featurebase/sql3/planner/compileinsert.go
Travis Turner bcb32addaf Batch insert via SQL (multiple tuples) (#2243)
* Formatting adjustments made during code review.

While reviewing the BULK INSERT logic (in order to decide how best to
approach "ingest via sql" in the cloud), I made a few formatting and
comment changes. I'm just adding them here as a separate commit so they
don't muddy up my actual work.

* Parser modifications to support mulitple tuples in INSERT INTO

This commit doesn't include all of the changes required in the
planner. Fow now, the planner is simply modified to continue supporting
a single tuple (the first tuple in the list).

* Update the planner to handle multiple INSERT INTO tuples

This is part 1. It's still using the existing logic which builds an
ImportRequest for every record (and every field!).

The next step will involve using a client.Batch to handle the records.

* Introduce client.Importer interface (used by client.Batch)

Instead of the Batch having a pointer to a client, this puts an
interface there instead (which the client implements). It also allows us
to inject a different importer (i.e. other than a featurebase.client)
into the Batch.

* Decouple batch from client

This commit pulls batch-specific code out of the client package and into
a new batch package. It introduces the batch.Importer interface, the
methods of which replace all the calls that batch was previously making
directly to client methods.

Finally, it contains two implementations of the batch.Importer
interface: one is a wrapper around client, and the other is a wrapper
around featurebase.API.

* Use docker (instead of MustRunCluster) for internal batch tests

Because the `batch` package tests are internal, using
test.MustRunCluster() resulted in an import loop (because it eventually
imports `server`, and we can't have that). So this commit replaces the
use of `test.MustRunCluster()` with docker. The setup is basically the
same as that used in the idk docker tests.

Here we also remove all client-side references to `UseIngestAPI`, which
is an experimental (json) ingest api. It's still suppored on the server,
but here we remove the external usage of it.

* cherry-pick fix

* Use batch.Import() for sql3 INSERT INTO statements

* Thread logger into sql3

* fix batch test

* Fix some shadowing complaint by linter

* Address some test issues related to stringsets

* Exclude batch integration tests from CI

* Address PR feedback

- Added description to batch.README
- Consolidated grep commands in .gitlab-ci.yml
- Removed some debugging comments
- Replaces some inadvertantly removed license headers

* Add batch package to gitlab CI

* Updated CI for batch package

Updated CI include path

Update gitlab ci

Update CI

Update CI

Trying new include path for ci

Updated gitlab ci include path

Made idk race job optional for sonarcloud upload

add testdata directory

remove testenv from dockercompose file

use GIT_STRATEGY clone in batch CI

add testdata volume to dockercompose

Co-authored-by: Fletcher Haynes <fletcher.haynes@generalassemb.ly>
(cherry picked from commit 00ef2380e5)
2022-11-15 11:25:36 -08:00

187 lines
5.8 KiB
Go

// Copyright 2022 Molecula Corp. All rights reserved.
package planner
import (
"context"
"strings"
pilosa "github.com/featurebasedb/featurebase/v3"
"github.com/featurebasedb/featurebase/v3/sql3"
"github.com/featurebasedb/featurebase/v3/sql3/parser"
"github.com/featurebasedb/featurebase/v3/sql3/planner/types"
"github.com/pkg/errors"
)
// compileInsertStatement compiles an INSERT statement into a PlanOperator.
func (p *ExecutionPlanner) compileInsertStatement(stmt *parser.InsertStatement) (_ types.PlanOperator, err error) {
tableName := parser.IdentName(stmt.Table)
targetColumns := []*qualifiedRefPlanExpression{}
insertValues := [][]types.PlanExpression{}
table, err := p.schemaAPI.IndexInfo(context.Background(), tableName)
if err != nil {
if errors.Is(err, pilosa.ErrIndexNotFound) {
return nil, sql3.NewErrTableNotFound(stmt.Table.NamePos.Line, stmt.Table.NamePos.Column, tableName)
}
return nil, err
}
if len(stmt.Columns) > 0 {
for _, columnIdent := range stmt.Columns {
colName := parser.IdentName(columnIdent)
if strings.EqualFold(colName, "_id") {
targetColumns = append(targetColumns, newQualifiedRefPlanExpression(tableName, colName, 0, parser.NewDataTypeID()))
continue
}
for idx, field := range table.Fields {
if strings.EqualFold(colName, field.Name) {
targetColumns = append(targetColumns, newQualifiedRefPlanExpression(tableName, colName, idx, fieldSQLDataType(field)))
break
}
}
}
} else {
for idx, field := range table.Fields {
if strings.EqualFold("_exists", field.Name) {
continue
}
targetColumns = append(targetColumns, newQualifiedRefPlanExpression(tableName, field.Name, idx, fieldSQLDataType(field)))
}
}
//add expressions from values list
for _, tuple := range stmt.TupleList {
tupleValues := []types.PlanExpression{}
for _, expr := range tuple.Exprs {
e, err := p.compileExpr(expr)
if err != nil {
return nil, err
}
tupleValues = append(tupleValues, e)
}
insertValues = append(insertValues, tupleValues)
}
return NewPlanOpInsert(p, tableName, targetColumns, insertValues), nil
}
// analyzeInsertStatement analyzes an INSERT statement and returns and error if
// anything is invalid.
func (p *ExecutionPlanner) analyzeInsertStatement(stmt *parser.InsertStatement) error {
// Check that referred table exists.
tableName := parser.IdentName(stmt.Table)
table, err := p.schemaAPI.IndexInfo(context.Background(), tableName)
if err != nil {
if errors.Is(err, pilosa.ErrIndexNotFound) {
return sql3.NewErrTableNotFound(stmt.Table.NamePos.Line, stmt.Table.NamePos.Column, tableName)
}
return err
}
typeNames := make([]parser.ExprDataType, 0)
// If the insert statement does not provide the list of columns in which to
// insert the values, then the assumption is that the values apply to ALL
// fields in the table.
if len(stmt.Columns) == 0 {
// Generate the list of types from the FeatureBase index.
for _, field := range table.Fields {
if strings.EqualFold("_exists", field.Name) {
continue
}
typeNames = append(typeNames, fieldSQLDataType(field))
}
// Make sure (implicit) insert list and expression list have the same
// number of items.
for _, tuple := range stmt.TupleList {
if len(typeNames) != len(tuple.Exprs) {
return sql3.NewErrInsertExprTargetCountMismatch(tuple.Lparen.Line, tuple.Lparen.Column)
}
}
} else {
// Check column list refers to actual columns, and that there are no
// dupes.
columnNameMap := make(map[string]struct{})
for _, columnIdent := range stmt.Columns {
colName := parser.IdentName(columnIdent)
var typeName parser.ExprDataType
if strings.EqualFold(colName, "_id") {
columnNameMap["_id"] = struct{}{}
// Determine, from the existing table, whether the _id is of
// type ID or STRING.
var idType parser.ExprDataType
if table.Options.Keys {
idType = parser.NewDataTypeString()
} else {
idType = parser.NewDataTypeID()
}
typeNames = append(typeNames, idType)
continue
}
// Find the column in the existing table.
columnFound := false
for _, field := range table.Fields {
if strings.EqualFold(colName, field.Name) {
typeName = fieldSQLDataType(field)
columnFound = true
break
}
}
if !columnFound {
return sql3.NewErrColumnNotFound(columnIdent.NamePos.Line, columnIdent.NamePos.Column, colName)
}
// Ensure the column name hasn't already appeared in the list of
// columns.
if _, found := columnNameMap[colName]; found {
return sql3.NewErrDuplicateColumn(columnIdent.NamePos.Line, columnIdent.NamePos.Column, colName)
}
typeNames = append(typeNames, typeName)
columnNameMap[colName] = struct{}{}
}
// Ensure we have an _id column.
if _, ok := columnNameMap["_id"]; !ok {
return sql3.NewErrInsertMustHaveIDColumn(stmt.ColumnsLparen.Line, stmt.ColumnsLparen.Column)
}
// Ensure we have at least one more than just the _id column.
if len(stmt.Columns) < 2 {
return sql3.NewErrInsertMustAtLeastOneNonIDColumn(stmt.ColumnsLparen.Line, stmt.ColumnsLparen.Column)
}
// Make sure insert list and expression list have the same number of items.
for _, tuple := range stmt.TupleList {
if len(stmt.Columns) != len(tuple.Exprs) {
return sql3.NewErrInsertExprTargetCountMismatch(tuple.Lparen.Line, tuple.Lparen.Column)
}
}
}
// Check each of the expressions.
for _, tuple := range stmt.TupleList {
for i, expr := range tuple.Exprs {
e, err := p.analyzeExpression(expr, stmt)
if err != nil {
return err
}
// Type check against same ordinal position in column type list.
if !typesAreAssignmentCompatible(typeNames[i], e.DataType()) {
return sql3.NewErrTypeAssignmentIncompatible(expr.Pos().Line, expr.Pos().Column, e.DataType().TypeName(), typeNames[i].TypeName())
}
tuple.Exprs[i] = e
}
}
return nil
}