From fe87946eaae631ac4f1a020b74c92c3f1c981caa Mon Sep 17 00:00:00 2001 From: Ti Chi Robot Date: Wed, 2 Aug 2023 13:12:32 +0800 Subject: [PATCH] Fix TiFlash overflow (#2740) (#2743) * This is an automated cherry-pick of #2740 Signed-off-by: ti-chi-bot * Update TiBlockColumnVector.java * fmt --------- Signed-off-by: ti-chi-bot Co-authored-by: shi yuhang <52435083+shiyuhang0@users.noreply.github.com> Co-authored-by: shiyuhang <1136742008@qq.com> --- .../org/apache/spark/sql/IssueTestSuite.scala | 27 +++++++++++++++++++ .../tikv/columnar/TiBlockColumnVector.java | 6 +++++ 2 files changed, 33 insertions(+) diff --git a/core/src/test/scala/org/apache/spark/sql/IssueTestSuite.scala b/core/src/test/scala/org/apache/spark/sql/IssueTestSuite.scala index f1d4f42eb2..e510bb17a3 100644 --- a/core/src/test/scala/org/apache/spark/sql/IssueTestSuite.scala +++ b/core/src/test/scala/org/apache/spark/sql/IssueTestSuite.scala @@ -22,6 +22,33 @@ import org.apache.spark.sql.functions.{col, sum} class IssueTestSuite extends BaseTiSparkTest { + test("test tiflash overflow in unsigned bigint") { + if (!enableTiFlashTest) { + cancel("tiflash test not enabled") + } + val dbTable = "tispark_test.tiflash_overflow" + tidbStmt.execute(s"drop table if exists $dbTable") + tidbStmt.execute( + s"CREATE TABLE $dbTable (`a` bigint(20) UNSIGNED NOT NULL,`b` bigint(20) UNSIGNED NOT NULL)") + tidbStmt.execute(s"insert into $dbTable values(16717361816800086255, 16717361816800086255)") + tidbStmt.execute(s"ALTER TABLE $dbTable SET TIFLASH REPLICA 1") + + Thread.sleep(5 * 1000) + + val prev = spark.conf.getOption(TiConfigConst.ISOLATION_READ_ENGINES) + try { + spark.conf + .set(TiConfigConst.ISOLATION_READ_ENGINES, TiConfigConst.TIFLASH_STORAGE_ENGINE) + val df = spark.sql(s"select * from $dbTable") + val row = Row(BigDecimal("16717361816800086255"), BigDecimal("16717361816800086255")) + checkAnswer(df, Seq(row)) + } finally { + spark.conf.set( + TiConfigConst.ISOLATION_READ_ENGINES, + prev.getOrElse(TiConfigConst.DEFAULT_STORAGE_ENGINES)) + } + } + test("test issue 2649") { val dbTable = "tispark_test.mutil_uniq" tidbStmt.execute(s"drop table if exists $dbTable") diff --git a/tikv-client/src/main/java/com/pingcap/tikv/columnar/TiBlockColumnVector.java b/tikv-client/src/main/java/com/pingcap/tikv/columnar/TiBlockColumnVector.java index 906d1682fa..737156eb36 100644 --- a/tikv-client/src/main/java/com/pingcap/tikv/columnar/TiBlockColumnVector.java +++ b/tikv-client/src/main/java/com/pingcap/tikv/columnar/TiBlockColumnVector.java @@ -18,6 +18,7 @@ import static com.pingcap.tikv.util.MemoryUtil.EMPTY_BYTE_BUFFER_DIRECT; +import com.google.common.primitives.UnsignedLong; import com.pingcap.tikv.ExtendedDateTime; import com.pingcap.tikv.codec.Codec.DateCodec; import com.pingcap.tikv.codec.Codec.DateTimeCodec; @@ -25,6 +26,7 @@ import com.pingcap.tikv.types.AbstractDateTimeType; import com.pingcap.tikv.types.BytesType; import com.pingcap.tikv.types.DateType; +import com.pingcap.tikv.types.DecimalType; import com.pingcap.tikv.util.MemoryUtil; import java.math.BigDecimal; import java.nio.ByteBuffer; @@ -240,6 +242,10 @@ public double getDouble(int rowId) { @Override public BigDecimal getDecimal(int rowId, int precision, int scale) { long rowIdAddr = (long) rowId * fixedLength + dataAddr; + // avoid unsigned long overflow here + if (type == DecimalType.BIG_INT_DECIMAL) { + return new BigDecimal(UnsignedLong.fromLongBits(this.getLong(rowId)).bigIntegerValue()); + } if (fixedLength == 4) { return MemoryUtil.getDecimal32(rowIdAddr, scale); } else if (fixedLength == 8) {