import { Knex, knex } from "knex" import * as dbCore from "../db" import { getNativeSql, isExternalTable, isInvalidISODateString, isValidFilter, isValidISODateString, sqlLog, validateManyToMany, } from "./utils" import SqlTableQueryBuilder from "./sqlTable" import { Aggregation, AnySearchFilter, ArrayOperator, BasicOperator, BBReferenceFieldMetadata, CalculationType, FieldSchema, FieldType, INTERNAL_TABLE_SOURCE_ID, InternalSearchFilterOperator, JsonFieldMetadata, JsonTypes, Operation, prefixed, QueryJson, QueryOptions, RangeOperator, RelationshipsJson, SearchFilters, SortOrder, SqlClient, SqlQuery, SqlQueryBinding, Table, TableSourceType, } from "@budibase/types" import environment from "../environment" import { dataFilters, helpers } from "@budibase/shared-core" import { cloneDeep } from "lodash" type QueryFunction = (query: SqlQuery | SqlQuery[], operation: Operation) => any export const COUNT_FIELD_NAME = "__bb_total" function getBaseLimit() { const envLimit = environment.SQL_MAX_ROWS ? parseInt(environment.SQL_MAX_ROWS) : null return envLimit || 5000 } function getRelationshipLimit() { const envLimit = environment.SQL_MAX_RELATED_ROWS ? parseInt(environment.SQL_MAX_RELATED_ROWS) : null return envLimit || 500 } function prioritisedArraySort(toSort: string[], priorities: string[]) { return toSort.sort((a, b) => { const aPriority = priorities.find(field => field && a.endsWith(field)) const bPriority = priorities.find(field => field && b.endsWith(field)) if (aPriority && !bPriority) { return -1 } if (!aPriority && bPriority) { return 1 } return a.localeCompare(b) }) } function convertBooleans(query: SqlQuery | SqlQuery[]): SqlQuery | SqlQuery[] { if (Array.isArray(query)) { return query.map((q: SqlQuery) => convertBooleans(q) as SqlQuery) } else { if (query.bindings) { query.bindings = query.bindings.map(binding => { if (typeof binding === "boolean") { return binding ? 1 : 0 } return binding }) } } return query } function isSqs(table: Table): boolean { return ( table.sourceType === TableSourceType.INTERNAL || table.sourceId === INTERNAL_TABLE_SOURCE_ID ) } class InternalBuilder { private readonly client: SqlClient private readonly query: QueryJson private readonly splitter: dataFilters.ColumnSplitter private readonly knex: Knex constructor(client: SqlClient, knex: Knex, query: QueryJson) { this.client = client this.query = query this.knex = knex this.splitter = new dataFilters.ColumnSplitter([this.table], { aliases: this.query.tableAliases, columnPrefix: this.query.meta.columnPrefix, }) } // states the various situations in which we need a full mapped select statement private readonly SPECIAL_SELECT_CASES = { POSTGRES_MONEY: (field: FieldSchema | undefined) => { return ( this.client === SqlClient.POSTGRES && field?.externalType?.includes("money") ) }, MSSQL_DATES: (field: FieldSchema | undefined) => { return ( this.client === SqlClient.MS_SQL && field?.type === FieldType.DATETIME && field.timeOnly ) }, } get table(): Table { return this.query.meta.table } getFieldSchema(key: string): FieldSchema | undefined { const { column } = this.splitter.run(key) return this.table.schema[column] } private quoteChars(): [string, string] { switch (this.client) { case SqlClient.ORACLE: case SqlClient.POSTGRES: return ['"', '"'] case SqlClient.MS_SQL: return ["[", "]"] case SqlClient.MARIADB: case SqlClient.MY_SQL: case SqlClient.SQL_LITE: return ["`", "`"] } } // Takes a string like foo and returns a quoted string like [foo] for SQL Server // and "foo" for Postgres. private quote(str: string): string { const [start, end] = this.quoteChars() return `${start}${str}${end}` } private isQuoted(key: string): boolean { const [start, end] = this.quoteChars() return key.startsWith(start) && key.endsWith(end) } // Takes a string like a.b.c or an array like ["a", "b", "c"] and returns a // quoted identifier like [a].[b].[c] for SQL Server and `a`.`b`.`c` for // MySQL. private quotedIdentifier(key: string | string[]): string { if (!Array.isArray(key)) { key = this.splitIdentifier(key) } return key.map(part => this.quote(part)).join(".") } // Turns an identifier like a.b.c or `a`.`b`.`c` into ["a", "b", "c"] private splitIdentifier(key: string): string[] { const [start, end] = this.quoteChars() if (this.isQuoted(key)) { return key.slice(1, -1).split(`${end}.${start}`) } return key.split(".") } private qualifyIdentifier(key: string): string { const tableName = this.getTableName() const parts = this.splitIdentifier(key) if (parts[0] !== tableName) { parts.unshift(tableName) } if (this.isQuoted(key)) { return this.quotedIdentifier(parts) } return parts.join(".") } private isFullSelectStatementRequired(): boolean { const { meta } = this.query for (let column of Object.values(meta.table.schema)) { if (this.SPECIAL_SELECT_CASES.POSTGRES_MONEY(column)) { return true } else if (this.SPECIAL_SELECT_CASES.MSSQL_DATES(column)) { return true } } return false } private generateSelectStatement(): (string | Knex.Raw)[] | "*" { const { meta, endpoint, resource } = this.query if (!resource || !resource.fields || resource.fields.length === 0) { return "*" } const alias = this.getTableName(endpoint.entityId) const schema = meta.table.schema if (!this.isFullSelectStatementRequired()) { return [this.knex.raw(`${this.quote(alias)}.*`)] } // get just the fields for this table return resource.fields .map(field => { const parts = field.split(/\./g) let table: string | undefined = undefined let column = parts[0] // Just a column name, e.g.: "column" if (parts.length > 1) { table = parts[0] column = parts.slice(1).join(".") } return { table, column, field } }) .filter(({ table }) => !table || table === alias) .map(({ table, column, field }) => { const columnSchema = schema[column] if (this.SPECIAL_SELECT_CASES.POSTGRES_MONEY(columnSchema)) { return this.knex.raw( `${this.quotedIdentifier( [table, column].join(".") )}::money::numeric as ${this.quote(field)}` ) } if (this.SPECIAL_SELECT_CASES.MSSQL_DATES(columnSchema)) { // Time gets returned as timestamp from mssql, not matching the expected // HH:mm format return this.knex.raw(`CONVERT(varchar, ${field}, 108) as "${field}"`) } const quoted = table ? `${this.quote(table)}.${this.quote(column)}` : this.quote(field) return this.knex.raw(quoted) }) } // OracleDB can't use character-large-objects (CLOBs) in WHERE clauses, // so when we use them we need to wrap them in to_char(). This function // converts a field name to the appropriate identifier. private convertClobs(field: string, opts?: { forSelect?: boolean }): string { if (this.client !== SqlClient.ORACLE) { throw new Error( "you've called convertClobs on a DB that's not Oracle, this is a mistake" ) } const parts = this.splitIdentifier(field) const col = parts.pop()! const schema = this.table.schema[col] let identifier = this.quotedIdentifier(field) if ( schema.type === FieldType.STRING || schema.type === FieldType.LONGFORM || schema.type === FieldType.BB_REFERENCE_SINGLE || schema.type === FieldType.BB_REFERENCE || schema.type === FieldType.OPTIONS || schema.type === FieldType.BARCODEQR ) { if (opts?.forSelect) { identifier = `to_char(${identifier}) as ${this.quotedIdentifier(col)}` } else { identifier = `to_char(${identifier})` } } return identifier } private parse(input: any, schema: FieldSchema) { if (Array.isArray(input)) { return JSON.stringify(input) } if (input == undefined) { return null } if ( this.client === SqlClient.ORACLE && schema.type === FieldType.DATETIME && schema.timeOnly ) { if (input instanceof Date) { const hours = input.getHours().toString().padStart(2, "0") const minutes = input.getMinutes().toString().padStart(2, "0") const seconds = input.getSeconds().toString().padStart(2, "0") return `${hours}:${minutes}:${seconds}` } if (typeof input === "string") { return new Date(`1970-01-01T${input}Z`) } } if (typeof input === "string") { if (isInvalidISODateString(input)) { return null } if (isValidISODateString(input)) { return new Date(input.trim()) } } return input } private parseBody(body: Record) { for (let [key, value] of Object.entries(body)) { const { column } = this.splitter.run(key) const schema = this.table.schema[column] if (!schema) { continue } body[key] = this.parse(value, schema) } return body } private parseFilters(filters: SearchFilters): SearchFilters { filters = cloneDeep(filters) for (const op of Object.values(BasicOperator)) { const filter = filters[op] if (!filter) { continue } for (const key of Object.keys(filter)) { if (Array.isArray(filter[key])) { filter[key] = JSON.stringify(filter[key]) continue } const { column } = this.splitter.run(key) const schema = this.table.schema[column] if (!schema) { continue } filter[key] = this.parse(filter[key], schema) } } for (const op of Object.values(ArrayOperator)) { const filter = filters[op] if (!filter) { continue } for (const key of Object.keys(filter)) { const { column } = this.splitter.run(key) const schema = this.table.schema[column] if (!schema) { continue } filter[key] = filter[key].map(v => this.parse(v, schema)) } } for (const op of Object.values(RangeOperator)) { const filter = filters[op] if (!filter) { continue } for (const key of Object.keys(filter)) { const { column } = this.splitter.run(key) const schema = this.table.schema[column] if (!schema) { continue } const value = filter[key] if ("low" in value) { value.low = this.parse(value.low, schema) } if ("high" in value) { value.high = this.parse(value.high, schema) } } } return filters } addJoinFieldCheck(query: Knex.QueryBuilder, relationship: RelationshipsJson) { const document = relationship.from?.split(".")[0] || "" return query.andWhere(`${document}.fieldName`, "=", relationship.column) } addRelationshipForFilter( query: Knex.QueryBuilder, filterKey: string, whereCb: (query: Knex.QueryBuilder) => Knex.QueryBuilder ): Knex.QueryBuilder { const mainKnex = this.knex const { relationships, endpoint, tableAliases: aliases } = this.query const tableName = endpoint.entityId const fromAlias = aliases?.[tableName] || tableName const matches = (possibleTable: string) => filterKey.startsWith(`${possibleTable}`) if (!relationships) { return query } for (const relationship of relationships) { const relatedTableName = relationship.tableName const toAlias = aliases?.[relatedTableName] || relatedTableName // this is the relationship which is being filtered if ( (matches(relatedTableName) || matches(toAlias)) && relationship.to && relationship.tableName ) { let subQuery = mainKnex .select(mainKnex.raw(1)) .from({ [toAlias]: relatedTableName }) const manyToMany = validateManyToMany(relationship) if (manyToMany) { const throughAlias = aliases?.[manyToMany.through] || relationship.through let throughTable = this.tableNameWithSchema(manyToMany.through, { alias: throughAlias, schema: endpoint.schema, }) subQuery = subQuery // add a join through the junction table .innerJoin(throughTable, function () { // @ts-ignore this.on( `${toAlias}.${manyToMany.toPrimary}`, "=", `${throughAlias}.${manyToMany.to}` ) }) // check the document in the junction table points to the main table .where( `${throughAlias}.${manyToMany.from}`, "=", mainKnex.raw( this.quotedIdentifier(`${fromAlias}.${manyToMany.fromPrimary}`) ) ) // in SQS the same junction table is used for different many-to-many relationships between the // two same tables, this is needed to avoid rows ending up in all columns if (this.client === SqlClient.SQL_LITE) { subQuery = this.addJoinFieldCheck(subQuery, manyToMany) } } else { // "join" to the main table, making sure the ID matches that of the main subQuery = subQuery.where( `${toAlias}.${relationship.to}`, "=", mainKnex.raw( this.quotedIdentifier(`${fromAlias}.${relationship.from}`) ) ) } query = query.whereExists(whereCb(subQuery)) break } } return query } // right now we only do filters on the specific table being queried addFilters( query: Knex.QueryBuilder, filters: SearchFilters | undefined, opts?: { relationship?: boolean } ): Knex.QueryBuilder { if (!filters) { return query } const builder = this filters = this.parseFilters({ ...filters }) const aliases = this.query.tableAliases // if all or specified in filters, then everything is an or const allOr = filters.allOr const isSqlite = this.client === SqlClient.SQL_LITE const tableName = isSqlite ? this.table._id! : this.table.name function getTableAlias(name: string) { const alias = aliases?.[name] return alias || name } function iterate( structure: AnySearchFilter, fn: ( query: Knex.QueryBuilder, key: string, value: any ) => Knex.QueryBuilder, complexKeyFn?: ( query: Knex.QueryBuilder, key: string[], value: any ) => Knex.QueryBuilder ) { const handleRelationship = ( q: Knex.QueryBuilder, key: string, value: any ) => { const [filterTableName, ...otherProperties] = key.split(".") const property = otherProperties.join(".") const alias = getTableAlias(filterTableName) return fn(q, alias ? `${alias}.${property}` : property, value) } for (const key in structure) { const value = structure[key] const updatedKey = dbCore.removeKeyNumbering(key) const isRelationshipField = updatedKey.includes(".") const shouldProcessRelationship = opts?.relationship && isRelationshipField let castedTypeValue if ( key === InternalSearchFilterOperator.COMPLEX_ID_OPERATOR && (castedTypeValue = structure[key]) && complexKeyFn ) { const alias = getTableAlias(tableName) query = complexKeyFn( query, castedTypeValue.id.map((x: string) => alias ? `${alias}.${x}` : x ), castedTypeValue.values ) } else if (!isRelationshipField) { const alias = getTableAlias(tableName) query = fn( query, alias ? `${alias}.${updatedKey}` : updatedKey, value ) } else if (shouldProcessRelationship) { query = builder.addRelationshipForFilter(query, updatedKey, q => { return handleRelationship(q, updatedKey, value) }) } } } const like = (q: Knex.QueryBuilder, key: string, value: any) => { const fuzzyOr = filters?.fuzzyOr const fnc = fuzzyOr || allOr ? "orWhere" : "where" // postgres supports ilike, nothing else does if (this.client === SqlClient.POSTGRES) { return q[fnc](key, "ilike", `%${value}%`) } else { const rawFnc = `${fnc}Raw` // @ts-ignore return q[rawFnc](`LOWER(${this.quotedIdentifier(key)}) LIKE ?`, [ `%${value.toLowerCase()}%`, ]) } } const contains = (mode: AnySearchFilter, any: boolean = false) => { const rawFnc = allOr ? "orWhereRaw" : "whereRaw" const not = mode === filters?.notContains ? "NOT " : "" function stringifyArray(value: Array, quoteStyle = '"'): string { for (let i in value) { if (typeof value[i] === "string") { value[i] = `${quoteStyle}${value[i]}${quoteStyle}` } } return `[${value.join(",")}]` } if (this.client === SqlClient.POSTGRES) { iterate(mode, (q, key, value) => { const wrap = any ? "" : "'" const op = any ? "\\?| array" : "@>" const fieldNames = key.split(/\./g) const table = fieldNames[0] const col = fieldNames[1] return q[rawFnc]( `${not}COALESCE("${table}"."${col}"::jsonb ${op} ${wrap}${stringifyArray( value, any ? "'" : '"' )}${wrap}, FALSE)` ) }) } else if ( this.client === SqlClient.MY_SQL || this.client === SqlClient.MARIADB ) { const jsonFnc = any ? "JSON_OVERLAPS" : "JSON_CONTAINS" iterate(mode, (q, key, value) => { return q[rawFnc]( `${not}COALESCE(${jsonFnc}(${key}, '${stringifyArray( value )}'), FALSE)` ) }) } else { const andOr = mode === filters?.containsAny ? " OR " : " AND " iterate(mode, (q, key, value) => { let statement = "" const identifier = this.quotedIdentifier(key) for (let i in value) { if (typeof value[i] === "string") { value[i] = `%"${value[i].toLowerCase()}"%` } else { value[i] = `%${value[i]}%` } statement += `${ statement ? andOr : "" }COALESCE(LOWER(${identifier}), '') LIKE ?` } if (statement === "") { return q } if (not) { return q[rawFnc]( `(NOT (${statement}) OR ${identifier} IS NULL)`, value ) } else { return q[rawFnc](statement, value) } }) } } if (filters.$and) { const { $and } = filters for (const condition of $and.conditions) { query = query.where(b => { this.addFilters(b, condition, opts) }) } } if (filters.$or) { const { $or } = filters query = query.where(b => { for (const condition of $or.conditions) { b.orWhere(c => this.addFilters(c, { ...condition, allOr: true }, opts) ) } }) } if (filters.oneOf) { const fnc = allOr ? "orWhereIn" : "whereIn" iterate( filters.oneOf, (q, key: string, array) => { if (this.client === SqlClient.ORACLE) { key = this.convertClobs(key) array = Array.isArray(array) ? array : [array] const binding = new Array(array.length).fill("?").join(",") return q.whereRaw(`${key} IN (${binding})`, array) } else { return q[fnc](key, Array.isArray(array) ? array : [array]) } }, (q, key: string[], array) => { if (this.client === SqlClient.ORACLE) { const keyStr = `(${key.map(k => this.convertClobs(k)).join(",")})` const binding = `(${array .map((a: any) => `(${new Array(a.length).fill("?").join(",")})`) .join(",")})` return q.whereRaw(`${keyStr} IN ${binding}`, array.flat()) } else { return q[fnc](key, Array.isArray(array) ? array : [array]) } } ) } if (filters.string) { iterate(filters.string, (q, key, value) => { const fnc = allOr ? "orWhere" : "where" // postgres supports ilike, nothing else does if (this.client === SqlClient.POSTGRES) { return q[fnc](key, "ilike", `${value}%`) } else { const rawFnc = `${fnc}Raw` // @ts-ignore return q[rawFnc](`LOWER(${this.quotedIdentifier(key)}) LIKE ?`, [ `${value.toLowerCase()}%`, ]) } }) } if (filters.fuzzy) { iterate(filters.fuzzy, like) } if (filters.range) { iterate(filters.range, (q, key, value) => { const isEmptyObject = (val: any) => { return ( val && Object.keys(val).length === 0 && Object.getPrototypeOf(val) === Object.prototype ) } if (isEmptyObject(value.low)) { value.low = "" } if (isEmptyObject(value.high)) { value.high = "" } const lowValid = isValidFilter(value.low), highValid = isValidFilter(value.high) const schema = this.getFieldSchema(key) if (this.client === SqlClient.ORACLE) { // @ts-ignore key = this.knex.raw(this.convertClobs(key)) } if (lowValid && highValid) { if ( schema?.type === FieldType.BIGINT && this.client === SqlClient.SQL_LITE ) { return q.whereRaw( `CAST(${key} AS INTEGER) BETWEEN CAST(? AS INTEGER) AND CAST(? AS INTEGER)`, [value.low, value.high] ) } else { const fnc = allOr ? "orWhereBetween" : "whereBetween" return q[fnc](key, [value.low, value.high]) } } else if (lowValid) { if ( schema?.type === FieldType.BIGINT && this.client === SqlClient.SQL_LITE ) { return q.whereRaw(`CAST(${key} AS INTEGER) >= CAST(? AS INTEGER)`, [ value.low, ]) } else { const fnc = allOr ? "orWhere" : "where" return q[fnc](key, ">=", value.low) } } else if (highValid) { if ( schema?.type === FieldType.BIGINT && this.client === SqlClient.SQL_LITE ) { return q.whereRaw(`CAST(${key} AS INTEGER) <= CAST(? AS INTEGER)`, [ value.high, ]) } else { const fnc = allOr ? "orWhere" : "where" return q[fnc](key, "<=", value.high) } } return q }) } if (filters.equal) { iterate(filters.equal, (q, key, value) => { const fnc = allOr ? "orWhereRaw" : "whereRaw" if (this.client === SqlClient.MS_SQL) { return q[fnc]( `CASE WHEN ${this.quotedIdentifier(key)} = ? THEN 1 ELSE 0 END = 1`, [value] ) } else if (this.client === SqlClient.ORACLE) { const identifier = this.convertClobs(key) return q[fnc](`(${identifier} IS NOT NULL AND ${identifier} = ?)`, [ value, ]) } else { return q[fnc](`COALESCE(${this.quotedIdentifier(key)} = ?, FALSE)`, [ value, ]) } }) } if (filters.notEqual) { iterate(filters.notEqual, (q, key, value) => { const fnc = allOr ? "orWhereRaw" : "whereRaw" if (this.client === SqlClient.MS_SQL) { return q[fnc]( `CASE WHEN ${this.quotedIdentifier(key)} = ? THEN 1 ELSE 0 END = 0`, [value] ) } else if (this.client === SqlClient.ORACLE) { const identifier = this.convertClobs(key) return q[fnc]( `(${identifier} IS NOT NULL AND ${identifier} != ?) OR ${identifier} IS NULL`, [value] ) } else { return q[fnc](`COALESCE(${this.quotedIdentifier(key)} != ?, TRUE)`, [ value, ]) } }) } if (filters.empty) { iterate(filters.empty, (q, key) => { const fnc = allOr ? "orWhereNull" : "whereNull" return q[fnc](key) }) } if (filters.notEmpty) { iterate(filters.notEmpty, (q, key) => { const fnc = allOr ? "orWhereNotNull" : "whereNotNull" return q[fnc](key) }) } if (filters.contains) { contains(filters.contains) } if (filters.notContains) { contains(filters.notContains) } if (filters.containsAny) { contains(filters.containsAny, true) } const tableRef = aliases?.[this.table._id!] || this.table._id // when searching internal tables make sure long looking for rows if (filters.documentType && !isExternalTable(this.table) && tableRef) { // has to be its own option, must always be AND onto the search query.andWhereLike( `${tableRef}._id`, `${prefixed(filters.documentType)}%` ) } return query } isSqs(): boolean { return isSqs(this.table) } getTableName(tableOrName?: Table | string): string { let table: Table if (typeof tableOrName === "string") { const name = tableOrName if (this.query.table?.name === name) { table = this.query.table } else if (this.query.meta.table?.name === name) { table = this.query.meta.table } else if (!this.query.meta.tables?.[name]) { // This can legitimately happen in custom queries, where the user is // querying against a table that may not have been imported into // Budibase. return name } else { table = this.query.meta.tables[name] } } else if (tableOrName) { table = tableOrName } else { table = this.table } let name = table.name if (isSqs(table) && table._id) { // SQS uses the table ID rather than the table name name = table._id } const aliases = this.query.tableAliases || {} return aliases[name] ? aliases[name] : name } addDistinctCount(query: Knex.QueryBuilder): Knex.QueryBuilder { if (!this.table.primary) { throw new Error("SQL counting requires primary key to be supplied") } return query.countDistinct( `${this.getTableName()}.${this.table.primary[0]} as ${COUNT_FIELD_NAME}` ) } addAggregations( query: Knex.QueryBuilder, aggregations: Aggregation[] ): Knex.QueryBuilder { const fields = this.query.resource?.fields || [] const tableName = this.getTableName() if (fields.length > 0) { const qualifiedFields = fields.map(field => this.qualifyIdentifier(field)) if (this.client === SqlClient.ORACLE) { const groupByFields = qualifiedFields.map(field => this.convertClobs(field) ) const selectFields = qualifiedFields.map(field => this.convertClobs(field, { forSelect: true }) ) query = query .groupByRaw(groupByFields.join(", ")) .select(this.knex.raw(selectFields.join(", "))) } else { query = query.groupBy(qualifiedFields).select(qualifiedFields) } } for (const aggregation of aggregations) { const op = aggregation.calculationType if (op === CalculationType.COUNT) { if ("distinct" in aggregation && aggregation.distinct) { if (this.client === SqlClient.ORACLE) { const field = this.convertClobs(`${tableName}.${aggregation.field}`) query = query.select( this.knex.raw( `COUNT(DISTINCT ${field}) as ${this.quotedIdentifier( aggregation.name )}` ) ) } else { query = query.countDistinct( `${tableName}.${aggregation.field} as ${aggregation.name}` ) } } else { query = query.count(`* as ${aggregation.name}`) } } else { const field = `${tableName}.${aggregation.field} as ${aggregation.name}` switch (op) { case CalculationType.SUM: query = query.sum(field) break case CalculationType.AVG: query = query.avg(field) break case CalculationType.MIN: query = query.min(field) break case CalculationType.MAX: query = query.max(field) break } } } return query } addSorting(query: Knex.QueryBuilder): Knex.QueryBuilder { let { sort, resource } = this.query const primaryKey = this.table.primary const aliased = this.getTableName() if (!Array.isArray(primaryKey)) { throw new Error("Sorting requires primary key to be specified for table") } if (sort && Object.keys(sort || {}).length > 0) { for (let [key, value] of Object.entries(sort)) { const direction = value.direction === SortOrder.ASCENDING ? "asc" : "desc" // TODO: figure out a way to remove this conditional, not relying on // the defaults of each datastore. let nulls: "first" | "last" | undefined = undefined if ( this.client === SqlClient.POSTGRES || this.client === SqlClient.ORACLE ) { nulls = value.direction === SortOrder.ASCENDING ? "first" : "last" } let composite = `${aliased}.${key}` if (this.client === SqlClient.ORACLE) { query = query.orderByRaw( `${this.convertClobs(composite)} ${direction} nulls ${nulls}` ) } else { query = query.orderBy(composite, direction, nulls) } } } // add sorting by the primary key if the result isn't already sorted by it, // to make sure result is deterministic const hasAggregations = (resource?.aggregations?.length ?? 0) > 0 if (!hasAggregations && (!sort || sort[primaryKey[0]] === undefined)) { query = query.orderBy(`${aliased}.${primaryKey[0]}`) } return query } tableNameWithSchema( tableName: string, opts?: { alias?: string; schema?: string } ) { let withSchema = opts?.schema ? `${opts.schema}.${tableName}` : tableName if (opts?.alias) { withSchema += ` as ${opts.alias}` } return withSchema } private buildJsonField(field: string): string { const parts = field.split(".") let tableField: string, unaliased: string if (parts.length > 1) { const alias = parts.shift()! unaliased = parts.join(".") tableField = `${this.quote(alias)}.${this.quote(unaliased)}` } else { unaliased = parts.join(".") tableField = this.quote(unaliased) } const separator = this.client === SqlClient.ORACLE ? " VALUE " : "," return `'${unaliased}'${separator}${tableField}` } maxFunctionParameters() { // functions like say json_build_object() in SQL have a limit as to how many can be performed // before a limit is met, this limit exists in Postgres/SQLite. This can be very important, such as // for JSON column building as part of relationships. We also have a default limit to avoid very complex // functions being built - it is likely this is not necessary or the best way to do it. switch (this.client) { case SqlClient.SQL_LITE: return 127 case SqlClient.POSTGRES: return 100 // other DBs don't have a limit, but set some sort of limit default: return 200 } } addJsonRelationships( query: Knex.QueryBuilder, fromTable: string, relationships: RelationshipsJson[] ): Knex.QueryBuilder { const sqlClient = this.client const knex = this.knex const { resource, tableAliases: aliases, endpoint, meta } = this.query const fields = resource?.fields || [] for (let relationship of relationships) { const { tableName: toTable, through: throughTable, to: toKey, from: fromKey, fromPrimary, toPrimary, } = relationship // skip invalid relationships if (!toTable || !fromTable) { continue } const relatedTable = meta.tables?.[toTable] const toAlias = aliases?.[toTable] || toTable, fromAlias = aliases?.[fromTable] || fromTable, throughAlias = (throughTable && aliases?.[throughTable]) || throughTable let toTableWithSchema = this.tableNameWithSchema(toTable, { alias: toAlias, schema: endpoint.schema, }) const requiredFields = [ ...(relatedTable?.primary || []), relatedTable?.primaryDisplay, ].filter(field => field) as string[] // sort the required fields to first in the list, so they don't get sliced out let relationshipFields = prioritisedArraySort( fields.filter(field => field.split(".")[0] === toAlias), requiredFields ) relationshipFields = relationshipFields.slice( 0, Math.floor(this.maxFunctionParameters() / 2) ) const fieldList: string = relationshipFields .map(field => this.buildJsonField(field)) .join(",") // SQL Server uses TOP - which performs a little differently to the normal LIMIT syntax // it reduces the result set rather than limiting how much data it filters over const primaryKey = `${toAlias}.${toPrimary || toKey}` let subQuery: Knex.QueryBuilder = knex .from(toTableWithSchema) // add sorting to get consistent order .orderBy(primaryKey) const isManyToMany = throughTable && toPrimary && fromPrimary let correlatedTo = isManyToMany ? `${throughAlias}.${fromKey}` : `${toAlias}.${toKey}`, correlatedFrom = isManyToMany ? `${fromAlias}.${fromPrimary}` : `${fromAlias}.${fromKey}` // many-to-many relationship needs junction table join if (isManyToMany) { let throughTableWithSchema = this.tableNameWithSchema(throughTable, { alias: throughAlias, schema: endpoint.schema, }) subQuery = subQuery.join(throughTableWithSchema, function () { this.on(`${toAlias}.${toPrimary}`, "=", `${throughAlias}.${toKey}`) }) } // add the correlation to the overall query subQuery = subQuery.where( correlatedTo, "=", knex.raw(this.quotedIdentifier(correlatedFrom)) ) const standardWrap = (select: string): Knex.QueryBuilder => { subQuery = subQuery.select(`${toAlias}.*`).limit(getRelationshipLimit()) // @ts-ignore - the from alias syntax isn't in Knex typing return knex.select(knex.raw(select)).from({ [toAlias]: subQuery, }) } let wrapperQuery: Knex.QueryBuilder | Knex.Raw switch (sqlClient) { case SqlClient.SQL_LITE: // need to check the junction table document is to the right column, this is just for SQS subQuery = this.addJoinFieldCheck(subQuery, relationship) wrapperQuery = standardWrap( `json_group_array(json_object(${fieldList}))` ) break case SqlClient.POSTGRES: wrapperQuery = standardWrap( `json_agg(json_build_object(${fieldList}))` ) break case SqlClient.MARIADB: // can't use the standard wrap due to correlated sub-query limitations in MariaDB wrapperQuery = subQuery.select( knex.raw( `json_arrayagg(json_object(${fieldList}) LIMIT ${getRelationshipLimit()})` ) ) break case SqlClient.MY_SQL: case SqlClient.ORACLE: wrapperQuery = standardWrap( `json_arrayagg(json_object(${fieldList}))` ) break case SqlClient.MS_SQL: wrapperQuery = knex.raw( `(SELECT ${this.quote(toAlias)} = (${knex .select(`${fromAlias}.*`) // @ts-ignore - from alias syntax not TS supported .from({ [fromAlias]: subQuery .select(`${toAlias}.*`) .limit(getRelationshipLimit()), })} FOR JSON PATH))` ) break default: throw new Error(`JSON relationships not implement for ${sqlClient}`) } query = query.select({ [relationship.column]: wrapperQuery }) } return query } addJoin( query: Knex.QueryBuilder, tables: { from: string; to: string; through?: string }, columns: { from?: string to?: string fromPrimary?: string toPrimary?: string }[] ): Knex.QueryBuilder { const { tableAliases: aliases, endpoint } = this.query const schema = endpoint.schema const toTable = tables.to, fromTable = tables.from, throughTable = tables.through const toAlias = aliases?.[toTable] || toTable, throughAlias = (throughTable && aliases?.[throughTable]) || throughTable, fromAlias = aliases?.[fromTable] || fromTable let toTableWithSchema = this.tableNameWithSchema(toTable, { alias: toAlias, schema, }) let throughTableWithSchema = throughTable ? this.tableNameWithSchema(throughTable, { alias: throughAlias, schema, }) : undefined if (!throughTable) { // @ts-ignore query = query.leftJoin(toTableWithSchema, function () { for (let relationship of columns) { const from = relationship.from, to = relationship.to // @ts-ignore this.orOn(`${fromAlias}.${from}`, "=", `${toAlias}.${to}`) } }) } else { query = query // @ts-ignore .leftJoin(throughTableWithSchema, function () { for (let relationship of columns) { const fromPrimary = relationship.fromPrimary const from = relationship.from // @ts-ignore this.orOn( `${fromAlias}.${fromPrimary}`, "=", `${throughAlias}.${from}` ) } }) .leftJoin(toTableWithSchema, function () { for (let relationship of columns) { const toPrimary = relationship.toPrimary const to = relationship.to // @ts-ignore this.orOn(`${toAlias}.${toPrimary}`, `${throughAlias}.${to}`) } }) } return query } qualifiedKnex(opts?: { alias?: string | boolean }): Knex.QueryBuilder { let alias = this.query.tableAliases?.[this.query.endpoint.entityId] if (opts?.alias === false) { alias = undefined } else if (typeof opts?.alias === "string") { alias = opts.alias } return this.knex( this.tableNameWithSchema(this.query.endpoint.entityId, { alias, schema: this.query.endpoint.schema, }) ) } create(opts: QueryOptions): Knex.QueryBuilder { const { body } = this.query if (!body) { throw new Error("Cannot create without row body") } let query = this.qualifiedKnex({ alias: false }) const parsedBody = this.parseBody(body) if (this.client === SqlClient.ORACLE) { // Oracle doesn't seem to automatically insert nulls // if we don't specify them, so we need to do that here for (const [column, schema] of Object.entries( this.query.meta.table.schema )) { if ( schema.constraints?.presence === true || schema.type === FieldType.FORMULA || schema.type === FieldType.AUTO || schema.type === FieldType.LINK ) { continue } const value = parsedBody[column] if (value == null) { parsedBody[column] = null } } } else { // make sure no null values in body for creation for (let [key, value] of Object.entries(parsedBody)) { if (value == null) { delete parsedBody[key] } } } // mysql can't use returning if (opts.disableReturning) { return query.insert(parsedBody) } else { return query.insert(parsedBody).returning("*") } } bulkCreate(): Knex.QueryBuilder { const { body } = this.query let query = this.qualifiedKnex({ alias: false }) if (!Array.isArray(body)) { return query } const parsedBody = body.map(row => this.parseBody(row)) return query.insert(parsedBody) } bulkUpsert(): Knex.QueryBuilder { const { body } = this.query let query = this.qualifiedKnex({ alias: false }) if (!Array.isArray(body)) { return query } const parsedBody = body.map(row => this.parseBody(row)) if ( this.client === SqlClient.POSTGRES || this.client === SqlClient.SQL_LITE || this.client === SqlClient.MY_SQL || this.client === SqlClient.MARIADB ) { const primary = this.table.primary if (!primary) { throw new Error("Primary key is required for upsert") } return query.insert(parsedBody).onConflict(primary).merge() } else if ( this.client === SqlClient.MS_SQL || this.client === SqlClient.ORACLE ) { // No upsert or onConflict support in MSSQL/Oracle yet, see: // https://github.com/knex/knex/pull/6050 return query.insert(parsedBody) } return query.upsert(parsedBody) } read( opts: { limits?: { base: number; query: number } } = {} ): Knex.QueryBuilder { let { endpoint, filters, paginate, relationships } = this.query const { limits } = opts const counting = endpoint.operation === Operation.COUNT const tableName = endpoint.entityId // start building the query let query = this.qualifiedKnex() // handle pagination let foundOffset: number | null = null let foundLimit = limits?.query || limits?.base if (paginate && paginate.page && paginate.limit) { // @ts-ignore const page = paginate.page <= 1 ? 0 : paginate.page - 1 const offset = page * paginate.limit foundLimit = paginate.limit foundOffset = offset } else if (paginate && paginate.offset && paginate.limit) { foundLimit = paginate.limit foundOffset = paginate.offset } else if (paginate && paginate.limit) { foundLimit = paginate.limit } // counting should not sort, limit or offset if (!counting) { // add the found limit if supplied if (foundLimit != null) { query = query.limit(foundLimit) } // add overall pagination if (foundOffset != null) { query = query.offset(foundOffset) } } const aggregations = this.query.resource?.aggregations || [] if (counting) { query = this.addDistinctCount(query) } else if (aggregations.length > 0) { query = this.addAggregations(query, aggregations) } else { query = query.select(this.generateSelectStatement()) } // have to add after as well (this breaks MS-SQL) if (!counting) { query = this.addSorting(query) } query = this.addFilters(query, filters, { relationship: true }) // handle relationships with a CTE for all others if (relationships?.length) { const mainTable = this.query.tableAliases?.[this.query.endpoint.entityId] || this.query.endpoint.entityId const cte = this.addSorting( this.knex .with("paginated", query) .select(this.generateSelectStatement()) .from({ [mainTable]: "paginated", }) ) // add JSON aggregations attached to the CTE return this.addJsonRelationships(cte, tableName, relationships) } // no relationships found - return query else { return query } } update(opts: QueryOptions): Knex.QueryBuilder { const { body, filters } = this.query if (!body) { throw new Error("Cannot update without row body") } let query = this.qualifiedKnex() const parsedBody = this.parseBody(body) query = this.addFilters(query, filters) // mysql can't use returning if (opts.disableReturning) { return query.update(parsedBody) } else { return query.update(parsedBody).returning("*") } } delete(opts: QueryOptions): Knex.QueryBuilder { const { filters } = this.query let query = this.qualifiedKnex() query = this.addFilters(query, filters) // mysql can't use returning if (opts.disableReturning) { return query.delete() } else { return query.delete().returning(this.generateSelectStatement()) } } } class SqlQueryBuilder extends SqlTableQueryBuilder { private readonly limit: number // pass through client to get flavour of SQL constructor(client: SqlClient, limit: number = getBaseLimit()) { super(client) this.limit = limit } private convertToNative(query: Knex.QueryBuilder, opts: QueryOptions = {}) { const sqlClient = this.getSqlClient() if (opts?.disableBindings) { return { sql: query.toString() } } else { let native = getNativeSql(query) if (sqlClient === SqlClient.SQL_LITE) { native = convertBooleans(native) } return native } } /** * @param json The JSON query DSL which is to be converted to SQL. * @param opts extra options which are to be passed into the query builder, e.g. disableReturning * which for the sake of mySQL stops adding the returning statement to inserts, updates and deletes. * @return the query ready to be passed to the driver. */ _query(json: QueryJson, opts: QueryOptions = {}): SqlQuery | SqlQuery[] { const sqlClient = this.getSqlClient() const config: Knex.Config = { client: this.getBaseSqlClient(), } if (sqlClient === SqlClient.SQL_LITE || sqlClient === SqlClient.ORACLE) { config.useNullAsDefault = true } const client = knex(config) let query: Knex.QueryBuilder const builder = new InternalBuilder(sqlClient, client, json) switch (this._operation(json)) { case Operation.CREATE: query = builder.create(opts) break case Operation.READ: query = builder.read({ limits: { query: this.limit, base: getBaseLimit(), }, }) break case Operation.COUNT: // read without any limits to count query = builder.read() break case Operation.UPDATE: query = builder.update(opts) break case Operation.DELETE: query = builder.delete(opts) break case Operation.BULK_CREATE: query = builder.bulkCreate() break case Operation.BULK_UPSERT: query = builder.bulkUpsert() break case Operation.CREATE_TABLE: case Operation.UPDATE_TABLE: case Operation.DELETE_TABLE: return this._tableQuery(json) default: throw `Operation type is not supported by SQL query builder` } return this.convertToNative(query, opts) } async getReturningRow(queryFn: QueryFunction, json: QueryJson) { if (!json.extra || !json.extra.idFilter) { return {} } const input = this._query({ endpoint: { ...json.endpoint, operation: Operation.READ, }, resource: { fields: [], }, filters: json.extra?.idFilter, paginate: { limit: 1, }, meta: json.meta, }) return queryFn(input, Operation.READ) } // when creating if an ID has been inserted need to make sure // the id filter is enriched with it before trying to retrieve the row checkLookupKeys(id: any, json: QueryJson) { if (!id || !json.meta.table || !json.meta.table.primary) { return json } const primaryKey = json.meta.table.primary?.[0] json.extra = { idFilter: { equal: { [primaryKey]: id, }, }, } return json } // this function recreates the returning functionality of postgres async queryWithReturning( json: QueryJson, queryFn: QueryFunction, processFn: Function = (result: any) => result ) { const sqlClient = this.getSqlClient() const operation = this._operation(json) const input = this._query(json, { disableReturning: true }) if (Array.isArray(input)) { const responses = [] for (let query of input) { responses.push(await queryFn(query, operation)) } return responses } let row // need to manage returning, a feature mySQL can't do if (operation === Operation.DELETE) { row = processFn(await this.getReturningRow(queryFn, json)) } const response = await queryFn(input, operation) const results = processFn(response) // same as delete, manage returning if (operation === Operation.CREATE || operation === Operation.UPDATE) { let id if (sqlClient === SqlClient.MS_SQL) { id = results?.[0].id } else if ( sqlClient === SqlClient.MY_SQL || sqlClient === SqlClient.MARIADB ) { id = results?.insertId } row = processFn( await this.getReturningRow(queryFn, this.checkLookupKeys(id, json)) ) } if (operation === Operation.COUNT) { return results } if (operation !== Operation.READ) { return row } return results.length ? results : [{ [operation.toLowerCase()]: true }] } private getTableName( table: Table, aliases?: Record ): string | undefined { let name = table.name if ( table.sourceType === TableSourceType.INTERNAL || table.sourceId === INTERNAL_TABLE_SOURCE_ID ) { if (!table._id) { return } // SQS uses the table ID rather than the table name name = table._id } return aliases?.[name] || name } convertJsonStringColumns>( table: Table, results: T[], aliases?: Record ): T[] { const tableName = this.getTableName(table, aliases) for (const [name, field] of Object.entries(table.schema)) { if (!this._isJsonColumn(field)) { continue } const fullName = `${tableName}.${name}` as keyof T for (let row of results) { if (typeof row[fullName] === "string") { row[fullName] = JSON.parse(row[fullName]) } if (typeof row[name] === "string") { row[name as keyof T] = JSON.parse(row[name]) } } } return results } _isJsonColumn( field: FieldSchema ): field is JsonFieldMetadata | BBReferenceFieldMetadata { return ( JsonTypes.includes(field.type) && !helpers.schema.isDeprecatedSingleUserColumn(field) ) } log(query: string, values?: SqlQueryBinding) { sqlLog(this.getSqlClient(), query, values) } } export default SqlQueryBuilder