[SPARK-18856] Newly created catalog table assumed to have 0 rows and 0 bytes

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Fix Version/s: 2.1.0
• Component/s: SQL
• Target Version/s: 2.1.0
• Assignee: Wenchen Fan
• Reporter: Reynold Xin
• Created: 14/Dec/16 07:55
• Updated: 15/Dec/16 05:03
• Resolved: 15/Dec/16 05:03

Description

scala> spark.range(100).selectExpr("id % 10 p", "id").write.partitionBy("p").format("json").saveAsTable("testjson")

scala> spark.table("testjson").queryExecution.optimizedPlan.statistics
res6: org.apache.spark.sql.catalyst.plans.logical.Statistics = Statistics(sizeInBytes=0, isBroadcastable=false)

It shouldn't be 0. The issue is that in DataSource.scala, we do:

        val fileCatalog = if (sparkSession.sqlContext.conf.manageFilesourcePartitions &&
            catalogTable.isDefined && catalogTable.get.tracksPartitionsInCatalog) {
          new CatalogFileIndex(
            sparkSession,
            catalogTable.get,
            catalogTable.get.stats.map(_.sizeInBytes.toLong).getOrElse(0L))
        } else {
          new InMemoryFileIndex(sparkSession, globbedPaths, options, Some(partitionSchema))
        }

We shouldn't use 0L as the fallback.

Issue Links

• links to: [Github] Pull Request #16280 (cloud-fan)