[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Lefty Leverenz updated HIVE-6455: - Labels: optimization (was: TODOC13 TODOC14 optimization) Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Fix For: 0.13.0, 0.14.0 Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.17.patch.txt, HIVE-6455.18.patch, HIVE-6455.19.patch, HIVE-6455.2.patch, HIVE-6455.20.patch, HIVE-6455.21.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Release Note: Added the new config to https://cwiki.apache.org/confluence/display/Hive/Configuration+Properties Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: TODOC13, TODOC14, optimization Fix For: 0.13.0, 0.14.0 Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.17.patch.txt, HIVE-6455.18.patch, HIVE-6455.19.patch, HIVE-6455.2.patch, HIVE-6455.20.patch, HIVE-6455.21.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Lefty Leverenz updated HIVE-6455: - Labels: TODOC13 TODOC14 optimization (was: TODOC14 optimization) Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: TODOC13, TODOC14, optimization Fix For: 0.13.0, 0.14.0 Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.17.patch.txt, HIVE-6455.18.patch, HIVE-6455.19.patch, HIVE-6455.2.patch, HIVE-6455.20.patch, HIVE-6455.21.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Swarnim Kulkarni updated HIVE-6455: --- Labels: TODOC14 optimization (was: optimization) Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: TODOC14, optimization Fix For: 0.13.0, 0.14.0 Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.17.patch.txt, HIVE-6455.18.patch, HIVE-6455.19.patch, HIVE-6455.2.patch, HIVE-6455.20.patch, HIVE-6455.21.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Vikram Dixit K updated HIVE-6455: - Fix Version/s: 0.14.0 0.13.0 Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Fix For: 0.13.0, 0.14.0 Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.17.patch.txt, HIVE-6455.18.patch, HIVE-6455.19.patch, HIVE-6455.2.patch, HIVE-6455.20.patch, HIVE-6455.21.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Vikram Dixit K updated HIVE-6455: - Resolution: Fixed Status: Resolved (was: Patch Available) Committed to trunk and branch-0.13. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.17.patch.txt, HIVE-6455.18.patch, HIVE-6455.19.patch, HIVE-6455.2.patch, HIVE-6455.20.patch, HIVE-6455.21.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.21.patch Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.17.patch.txt, HIVE-6455.18.patch, HIVE-6455.19.patch, HIVE-6455.2.patch, HIVE-6455.20.patch, HIVE-6455.21.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.20.patch Changes in PartitionDesc related to Interners are no longer required. The diffs are consistent if all the test for TestParse are run together which I believe is what precommit QA is doing. Refreshed the patch to trunk. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.17.patch.txt, HIVE-6455.18.patch, HIVE-6455.19.patch, HIVE-6455.2.patch, HIVE-6455.20.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.19.patch Rebased the patch to trunk. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.17.patch.txt, HIVE-6455.18.patch, HIVE-6455.19.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.18.patch Fix for test failures. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.17.patch.txt, HIVE-6455.18.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.17.patch.txt Reuploading for jenkins Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.17.patch.txt, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.17.patch Addressed [~vikram.dixit]'s review comments. Fixed the test failures. [~hagleitn]'s Can you take a look at PartitionDesc changes as it replaces the Interners from HIVE-6262? Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.17.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.16.patch Thanks [~gopalv] for reporting the issue. This patch has fix for Tez and the reported failure. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.16.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.15.patch Fixed the test failures caused by ReduceSinkDeDuplication Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.15.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.13.patch Reuploading .13 patch for precommit tests to run. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.14.patch Precommit tests did not run. So ran some tests manually after ReducerDeDuplication changes. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.13.patch, HIVE-6455.14.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.12.patch Added fixes for test failures. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.13.patch Removed some dead code. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.12.patch, HIVE-6455.13.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.11.patch Addressed [~hagleitn]'s code review comments. This is intermediate checkin to look for precommit test failures. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.11.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.10.patch To kick-off precommit tests earlier I reuploaded .9 version instead of .10 version. Reuploading .10 version again to kick off another precommit test run. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.10.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.2#6252)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.9.patch Reuploading for HIVE QA to pick it up. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.9.patch Addressed [~hagleitn]'s review comments. Following changes were made in this patch 1) Updated optimization rule operator regex 2) Removed redundant addition of bucket number to values 3) Using startGroup() to look for changes in the key Not all test cases are updated. Will wait for precommit test failures. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.10.patch Fixed some test failures. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.10.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch, HIVE-6455.9.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.8.patch Added fix when dynamic partition context is null for bucketed tables. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch, HIVE-6455.8.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.7.patch Added a fix that solved an issue with stats aggregation when stats aggregation key exceeds the max key prefix length. Fixed other failing tests. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch, HIVE-6455.7.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.6.patch Fixed the test failures reported by HIVE QA. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch, HIVE-6455.6.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.4.patch Reuploading patch as HIVE QA did not run yesterday. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.5.patch Fixed more tests. Not sure why precommit tests is not picking up this patch. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Lefty Leverenz updated HIVE-6455: - Assignee: Prasanth J (was: Lefty Leverenz) Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch, HIVE-6455.4.patch, HIVE-6455.5.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.1.patch Reuploading patch as HIVE-6037 changes prevented this patch to run precommit tests. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.2.patch Reuploading the same patch with different version. (not sure if precommit tests will be triggered if the file names are same but with different upload times) Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.2.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.3.patch Fixed some test failures by regenerating golden files which are expected to happen with this patch. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.2.patch, HIVE-6455.3.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.4.patch Unrelated HiveConf slipped in earlier patch. Removed it in this patch. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.4.patch Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch, HIVE-6455.1.patch, HIVE-6455.2.patch, HIVE-6455.3.patch, HIVE-6455.4.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Description: The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. was: The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Attachment: HIVE-6455.1.patch Attaching the initial patch. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
[jira] [Updated] (HIVE-6455) Scalable dynamic partitioning and bucketing optimization
[ https://issues.apache.org/jira/browse/HIVE-6455?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ] Prasanth J updated HIVE-6455: - Status: Patch Available (was: Open) Making it as patch available for HIVE QA to run precommit tests. Scalable dynamic partitioning and bucketing optimization Key: HIVE-6455 URL: https://issues.apache.org/jira/browse/HIVE-6455 Project: Hive Issue Type: New Feature Components: Query Processor Affects Versions: 0.13.0 Reporter: Prasanth J Assignee: Prasanth J Labels: optimization Attachments: HIVE-6455.1.patch The current implementation of dynamic partition works by keeping at least one record writer open per dynamic partition directory. In case of bucketing there can be multispray file writers which further adds up to the number of open record writers. The record writers of column oriented file format (like ORC, RCFile etc.) keeps some sort of in-memory buffers (value buffer or compression buffers) open all the time to buffer up the rows and compress them before flushing it to disk. Since these buffers are maintained per column basis the amount of constant memory that will required at runtime increases as the number of partitions and number of columns per partition increases. This often leads to OutOfMemory (OOM) exception in mappers or reducers depending on the number of open record writers. Users often tune the JVM heapsize (runtime memory) to get over such OOM issues. With this optimization, the dynamic partition columns and bucketing columns (in case of bucketed tables) are sorted before being fed to the reducers. Since the partitioning and bucketing columns are sorted, each reducers can keep only one record writer open at any time thereby reducing the memory pressure on the reducers. This optimization is highly scalable as the number of partition and number of columns per partition increases at the cost of sorting the columns. -- This message was sent by Atlassian JIRA (v6.1.5#6160)
