[jira] [Commented] (SPARK-23852) Parquet MR bug can lead to incorrect SQL results
[ https://issues.apache.org/jira/browse/SPARK-23852?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16474819#comment-16474819 ] Apache Spark commented on SPARK-23852: -- User 'henryr' has created a pull request for this issue: https://github.com/apache/spark/pull/21323 > Parquet MR bug can lead to incorrect SQL results > > > Key: SPARK-23852 > URL: https://issues.apache.org/jira/browse/SPARK-23852 > Project: Spark > Issue Type: Bug > Components: SQL >Affects Versions: 2.3.0 >Reporter: Henry Robinson >Assignee: Ryan Blue >Priority: Blocker > Labels: correctness > Fix For: 2.3.1, 2.4.0 > > > Parquet MR 1.9.0 and 1.8.2 both have a bug, PARQUET-1217, that means that > pushing certain predicates to Parquet scanners can return fewer results than > they should. > The bug triggers in Spark when: > * The Parquet file being scanner has stats for the null count, but not the > max or min on the column with the predicate (Apache Impala writes files like > this). > * The vectorized Parquet reader path is not taken, and the parquet-mr reader > is used. > * A suitable <, <=, > or >= predicate is pushed down to Parquet. > The bug is that the parquet-mr interprets the max and min of a row-group's > column as 0 in the absence of stats. So {{col > 0}} will filter all results, > even if some are > 0. > There is no upstream release of Parquet that contains the fix for > PARQUET-1217, although a 1.10 release is planned. > The least impactful workaround is to set the Parquet configuration > {{parquet.filter.stats.enabled}} to {{false}}. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-23852) Parquet MR bug can lead to incorrect SQL results
[ https://issues.apache.org/jira/browse/SPARK-23852?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16472494#comment-16472494 ] Apache Spark commented on SPARK-23852: -- User 'henryr' has created a pull request for this issue: https://github.com/apache/spark/pull/21302 > Parquet MR bug can lead to incorrect SQL results > > > Key: SPARK-23852 > URL: https://issues.apache.org/jira/browse/SPARK-23852 > Project: Spark > Issue Type: Bug > Components: SQL >Affects Versions: 2.3.0 >Reporter: Henry Robinson >Assignee: Ryan Blue >Priority: Blocker > Labels: correctness > Fix For: 2.4.0 > > > Parquet MR 1.9.0 and 1.8.2 both have a bug, PARQUET-1217, that means that > pushing certain predicates to Parquet scanners can return fewer results than > they should. > The bug triggers in Spark when: > * The Parquet file being scanner has stats for the null count, but not the > max or min on the column with the predicate (Apache Impala writes files like > this). > * The vectorized Parquet reader path is not taken, and the parquet-mr reader > is used. > * A suitable <, <=, > or >= predicate is pushed down to Parquet. > The bug is that the parquet-mr interprets the max and min of a row-group's > column as 0 in the absence of stats. So {{col > 0}} will filter all results, > even if some are > 0. > There is no upstream release of Parquet that contains the fix for > PARQUET-1217, although a 1.10 release is planned. > The least impactful workaround is to set the Parquet configuration > {{parquet.filter.stats.enabled}} to {{false}}. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-23852) Parquet MR bug can lead to incorrect SQL results
[ https://issues.apache.org/jira/browse/SPARK-23852?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16469330#comment-16469330 ] Apache Spark commented on SPARK-23852: -- User 'henryr' has created a pull request for this issue: https://github.com/apache/spark/pull/21284 > Parquet MR bug can lead to incorrect SQL results > > > Key: SPARK-23852 > URL: https://issues.apache.org/jira/browse/SPARK-23852 > Project: Spark > Issue Type: Bug > Components: SQL >Affects Versions: 2.3.0 >Reporter: Henry Robinson >Priority: Blocker > Labels: correctness > > Parquet MR 1.9.0 and 1.8.2 both have a bug, PARQUET-1217, that means that > pushing certain predicates to Parquet scanners can return fewer results than > they should. > The bug triggers in Spark when: > * The Parquet file being scanner has stats for the null count, but not the > max or min on the column with the predicate (Apache Impala writes files like > this). > * The vectorized Parquet reader path is not taken, and the parquet-mr reader > is used. > * A suitable <, <=, > or >= predicate is pushed down to Parquet. > The bug is that the parquet-mr interprets the max and min of a row-group's > column as 0 in the absence of stats. So {{col > 0}} will filter all results, > even if some are > 0. > There is no upstream release of Parquet that contains the fix for > PARQUET-1217, although a 1.10 release is planned. > The least impactful workaround is to set the Parquet configuration > {{parquet.filter.stats.enabled}} to {{false}}. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-23852) Parquet MR bug can lead to incorrect SQL results
[ https://issues.apache.org/jira/browse/SPARK-23852?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16451263#comment-16451263 ] Henry Robinson commented on SPARK-23852: Yes it has - the Parquet community are going to do a 1.8.3 release, mostly just for us for this issue. Parquet 1.10 has already been released, and includes this fix. Upgrading Spark trunk to that version is the subject of SPARK-23972. > Parquet MR bug can lead to incorrect SQL results > > > Key: SPARK-23852 > URL: https://issues.apache.org/jira/browse/SPARK-23852 > Project: Spark > Issue Type: Bug > Components: SQL >Affects Versions: 2.3.0 >Reporter: Henry Robinson >Priority: Blocker > Labels: correctness > > Parquet MR 1.9.0 and 1.8.2 both have a bug, PARQUET-1217, that means that > pushing certain predicates to Parquet scanners can return fewer results than > they should. > The bug triggers in Spark when: > * The Parquet file being scanner has stats for the null count, but not the > max or min on the column with the predicate (Apache Impala writes files like > this). > * The vectorized Parquet reader path is not taken, and the parquet-mr reader > is used. > * A suitable <, <=, > or >= predicate is pushed down to Parquet. > The bug is that the parquet-mr interprets the max and min of a row-group's > column as 0 in the absence of stats. So {{col > 0}} will filter all results, > even if some are > 0. > There is no upstream release of Parquet that contains the fix for > PARQUET-1217, although a 1.10 release is planned. > The least impactful workaround is to set the Parquet configuration > {{parquet.filter.stats.enabled}} to {{false}}. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-23852) Parquet MR bug can lead to incorrect SQL results
[ https://issues.apache.org/jira/browse/SPARK-23852?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16450018#comment-16450018 ] Eric Maynard commented on SPARK-23852: -- {color:#33}>There is no upstream release of Parquet that contains the fix for {color}PARQUET-1217{color:#33}, although a 1.10 release is planned.{color} PARQUET-1217 seems to have been merged into Parquet 1.8.3 today. > Parquet MR bug can lead to incorrect SQL results > > > Key: SPARK-23852 > URL: https://issues.apache.org/jira/browse/SPARK-23852 > Project: Spark > Issue Type: Bug > Components: SQL >Affects Versions: 2.3.0 >Reporter: Henry Robinson >Priority: Blocker > Labels: correctness > > Parquet MR 1.9.0 and 1.8.2 both have a bug, PARQUET-1217, that means that > pushing certain predicates to Parquet scanners can return fewer results than > they should. > The bug triggers in Spark when: > * The Parquet file being scanner has stats for the null count, but not the > max or min on the column with the predicate (Apache Impala writes files like > this). > * The vectorized Parquet reader path is not taken, and the parquet-mr reader > is used. > * A suitable <, <=, > or >= predicate is pushed down to Parquet. > The bug is that the parquet-mr interprets the max and min of a row-group's > column as 0 in the absence of stats. So {{col > 0}} will filter all results, > even if some are > 0. > There is no upstream release of Parquet that contains the fix for > PARQUET-1217, although a 1.10 release is planned. > The least impactful workaround is to set the Parquet configuration > {{parquet.filter.stats.enabled}} to {{false}}. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-23852) Parquet MR bug can lead to incorrect SQL results
[ https://issues.apache.org/jira/browse/SPARK-23852?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16436292#comment-16436292 ] Henry Robinson commented on SPARK-23852: [Here's a branch|https://github.com/henryr/spark/tree/spark-23852] with a test that fails if PARQUET-1217 is not fixed. > Parquet MR bug can lead to incorrect SQL results > > > Key: SPARK-23852 > URL: https://issues.apache.org/jira/browse/SPARK-23852 > Project: Spark > Issue Type: Bug > Components: SQL >Affects Versions: 2.3.0 >Reporter: Henry Robinson >Priority: Blocker > Labels: correctness > > Parquet MR 1.9.0 and 1.8.2 both have a bug, PARQUET-1217, that means that > pushing certain predicates to Parquet scanners can return fewer results than > they should. > The bug triggers in Spark when: > * The Parquet file being scanner has stats for the null count, but not the > max or min on the column with the predicate (Apache Impala writes files like > this). > * The vectorized Parquet reader path is not taken, and the parquet-mr reader > is used. > * A suitable <, <=, > or >= predicate is pushed down to Parquet. > The bug is that the parquet-mr interprets the max and min of a row-group's > column as 0 in the absence of stats. So {{col > 0}} will filter all results, > even if some are > 0. > There is no upstream release of Parquet that contains the fix for > PARQUET-1217, although a 1.10 release is planned. > The least impactful workaround is to set the Parquet configuration > {{parquet.filter.stats.enabled}} to {{false}}. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-23852) Parquet MR bug can lead to incorrect SQL results
[ https://issues.apache.org/jira/browse/SPARK-23852?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16423491#comment-16423491 ] Henry Robinson commented on SPARK-23852: Partly, but not completely. If the column is dictionary encoded, the filters are applied to the dictionary as well at the row-group level. See [RowGroupFilter|https://github.com/apache/parquet-mr/blob/master/parquet-hadoop/src/main/java/org/apache/parquet/filter2/compat/RowGroupFilter.java#L92]. > Parquet MR bug can lead to incorrect SQL results > > > Key: SPARK-23852 > URL: https://issues.apache.org/jira/browse/SPARK-23852 > Project: Spark > Issue Type: Bug > Components: SQL >Affects Versions: 2.3.0 >Reporter: Henry Robinson >Priority: Blocker > Labels: correctness > > Parquet MR 1.9.0 and 1.8.2 both have a bug, PARQUET-1217, that means that > pushing certain predicates to Parquet scanners can return fewer results than > they should. > The bug triggers in Spark when: > * The Parquet file being scanner has stats for the null count, but not the > max or min on the column with the predicate (Apache Impala writes files like > this). > * The vectorized Parquet reader path is not taken, and the parquet-mr reader > is used. > * A suitable <, <=, > or >= predicate is pushed down to Parquet. > The bug is that the parquet-mr interprets the max and min of a row-group's > column as 0 in the absence of stats. So {{col > 0}} will filter all results, > even if some are > 0. > There is no upstream release of Parquet that contains the fix for > PARQUET-1217, although a 1.10 release is planned. > The least impactful workaround is to set the Parquet configuration > {{parquet.filter.stats.enabled}} to {{false}}. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-23852) Parquet MR bug can lead to incorrect SQL results
[ https://issues.apache.org/jira/browse/SPARK-23852?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16423479#comment-16423479 ] Reynold Xin commented on SPARK-23852: - Does turning the flag parquet.filter.stats.enabled off also turning off row group level skipping? > Parquet MR bug can lead to incorrect SQL results > > > Key: SPARK-23852 > URL: https://issues.apache.org/jira/browse/SPARK-23852 > Project: Spark > Issue Type: Bug > Components: SQL >Affects Versions: 2.3.0 >Reporter: Henry Robinson >Priority: Critical > > Parquet MR 1.9.0 and 1.8.2 both have a bug, PARQUET-1217, that means that > pushing certain predicates to Parquet scanners can return fewer results than > they should. > The bug triggers in Spark when: > * The Parquet file being scanner has stats for the null count, but not the > max or min on the column with the predicate (Apache Impala writes files like > this). > * The vectorized Parquet reader path is not taken, and the parquet-mr reader > is used. > * A suitable <, <=, > or >= predicate is pushed down to Parquet. > The bug is that the parquet-mr interprets the max and min of a row-group's > column as 0 in the absence of stats. So {{col > 0}} will filter all results, > even if some are > 0. > There is no upstream release of Parquet that contains the fix for > PARQUET-1217, although a 1.10 release is planned. > The least impactful workaround is to set the Parquet configuration > {{parquet.filter.stats.enabled}} to {{false}}. -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org