[SPARK-57625][SDP] Manage Auxiliary Table Lifecycle from DatasetManager 2/2
#56686
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Contributor
Author
There was a problem hiding this comment. This file is mostly just moving over existing logic as-is from
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,335 @@ | ||
| /* | ||
| * Licensed to the Apache Software Foundation (ASF) under one or more | ||
| * contributor license agreements. See the NOTICE file distributed with | ||
| * this work for additional information regarding copyright ownership. | ||
| * The ASF licenses this file to You under the Apache License, Version 2.0 | ||
| * (the "License"); you may not use this file except in compliance with | ||
| * the License. You may obtain a copy of the License at | ||
| * | ||
| * http://www.apache.org/licenses/LICENSE-2.0 | ||
| * | ||
| * Unless required by applicable law or agreed to in writing, software | ||
| * distributed under the License is distributed on an "AS IS" BASIS, | ||
| * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| * See the License for the specific language governing permissions and | ||
| * limitations under the License. | ||
| */ | ||
|
|
||
| package org.apache.spark.sql.pipelines.graph | ||
|
|
||
| import scala.util.control.NonFatal | ||
|
|
||
| import org.json4s.JsonAST.{JArray, JString} | ||
| import org.json4s.jackson.JsonMethods.{compact, parse} | ||
|
|
||
| import org.apache.spark.SparkException | ||
| import org.apache.spark.sql.AnalysisException | ||
| import org.apache.spark.sql.catalyst.TableIdentifier | ||
| import org.apache.spark.sql.catalyst.analysis.Resolver | ||
| import org.apache.spark.sql.connector.catalog.{CatalogV2Util, Table => CatalogTable, TableCatalog} | ||
| import org.apache.spark.sql.pipelines.autocdc.{AutoCdcReservedNames, ScdType} | ||
| import org.apache.spark.sql.types.{StructField, StructType} | ||
|
|
||
| /** | ||
| * Helpers to construct and validate an AutoCDC flow's auxiliary table within the context of a | ||
| * dataflow graph. | ||
| */ | ||
| object AutoCdcAuxiliaryTable { | ||
|
Contributor
There was a problem hiding this comment. I wonder why this is an object and not a class. It feels as if the methods here are taking in and passing down the same parameters over and over again.
Contributor
There was a problem hiding this comment. I think we can refactor this - if desired - in a separate PR |
||
| /** | ||
| * Helper for deriving the auxiliary AutoCDC catalog table identifier from a target table. If a | ||
| * table exists with a name matching the name derived here, it is assumed to be an AutoCDC | ||
| * auxiliary table that should be managed by the pipeline. | ||
| */ | ||
| def identifier(destination: TableIdentifier): TableIdentifier = TableIdentifier( | ||
| table = s"${AutoCdcReservedNames.prefix}aux_state_${destination.table}", | ||
| database = destination.database, | ||
| catalog = destination.catalog | ||
| ) | ||
|
|
||
| /** | ||
| * Reserved table property key set on the auxiliary table to record which SCD strategy it | ||
| * serves. | ||
| */ | ||
| val scdTypePropertyKey: String = s"${PipelinesTableProperties.pipelinesPrefix}autocdc.scdType" | ||
|
Contributor
There was a problem hiding this comment. would it make sense to move this to
Contributor
There was a problem hiding this comment. On second thought, let's keep these here. |
||
|
|
||
| /** | ||
| * Table property recording the auxiliary table's unquoted AutoCDC key column names as a JSON | ||
| * string array (e.g. `["id","region"]`). Written once when the auxiliary table is created and is | ||
| * considered immutable; full-refresh is the only way to change it. | ||
| */ | ||
| val keyColumnNamesProperty: String = | ||
|
|
||
| s"${PipelinesTableProperties.pipelinesPrefix}autocdc.keyColumnNames" | ||
|
|
||
| /** | ||
| * Serialize key column names to the JSON form stored at [[keyColumnNamesProperty]]. | ||
| * Round-trips an empty list as `[]`; callers are expected to enforce a non-empty key set | ||
| * upstream. | ||
| */ | ||
| private[graph] def serializeKeyColumnNames(names: Seq[String]): String = { | ||
| compact(JArray(names.map(JString(_)).toList)) | ||
| } | ||
|
|
||
| /** | ||
| * Parse a [[keyColumnNamesProperty]] value. `None` if it is not a JSON array of strings. | ||
| * Round-trips an empty list as `[]`; callers are expected to enforce a non-empty key set | ||
| * upstream. | ||
| */ | ||
| private[graph] def parseKeyColumnNames(raw: String): Option[Seq[String]] = { | ||
| val parsed = try Some(parse(raw)) catch { case NonFatal(_) => None } | ||
|
Contributor
There was a problem hiding this comment. shouldn't this be an error rather than silently returning None?
Contributor
There was a problem hiding this comment. never mind, the caller throws the error if this returns None. |
||
| parsed.flatMap { | ||
| case JArray(elems) => | ||
| val names = elems.collect { case JString(s) => s } | ||
| if (names.size == elems.size) Some(names) else None | ||
| case _ => None | ||
| } | ||
| } | ||
|
|
||
| /** | ||
| * Build the auxiliary table spec given an AutoCdc flow and the target table it writes to. | ||
| * | ||
| * @param targetTable the dataset that owns the auxiliary table | ||
| * @param targetTableSchema the AutoCDC target's evolved schema as of the latest pipeline run | ||
| * (the union of all flows writing to the target after schema | ||
| * evolution, NOT the target's `specifiedSchema`) | ||
| * @param inputAutoCdcFlow the AutoCDC flow writing to `targetTable` | ||
| * @return the auxiliary-table spec | ||
| */ | ||
| def buildAuxiliaryTableSpecFor( | ||
| targetTable: Table, | ||
| targetTableSchema: StructType, | ||
| inputAutoCdcFlow: AutoCdcMergeFlow): AuxiliaryTableSpec = { | ||
| inputAutoCdcFlow.changeArgs.storedAsScdType match { | ||
| case ScdType.Type1 => | ||
| buildScd1AuxiliaryTableSpecFor( | ||
| targetTable, | ||
| targetTableSchema, | ||
| inputAutoCdcFlow | ||
| ) | ||
| case ScdType.Type2 => | ||
| // SCD2 auxiliary derivation lands with SCD2 support. AutoCdcMergeFlow rejects SCD2 at | ||
| // construction today, so a resolved SCD2 flow cannot exist and this branch is unreachable. | ||
| throw SparkException.internalError( | ||
| "SCD2 auxiliary table derivation is not yet implemented." | ||
| ) | ||
| } | ||
| } | ||
|
|
||
| /** | ||
| * Build the SCD1 auxiliary table spec given the AutoCdc flow's declared keys and the target | ||
| * table it writes to. | ||
| * | ||
| * @param targetTable the dataset that owns the SCD1 auxiliary table | ||
| * @param targetTableSchema the AutoCDC target's evolved schema as of the latest pipeline run | ||
| * (the union of all flows writing to the target after schema | ||
| * evolution), from which the key and CDC metadata fields are | ||
| * resolved | ||
| * @param inputAutoCdcFlow the AutoCDC flow writing to `targetTable` | ||
| * @return the SCD1 auxiliary-table spec | ||
| */ | ||
| private def buildScd1AuxiliaryTableSpecFor( | ||
| targetTable: Table, | ||
| targetTableSchema: StructType, | ||
| inputAutoCdcFlow: AutoCdcMergeFlow | ||
| ): AuxiliaryTableSpec = { | ||
| val scd1AuxiliaryTableIdentifier = identifier(targetTable.identifier) | ||
|
|
||
| val resolver = inputAutoCdcFlow.df.sparkSession.sessionState.conf.resolver | ||
| val autoCdcKeyColumnNames = inputAutoCdcFlow.changeArgs.keys.map(_.name) | ||
|
|
||
| // The auxiliary table should derive its schema from the exact same key/CDC metadata column | ||
| // schema in its corresponding target table. Retrieve those column schemas. | ||
| val keyFields = autoCdcKeyColumnNames.map { keyColumnName => | ||
| findFieldInTargetSchema( | ||
| targetTableSchema = targetTableSchema, | ||
| targetTableIdentifier = targetTable.identifier, | ||
| autoCdcFlowIdentifier = inputAutoCdcFlow.identifier, | ||
| fieldName = keyColumnName, | ||
| resolver = resolver | ||
| ) | ||
| } | ||
| val cdcMetadataField = findFieldInTargetSchema( | ||
| targetTableSchema = targetTableSchema, | ||
| targetTableIdentifier = targetTable.identifier, | ||
| autoCdcFlowIdentifier = inputAutoCdcFlow.identifier, | ||
| fieldName = AutoCdcReservedNames.cdcMetadataColName, | ||
| resolver = resolver | ||
| ) | ||
|
|
||
| val scd1AuxiliaryTableSchema = StructType(keyFields :+ cdcMetadataField) | ||
|
|
||
| val scd1AuxiliaryTableProperties = | ||
| // Record which SCD strategy this auxiliary table serves so downstream readers can identify it | ||
| // without inspecting the schema. | ||
| Map(scdTypePropertyKey -> ScdType.Type1.label) ++ | ||
| // Persist the AutoCDC key column names as a JSON list; immutable post-creation (full-refresh | ||
| // is the only way to change it). | ||
| Map(keyColumnNamesProperty -> serializeKeyColumnNames(keyFields.map(_.name))) ++ | ||
| // Inherit the target's format so MERGE semantics line up. When unspecified, omit the provider | ||
| // so the catalog falls back to its default. | ||
| targetTable.format.map(TableCatalog.PROP_PROVIDER -> _) | ||
|
|
||
| AutoCdcAuxiliaryTableSpec( | ||
| identifier = scd1AuxiliaryTableIdentifier, | ||
| schema = scd1AuxiliaryTableSchema, | ||
| properties = scd1AuxiliaryTableProperties, | ||
| targetTableIdentifier = targetTable.identifier, | ||
| expectedKeyFields = keyFields, | ||
| expectedScdType = ScdType.Type1 | ||
| ) | ||
| } | ||
|
|
||
| /** | ||
| * Resolve the [[StructField]] named `fieldName` in `targetTableSchema` (the AutoCDC target's | ||
| * evolved schema). The key columns and the CDC metadata column are always present in that schema, | ||
| * so a miss is an implementation invariant and surfaces as an internal error. | ||
| * | ||
| * @param targetTableSchema the AutoCDC target's evolved schema to resolve against | ||
| * @param fieldName the column name to resolve | ||
| * @param resolver the session resolver used for case-sensitivity-aware field lookups | ||
| * @param targetTableIdentifier the AutoCDC target's identifier, named in the error message | ||
| * @param autoCdcFlowIdentifier the AutoCDC flow writing to the target, named in the error message | ||
| * @return the matching field | ||
| */ | ||
| private def findFieldInTargetSchema( | ||
| targetTableSchema: StructType, | ||
| fieldName: String, | ||
| resolver: Resolver, | ||
| targetTableIdentifier: TableIdentifier, | ||
| autoCdcFlowIdentifier: TableIdentifier): StructField = { | ||
| targetTableSchema.fields | ||
| .find(field => resolver(field.name, fieldName)) | ||
| .getOrElse( | ||
| throw SparkException.internalError( | ||
|
Contributor
There was a problem hiding this comment. Is this the right way to format an error message. Shouldn't we have an error class definition for it?
Contributor
There was a problem hiding this comment. TIL that we don't use error classes for internal errors. |
||
| s"Expected but unable to find column $fieldName in target table " + | ||
| s"$targetTableIdentifier written to by AutoCDC flow $autoCdcFlowIdentifier." | ||
| ) | ||
| ) | ||
| } | ||
|
|
||
| /** | ||
| * Reject an existing auxiliary table whose key columns have drifted from `expectedKeyFields` as a | ||
| * set: same arity, same set of names (per `resolver`), same per-name `dataType`s. Nullability and | ||
| * metadata changes are intentionally tolerated. | ||
| * | ||
| * AutoCDC cannot change keys across incremental runs; a changed key set would otherwise be | ||
| * silently unioned into the schema by the additive evolve. The remedy is a full refresh, which | ||
| * recreates the auxiliary table. Errors name the AutoCDC target table rather than any single | ||
| * flow, since one auxiliary table is shared by every flow writing to that target. | ||
| */ | ||
| private[graph] def validateNoKeyColumnDrift( | ||
| existingAuxiliaryTable: CatalogTable, | ||
| targetTableIdentifier: TableIdentifier, | ||
| expectedKeyFields: Seq[StructField], | ||
| resolver: Resolver): Unit = { | ||
| val existingAuxSchema = CatalogV2Util.v2ColumnsToStructType(existingAuxiliaryTable.columns()) | ||
| val recordedKeyNames = | ||
|
Contributor
There was a problem hiding this comment. I would add a comment that this first validates that the existing aux table is consistent within itself, that is, all key columns exists. |
||
| parseRecordedKeyColumnNames(existingAuxiliaryTable, targetTableIdentifier) | ||
| // First validate the existing auxiliary table is internally consistent: every key column name | ||
| // recorded in its table property must still resolve to a field in its schema. A missing key | ||
| // column means the table was corrupted or modified externally, and is rejected before any | ||
| // drift comparison against the expected key fields. | ||
| val recordedKeyFields: Seq[StructField] = recordedKeyNames.map { name => | ||
| existingAuxSchema.fields | ||
| .find(field => resolver(field.name, name)) | ||
| .getOrElse( | ||
| // Either an implementation bug or, more likely, the user has corrupted the auxiliary | ||
| // table schema (e.g. dropped the key column). The remedy is full-refresh in either case. | ||
| throw new AnalysisException( | ||
| errorClass = "AUTOCDC_INVALID_STATE.AUXILIARY_TABLE_KEY_COLUMN_MISSING", | ||
| messageParameters = Map( | ||
| "tableName" -> targetTableIdentifier.unquotedString, | ||
| "keyColumnName" -> name, | ||
| "propertyName" -> keyColumnNamesProperty | ||
| ) | ||
| ) | ||
| ) | ||
| } | ||
|
|
||
| val drifted = | ||
| // Arity drift (added or dropped keys). | ||
| recordedKeyFields.length != expectedKeyFields.length || | ||
| // Name or dataType drift: every expected key must have a same-name (resolver-aware) recorded | ||
| // counterpart with an equivalent dataType. Columns changing nullability and metadata in the | ||
| // schema are intentionally tolerated, although null key values during microbatch execution | ||
| // will be invalidated regardless. | ||
| expectedKeyFields.exists { expected => | ||
| recordedKeyFields.find(rf => resolver(rf.name, expected.name)) match { | ||
| case None => true | ||
| case Some(recorded) => !recorded.dataType.sameType(expected.dataType) | ||
| } | ||
| } | ||
|
|
||
| if (drifted) { | ||
| throw new AnalysisException( | ||
| errorClass = "AUTOCDC_INVALID_STATE.KEY_SCHEMA_DRIFT", | ||
| messageParameters = Map( | ||
| "tableName" -> targetTableIdentifier.unquotedString, | ||
| "expectedKeySchema" -> StructType(expectedKeyFields).toDDL, | ||
| "recordedKeySchema" -> StructType(recordedKeyFields).toDDL | ||
| ) | ||
| ) | ||
| } | ||
| } | ||
|
|
||
| /** | ||
| * Reject an existing auxiliary table whose recorded `scdType` differs from the expected one. | ||
| * SCD1 and SCD2 auxiliary tables carry different state shapes, so an in-place flip is | ||
| * incompatible; the remedy is a full refresh, which recreates the auxiliary table. | ||
| */ | ||
| private[graph] def validateNoScdTypeDrift( | ||
| existingAuxiliaryTable: CatalogTable, | ||
| targetTableIdentifier: TableIdentifier, | ||
| expectedScdType: ScdType): Unit = { | ||
| val recordedScdType = Option( | ||
| existingAuxiliaryTable.properties().get(scdTypePropertyKey) | ||
| ).getOrElse { | ||
| throw new AnalysisException( | ||
| errorClass = "AUTOCDC_INVALID_STATE.AUXILIARY_TABLE_PROPERTY_MISSING", | ||
| messageParameters = Map( | ||
| "tableName" -> targetTableIdentifier.unquotedString, | ||
| "propertyName" -> scdTypePropertyKey | ||
| ) | ||
| ) | ||
| } | ||
| if (recordedScdType != expectedScdType.label) { | ||
| throw new AnalysisException( | ||
| errorClass = "AUTOCDC_INVALID_STATE.SCD_TYPE_DRIFT", | ||
| messageParameters = Map( | ||
| "tableName" -> targetTableIdentifier.unquotedString, | ||
| "expectedScdType" -> expectedScdType.label, | ||
| "recordedScdType" -> recordedScdType | ||
| ) | ||
| ) | ||
| } | ||
| } | ||
|
|
||
| /** | ||
| * Read [[keyColumnNamesProperty]] off an existing auxiliary table and parse it into the ordered | ||
| * list of recorded AutoCDC key column names. | ||
| */ | ||
| private def parseRecordedKeyColumnNames( | ||
| existingAuxiliaryTable: CatalogTable, | ||
| targetTableIdentifier: TableIdentifier): Seq[String] = { | ||
| val rawKeyColumnNamesStr = Option( | ||
| existingAuxiliaryTable.properties().get(keyColumnNamesProperty) | ||
| ).getOrElse { | ||
| throw new AnalysisException( | ||
| errorClass = "AUTOCDC_INVALID_STATE.AUXILIARY_TABLE_PROPERTY_MISSING", | ||
| messageParameters = Map( | ||
| "tableName" -> targetTableIdentifier.unquotedString, | ||
| "propertyName" -> keyColumnNamesProperty | ||
| ) | ||
| ) | ||
| } | ||
| parseKeyColumnNames(rawKeyColumnNamesStr).getOrElse { | ||
| throw new AnalysisException( | ||
| errorClass = "AUTOCDC_INVALID_STATE.AUXILIARY_TABLE_PROPERTY_MALFORMED", | ||
| messageParameters = Map( | ||
| "tableName" -> targetTableIdentifier.unquotedString, | ||
| "propertyName" -> keyColumnNamesProperty, | ||
| "rawValue" -> rawKeyColumnNamesStr | ||
| ) | ||
| ) | ||
| } | ||
| } | ||
| } | ||
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Small error message changes here, the idea being the messages should reflect that the auxiliary table is a companion to the target table, not necessarily to the AutoCDC flow.
This distinction doesn't really matter now because only one AutoCDC flow can write to a target table per pipeline today, but eventually we will want to support multiple AutoCDC flows writing to the same target table, and sharing the same auxiliary table.