dq_opt_join_hypergraph.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@


#include <numeric>
#include <util/string/join.h>
#include <util/string/printf.h>
#include "bitset.h"

Expand Down Expand Up @@ -272,21 +273,23 @@ class TJoinOrderHintsApplier {
{}

void Apply(const TJoinOrderHints& hints) {
auto labels = ApplyHintsToSubgraph(hints.HintsTree);
auto nodes = Graph_.GetNodesByRelNames(labels);

for (size_t i = 0; i < Graph_.GetEdges().size(); ++i) {
TNodeSet newLeft = Graph_.GetEdge(i).Left;
if (Overlaps(Graph_.GetEdge(i).Left, nodes) && !IsSubset(Graph_.GetEdge(i).Right, nodes)) {
newLeft |= nodes;
}
for (const auto& hintTree: hints.HintTrees) {
auto labels = ApplyHintsToSubgraph(hintTree);
auto nodes = Graph_.GetNodesByRelNames(labels);

for (size_t i = 0; i < Graph_.GetEdges().size(); ++i) {
TNodeSet newLeft = Graph_.GetEdge(i).Left;
if (Overlaps(Graph_.GetEdge(i).Left, nodes) && !IsSubset(Graph_.GetEdge(i).Right, nodes)) {
newLeft |= nodes;
}

TNodeSet newRight = Graph_.GetEdge(i).Right;
if (Overlaps(Graph_.GetEdge(i).Right, nodes) && !IsSubset(Graph_.GetEdge(i).Left, nodes)) {
newRight |= nodes;
}
TNodeSet newRight = Graph_.GetEdge(i).Right;
if (Overlaps(Graph_.GetEdge(i).Right, nodes) && !IsSubset(Graph_.GetEdge(i).Left, nodes)) {
newRight |= nodes;
}

Graph_.UpdateEdgeSides(i, newLeft, newRight);
Graph_.UpdateEdgeSides(i, newLeft, newRight);
}
}
}

Expand All @@ -302,14 +305,8 @@ class TJoinOrderHintsApplier {

auto* maybeEdge = Graph_.FindEdgeBetween(lhs, rhs);
if (maybeEdge == nullptr) {
auto str = [](const TVector<TString>& v) -> TString {
TString s;
for (auto& el : v) { s += (el + ", "); }
return s.empty()? s: s.substr(0, s.length() - 2);
};

const char* errStr = "There is no edge between {%s}, {%s}. The graf: %s";
Y_ENSURE(false, Sprintf(errStr, str(lhsLabels).c_str(), str(rhsLabels).c_str(), Graph_.String().c_str()));
Y_ENSURE(false, Sprintf(errStr, JoinSeq(", ", lhsLabels).c_str(), JoinSeq(", ", rhsLabels).c_str(), Graph_.String().c_str()));
}

size_t revEdgeIdx = maybeEdge->ReversedEdgeId;
Expand Down
2 changes: 1 addition & 1 deletion ydb/library/yql/dq/opt/dq_opt_make_join_hypergraph.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -100,7 +100,7 @@ TJoinHypergraph<TNodeSet> MakeJoinHypergraph(
YQL_CLOG(TRACE, CoreDq) << graph.String();
}

if (hints.HintsTree != nullptr) {
if (!hints.HintTrees.empty()) {
TJoinOrderHintsApplier joinHints(graph);
joinHints.Apply(hints);
if (NYql::NLog::YqlLogger().NeedToLog(NYql::NLog::EComponent::CoreDq, NYql::NLog::ELevel::TRACE)) {
Expand Down