sqlglot.lineage
1from __future__ import annotations 2 3import json 4import typing as t 5from dataclasses import dataclass, field 6 7from sqlglot import Schema, exp, maybe_parse 8from sqlglot.errors import SqlglotError 9from sqlglot.optimizer import Scope, build_scope, qualify 10 11if t.TYPE_CHECKING: 12 from sqlglot.dialects.dialect import DialectType 13 14 15@dataclass(frozen=True) 16class Node: 17 name: str 18 expression: exp.Expression 19 source: exp.Expression 20 downstream: t.List[Node] = field(default_factory=list) 21 alias: str = "" 22 23 def walk(self) -> t.Iterator[Node]: 24 yield self 25 26 for d in self.downstream: 27 if isinstance(d, Node): 28 yield from d.walk() 29 else: 30 yield d 31 32 def to_html(self, **opts) -> LineageHTML: 33 return LineageHTML(self, **opts) 34 35 36def lineage( 37 column: str | exp.Column, 38 sql: str | exp.Expression, 39 schema: t.Optional[t.Dict | Schema] = None, 40 sources: t.Optional[t.Dict[str, str | exp.Subqueryable]] = None, 41 dialect: DialectType = None, 42 **kwargs, 43) -> Node: 44 """Build the lineage graph for a column of a SQL query. 45 46 Args: 47 column: The column to build the lineage for. 48 sql: The SQL string or expression. 49 schema: The schema of tables. 50 sources: A mapping of queries which will be used to continue building lineage. 51 dialect: The dialect of input SQL. 52 **kwargs: Qualification optimizer kwargs. 53 54 Returns: 55 A lineage node. 56 """ 57 58 expression = maybe_parse(sql, dialect=dialect) 59 60 if sources: 61 expression = exp.expand( 62 expression, 63 { 64 k: t.cast(exp.Subqueryable, maybe_parse(v, dialect=dialect)) 65 for k, v in sources.items() 66 }, 67 ) 68 69 qualified = qualify.qualify( 70 expression, 71 dialect=dialect, 72 schema=schema, 73 **{"validate_qualify_columns": False, "identify": False, **kwargs}, # type: ignore 74 ) 75 76 scope = build_scope(qualified) 77 78 if not scope: 79 raise SqlglotError("Cannot build lineage, sql must be SELECT") 80 81 def to_node( 82 column: str | int, 83 scope: Scope, 84 scope_name: t.Optional[str] = None, 85 upstream: t.Optional[Node] = None, 86 alias: t.Optional[str] = None, 87 ) -> Node: 88 aliases = { 89 dt.alias: dt.comments[0].split()[1] 90 for dt in scope.derived_tables 91 if dt.comments and dt.comments[0].startswith("source: ") 92 } 93 94 # Find the specific select clause that is the source of the column we want. 95 # This can either be a specific, named select or a generic `*` clause. 96 select = ( 97 scope.expression.selects[column] 98 if isinstance(column, int) 99 else next( 100 (select for select in scope.expression.selects if select.alias_or_name == column), 101 exp.Star() if scope.expression.is_star else None, 102 ) 103 ) 104 105 if not select: 106 raise ValueError(f"Could not find {column} in {scope.expression}") 107 108 if isinstance(scope.expression, exp.Union): 109 upstream = upstream or Node(name="UNION", source=scope.expression, expression=select) 110 111 index = ( 112 column 113 if isinstance(column, int) 114 else next( 115 i 116 for i, select in enumerate(scope.expression.selects) 117 if select.alias_or_name == column 118 ) 119 ) 120 121 for s in scope.union_scopes: 122 to_node(index, scope=s, upstream=upstream) 123 124 return upstream 125 126 if isinstance(scope.expression, exp.Select): 127 # For better ergonomics in our node labels, replace the full select with 128 # a version that has only the column we care about. 129 # "x", SELECT x, y FROM foo 130 # => "x", SELECT x FROM foo 131 source = t.cast(exp.Expression, scope.expression.select(select, append=False)) 132 else: 133 source = scope.expression 134 135 # Create the node for this step in the lineage chain, and attach it to the previous one. 136 node = Node( 137 name=f"{scope_name}.{column}" if scope_name else str(column), 138 source=source, 139 expression=select, 140 alias=alias or "", 141 ) 142 if upstream: 143 upstream.downstream.append(node) 144 145 # Find all columns that went into creating this one to list their lineage nodes. 146 for c in set(select.find_all(exp.Column)): 147 table = c.table 148 source = scope.sources.get(table) 149 150 if isinstance(source, Scope): 151 # The table itself came from a more specific scope. Recurse into that one using the unaliased column name. 152 to_node( 153 c.name, scope=source, scope_name=table, upstream=node, alias=aliases.get(table) 154 ) 155 else: 156 # The source is not a scope - we've reached the end of the line. At this point, if a source is not found 157 # it means this column's lineage is unknown. This can happen if the definition of a source used in a query 158 # is not passed into the `sources` map. 159 source = source or exp.Placeholder() 160 node.downstream.append(Node(name=c.sql(), source=source, expression=source)) 161 162 return node 163 164 return to_node(column if isinstance(column, str) else column.name, scope) 165 166 167class LineageHTML: 168 """Node to HTML generator using vis.js. 169 170 https://visjs.github.io/vis-network/docs/network/ 171 """ 172 173 def __init__( 174 self, 175 node: Node, 176 dialect: DialectType = None, 177 imports: bool = True, 178 **opts: t.Any, 179 ): 180 self.node = node 181 self.imports = imports 182 183 self.options = { 184 "height": "500px", 185 "width": "100%", 186 "layout": { 187 "hierarchical": { 188 "enabled": True, 189 "nodeSpacing": 200, 190 "sortMethod": "directed", 191 }, 192 }, 193 "interaction": { 194 "dragNodes": False, 195 "selectable": False, 196 }, 197 "physics": { 198 "enabled": False, 199 }, 200 "edges": { 201 "arrows": "to", 202 }, 203 "nodes": { 204 "font": "20px monaco", 205 "shape": "box", 206 "widthConstraint": { 207 "maximum": 300, 208 }, 209 }, 210 **opts, 211 } 212 213 self.nodes = {} 214 self.edges = [] 215 216 for node in node.walk(): 217 if isinstance(node.expression, exp.Table): 218 label = f"FROM {node.expression.this}" 219 title = f"<pre>SELECT {node.name} FROM {node.expression.this}</pre>" 220 group = 1 221 else: 222 label = node.expression.sql(pretty=True, dialect=dialect) 223 source = node.source.transform( 224 lambda n: exp.Tag(this=n, prefix="<b>", postfix="</b>") 225 if n is node.expression 226 else n, 227 copy=False, 228 ).sql(pretty=True, dialect=dialect) 229 title = f"<pre>{source}</pre>" 230 group = 0 231 232 node_id = id(node) 233 234 self.nodes[node_id] = { 235 "id": node_id, 236 "label": label, 237 "title": title, 238 "group": group, 239 } 240 241 for d in node.downstream: 242 self.edges.append({"from": node_id, "to": id(d)}) 243 244 def __str__(self): 245 nodes = json.dumps(list(self.nodes.values())) 246 edges = json.dumps(self.edges) 247 options = json.dumps(self.options) 248 imports = ( 249 """<script type="text/javascript" src="https://unpkg.com/vis-data@latest/peer/umd/vis-data.min.js"></script> 250 <script type="text/javascript" src="https://unpkg.com/vis-network@latest/peer/umd/vis-network.min.js"></script> 251 <link rel="stylesheet" type="text/css" href="https://unpkg.com/vis-network/styles/vis-network.min.css" />""" 252 if self.imports 253 else "" 254 ) 255 256 return f"""<div> 257 <div id="sqlglot-lineage"></div> 258 {imports} 259 <script type="text/javascript"> 260 var nodes = new vis.DataSet({nodes}) 261 nodes.forEach(row => row["title"] = new DOMParser().parseFromString(row["title"], "text/html").body.childNodes[0]) 262 263 new vis.Network( 264 document.getElementById("sqlglot-lineage"), 265 {{ 266 nodes: nodes, 267 edges: new vis.DataSet({edges}) 268 }}, 269 {options}, 270 ) 271 </script> 272</div>""" 273 274 def _repr_html_(self) -> str: 275 return self.__str__()
@dataclass(frozen=True)
class
Node:
16@dataclass(frozen=True) 17class Node: 18 name: str 19 expression: exp.Expression 20 source: exp.Expression 21 downstream: t.List[Node] = field(default_factory=list) 22 alias: str = "" 23 24 def walk(self) -> t.Iterator[Node]: 25 yield self 26 27 for d in self.downstream: 28 if isinstance(d, Node): 29 yield from d.walk() 30 else: 31 yield d 32 33 def to_html(self, **opts) -> LineageHTML: 34 return LineageHTML(self, **opts)
Node( name: str, expression: sqlglot.expressions.Expression, source: sqlglot.expressions.Expression, downstream: List[sqlglot.lineage.Node] = <factory>, alias: str = '')
expression: sqlglot.expressions.Expression
source: sqlglot.expressions.Expression
downstream: List[sqlglot.lineage.Node]
def
lineage( column: str | sqlglot.expressions.Column, sql: str | sqlglot.expressions.Expression, schema: Union[Dict, sqlglot.schema.Schema, NoneType] = None, sources: Optional[Dict[str, str | sqlglot.expressions.Subqueryable]] = None, dialect: Union[str, sqlglot.dialects.dialect.Dialect, Type[sqlglot.dialects.dialect.Dialect], NoneType] = None, **kwargs) -> sqlglot.lineage.Node:
37def lineage( 38 column: str | exp.Column, 39 sql: str | exp.Expression, 40 schema: t.Optional[t.Dict | Schema] = None, 41 sources: t.Optional[t.Dict[str, str | exp.Subqueryable]] = None, 42 dialect: DialectType = None, 43 **kwargs, 44) -> Node: 45 """Build the lineage graph for a column of a SQL query. 46 47 Args: 48 column: The column to build the lineage for. 49 sql: The SQL string or expression. 50 schema: The schema of tables. 51 sources: A mapping of queries which will be used to continue building lineage. 52 dialect: The dialect of input SQL. 53 **kwargs: Qualification optimizer kwargs. 54 55 Returns: 56 A lineage node. 57 """ 58 59 expression = maybe_parse(sql, dialect=dialect) 60 61 if sources: 62 expression = exp.expand( 63 expression, 64 { 65 k: t.cast(exp.Subqueryable, maybe_parse(v, dialect=dialect)) 66 for k, v in sources.items() 67 }, 68 ) 69 70 qualified = qualify.qualify( 71 expression, 72 dialect=dialect, 73 schema=schema, 74 **{"validate_qualify_columns": False, "identify": False, **kwargs}, # type: ignore 75 ) 76 77 scope = build_scope(qualified) 78 79 if not scope: 80 raise SqlglotError("Cannot build lineage, sql must be SELECT") 81 82 def to_node( 83 column: str | int, 84 scope: Scope, 85 scope_name: t.Optional[str] = None, 86 upstream: t.Optional[Node] = None, 87 alias: t.Optional[str] = None, 88 ) -> Node: 89 aliases = { 90 dt.alias: dt.comments[0].split()[1] 91 for dt in scope.derived_tables 92 if dt.comments and dt.comments[0].startswith("source: ") 93 } 94 95 # Find the specific select clause that is the source of the column we want. 96 # This can either be a specific, named select or a generic `*` clause. 97 select = ( 98 scope.expression.selects[column] 99 if isinstance(column, int) 100 else next( 101 (select for select in scope.expression.selects if select.alias_or_name == column), 102 exp.Star() if scope.expression.is_star else None, 103 ) 104 ) 105 106 if not select: 107 raise ValueError(f"Could not find {column} in {scope.expression}") 108 109 if isinstance(scope.expression, exp.Union): 110 upstream = upstream or Node(name="UNION", source=scope.expression, expression=select) 111 112 index = ( 113 column 114 if isinstance(column, int) 115 else next( 116 i 117 for i, select in enumerate(scope.expression.selects) 118 if select.alias_or_name == column 119 ) 120 ) 121 122 for s in scope.union_scopes: 123 to_node(index, scope=s, upstream=upstream) 124 125 return upstream 126 127 if isinstance(scope.expression, exp.Select): 128 # For better ergonomics in our node labels, replace the full select with 129 # a version that has only the column we care about. 130 # "x", SELECT x, y FROM foo 131 # => "x", SELECT x FROM foo 132 source = t.cast(exp.Expression, scope.expression.select(select, append=False)) 133 else: 134 source = scope.expression 135 136 # Create the node for this step in the lineage chain, and attach it to the previous one. 137 node = Node( 138 name=f"{scope_name}.{column}" if scope_name else str(column), 139 source=source, 140 expression=select, 141 alias=alias or "", 142 ) 143 if upstream: 144 upstream.downstream.append(node) 145 146 # Find all columns that went into creating this one to list their lineage nodes. 147 for c in set(select.find_all(exp.Column)): 148 table = c.table 149 source = scope.sources.get(table) 150 151 if isinstance(source, Scope): 152 # The table itself came from a more specific scope. Recurse into that one using the unaliased column name. 153 to_node( 154 c.name, scope=source, scope_name=table, upstream=node, alias=aliases.get(table) 155 ) 156 else: 157 # The source is not a scope - we've reached the end of the line. At this point, if a source is not found 158 # it means this column's lineage is unknown. This can happen if the definition of a source used in a query 159 # is not passed into the `sources` map. 160 source = source or exp.Placeholder() 161 node.downstream.append(Node(name=c.sql(), source=source, expression=source)) 162 163 return node 164 165 return to_node(column if isinstance(column, str) else column.name, scope)
Build the lineage graph for a column of a SQL query.
Arguments:
- column: The column to build the lineage for.
- sql: The SQL string or expression.
- schema: The schema of tables.
- sources: A mapping of queries which will be used to continue building lineage.
- dialect: The dialect of input SQL.
- **kwargs: Qualification optimizer kwargs.
Returns:
A lineage node.
class
LineageHTML:
168class LineageHTML: 169 """Node to HTML generator using vis.js. 170 171 https://visjs.github.io/vis-network/docs/network/ 172 """ 173 174 def __init__( 175 self, 176 node: Node, 177 dialect: DialectType = None, 178 imports: bool = True, 179 **opts: t.Any, 180 ): 181 self.node = node 182 self.imports = imports 183 184 self.options = { 185 "height": "500px", 186 "width": "100%", 187 "layout": { 188 "hierarchical": { 189 "enabled": True, 190 "nodeSpacing": 200, 191 "sortMethod": "directed", 192 }, 193 }, 194 "interaction": { 195 "dragNodes": False, 196 "selectable": False, 197 }, 198 "physics": { 199 "enabled": False, 200 }, 201 "edges": { 202 "arrows": "to", 203 }, 204 "nodes": { 205 "font": "20px monaco", 206 "shape": "box", 207 "widthConstraint": { 208 "maximum": 300, 209 }, 210 }, 211 **opts, 212 } 213 214 self.nodes = {} 215 self.edges = [] 216 217 for node in node.walk(): 218 if isinstance(node.expression, exp.Table): 219 label = f"FROM {node.expression.this}" 220 title = f"<pre>SELECT {node.name} FROM {node.expression.this}</pre>" 221 group = 1 222 else: 223 label = node.expression.sql(pretty=True, dialect=dialect) 224 source = node.source.transform( 225 lambda n: exp.Tag(this=n, prefix="<b>", postfix="</b>") 226 if n is node.expression 227 else n, 228 copy=False, 229 ).sql(pretty=True, dialect=dialect) 230 title = f"<pre>{source}</pre>" 231 group = 0 232 233 node_id = id(node) 234 235 self.nodes[node_id] = { 236 "id": node_id, 237 "label": label, 238 "title": title, 239 "group": group, 240 } 241 242 for d in node.downstream: 243 self.edges.append({"from": node_id, "to": id(d)}) 244 245 def __str__(self): 246 nodes = json.dumps(list(self.nodes.values())) 247 edges = json.dumps(self.edges) 248 options = json.dumps(self.options) 249 imports = ( 250 """<script type="text/javascript" src="https://unpkg.com/vis-data@latest/peer/umd/vis-data.min.js"></script> 251 <script type="text/javascript" src="https://unpkg.com/vis-network@latest/peer/umd/vis-network.min.js"></script> 252 <link rel="stylesheet" type="text/css" href="https://unpkg.com/vis-network/styles/vis-network.min.css" />""" 253 if self.imports 254 else "" 255 ) 256 257 return f"""<div> 258 <div id="sqlglot-lineage"></div> 259 {imports} 260 <script type="text/javascript"> 261 var nodes = new vis.DataSet({nodes}) 262 nodes.forEach(row => row["title"] = new DOMParser().parseFromString(row["title"], "text/html").body.childNodes[0]) 263 264 new vis.Network( 265 document.getElementById("sqlglot-lineage"), 266 {{ 267 nodes: nodes, 268 edges: new vis.DataSet({edges}) 269 }}, 270 {options}, 271 ) 272 </script> 273</div>""" 274 275 def _repr_html_(self) -> str: 276 return self.__str__()
Node to HTML generator using vis.js.
LineageHTML( node: sqlglot.lineage.Node, dialect: Union[str, sqlglot.dialects.dialect.Dialect, Type[sqlglot.dialects.dialect.Dialect], NoneType] = None, imports: bool = True, **opts: Any)
174 def __init__( 175 self, 176 node: Node, 177 dialect: DialectType = None, 178 imports: bool = True, 179 **opts: t.Any, 180 ): 181 self.node = node 182 self.imports = imports 183 184 self.options = { 185 "height": "500px", 186 "width": "100%", 187 "layout": { 188 "hierarchical": { 189 "enabled": True, 190 "nodeSpacing": 200, 191 "sortMethod": "directed", 192 }, 193 }, 194 "interaction": { 195 "dragNodes": False, 196 "selectable": False, 197 }, 198 "physics": { 199 "enabled": False, 200 }, 201 "edges": { 202 "arrows": "to", 203 }, 204 "nodes": { 205 "font": "20px monaco", 206 "shape": "box", 207 "widthConstraint": { 208 "maximum": 300, 209 }, 210 }, 211 **opts, 212 } 213 214 self.nodes = {} 215 self.edges = [] 216 217 for node in node.walk(): 218 if isinstance(node.expression, exp.Table): 219 label = f"FROM {node.expression.this}" 220 title = f"<pre>SELECT {node.name} FROM {node.expression.this}</pre>" 221 group = 1 222 else: 223 label = node.expression.sql(pretty=True, dialect=dialect) 224 source = node.source.transform( 225 lambda n: exp.Tag(this=n, prefix="<b>", postfix="</b>") 226 if n is node.expression 227 else n, 228 copy=False, 229 ).sql(pretty=True, dialect=dialect) 230 title = f"<pre>{source}</pre>" 231 group = 0 232 233 node_id = id(node) 234 235 self.nodes[node_id] = { 236 "id": node_id, 237 "label": label, 238 "title": title, 239 "group": group, 240 } 241 242 for d in node.downstream: 243 self.edges.append({"from": node_id, "to": id(d)})