This is an automated email from the ASF dual-hosted git repository. sunchao pushed a commit to branch branch-3.4 in repository https://gitbox.apache.org/repos/asf/spark.git
The following commit(s) were added to refs/heads/branch-3.4 by this push: new d61bfb87e39 [SPARK-45054][SQL] HiveExternalCatalog.listPartitions should restore partition statistics d61bfb87e39 is described below commit d61bfb87e39fdce163a2e01b4f039e63d4830ab2 Author: Chao Sun <sunc...@apple.com> AuthorDate: Fri Sep 1 20:22:34 2023 -0700 [SPARK-45054][SQL] HiveExternalCatalog.listPartitions should restore partition statistics ### What changes were proposed in this pull request? Call `restorePartitionMetadata` in `listPartitions` to restore Spark SQL statistics. ### Why are the changes needed? Currently when `listPartitions` is called, it doesn't restore Spark SQL statistics stored in metastore, such as `spark.sql.statistics.totalSize`. This means callers who rely on stats from the method call may wrong results. In particular, when `spark.sql.statistics.size.autoUpdate.enabled` is turned on, during insert overwrite Spark will first list partitions and get old statistics, and then compare them with new statistics and see which partitions need to be updated. This issue will sometimes cause it to update all partitions instead of only those partitions that have been touched. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? Added a new test. ### Was this patch authored or co-authored using generative AI tooling? Closes #42777 from sunchao/list-partition-stat. Authored-by: Chao Sun <sunc...@apple.com> Signed-off-by: Chao Sun <sunc...@apple.com> --- .../catalyst/catalog/ExternalCatalogSuite.scala | 25 ++++++++++++++++++++++ .../spark/sql/hive/HiveExternalCatalog.scala | 7 ++++-- 2 files changed, 30 insertions(+), 2 deletions(-) diff --git a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/catalog/ExternalCatalogSuite.scala b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/catalog/ExternalCatalogSuite.scala index 32eb8849427..a8f73cebf31 100644 --- a/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/catalog/ExternalCatalogSuite.scala +++ b/sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/catalog/ExternalCatalogSuite.scala @@ -475,6 +475,31 @@ abstract class ExternalCatalogSuite extends SparkFunSuite { assert(catalog.listPartitions("db2", "tbl2", Some(Map("a" -> "unknown"))).isEmpty) } + test("SPARK-45054: list partitions should restore stats") { + val catalog = newBasicCatalog() + val stats = Some(CatalogStatistics(sizeInBytes = 1)) + val newPart = CatalogTablePartition(Map("a" -> "1", "b" -> "2"), storageFormat, stats = stats) + catalog.alterPartitions("db2", "tbl2", Seq(newPart)) + val parts = catalog.listPartitions("db2", "tbl2", Some(Map("a" -> "1"))) + + assert(parts.length == 1) + val part = parts.head + assert(part.stats.exists(_.sizeInBytes == 1)) + } + + test("SPARK-45054: list partitions by filter should restore stats") { + val catalog = newBasicCatalog() + val stats = Some(CatalogStatistics(sizeInBytes = 1)) + val newPart = CatalogTablePartition(Map("a" -> "1", "b" -> "2"), storageFormat, stats = stats) + catalog.alterPartitions("db2", "tbl2", Seq(newPart)) + val tz = TimeZone.getDefault.getID + val parts = catalog.listPartitionsByFilter("db2", "tbl2", Seq($"a".int === 1), tz) + + assert(parts.length == 1) + val part = parts.head + assert(part.stats.exists(_.sizeInBytes == 1)) + } + test("SPARK-21457: list partitions with special chars") { val catalog = newBasicCatalog() assert(catalog.listPartitions("db2", "tbl1").isEmpty) diff --git a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveExternalCatalog.scala b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveExternalCatalog.scala index 0c556cd34ed..d1355d4075f 100644 --- a/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveExternalCatalog.scala +++ b/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveExternalCatalog.scala @@ -1265,13 +1265,14 @@ private[spark] class HiveExternalCatalog(conf: SparkConf, hadoopConf: Configurat db: String, table: String, partialSpec: Option[TablePartitionSpec] = None): Seq[CatalogTablePartition] = withClient { - val partColNameMap = buildLowerCasePartColNameMap(getTable(db, table)) + val catalogTable = getTable(db, table) + val partColNameMap = buildLowerCasePartColNameMap(catalogTable) val metaStoreSpec = partialSpec.map(toMetaStorePartitionSpec) val res = client.getPartitions(db, table, metaStoreSpec) .map { part => part.copy(spec = restorePartitionSpec(part.spec, partColNameMap)) } - metaStoreSpec match { + val parts = metaStoreSpec match { // This might be a bug of Hive: When the partition value inside the partial partition spec // contains dot, and we ask Hive to list partitions w.r.t. the partial partition spec, Hive // treats dot as matching any single character and may return more partitions than we @@ -1280,6 +1281,7 @@ private[spark] class HiveExternalCatalog(conf: SparkConf, hadoopConf: Configurat res.filter(p => isPartialPartitionSpec(spec, toMetaStorePartitionSpec(p.spec))) case _ => res } + parts.map(restorePartitionMetadata(_, catalogTable)) } override def listPartitionsByFilter( @@ -1293,6 +1295,7 @@ private[spark] class HiveExternalCatalog(conf: SparkConf, hadoopConf: Configurat val clientPrunedPartitions = client.getPartitionsByFilter(rawHiveTable, predicates).map { part => part.copy(spec = restorePartitionSpec(part.spec, partColNameMap)) + restorePartitionMetadata(part, catalogTable) } prunePartitionsByFilter(catalogTable, clientPrunedPartitions, predicates, defaultTimeZoneId) } --------------------------------------------------------------------- To unsubscribe, e-mail: commits-unsubscr...@spark.apache.org For additional commands, e-mail: commits-h...@spark.apache.org