This document contains release information about VMware Tanzu Greenplum Database 6.x releases. For previous versions of the release notes for Greenplum Database, go to VMware Tanzu Greenplum Database Documentation. For information about Greenplum Database end of life, see VMware Tanzu Greenplum Database end of life policy.

VMware Tanzu Greenplum 6 software is available for download from the VMware Tanzu Greenplum page on VMware Tanzu Network.

VMware Tanzu Greenplum 6 is based on the open source Greenplum Database project code.

Upgrading Greenplum

See Upgrading to Greenplum 6 to upgrade your existing Greenplum software.

Release 6.23

Release 6.23.0

Release Date: 2022-12-21

VMware Tanzu Greenplum 6.23.0 is a minor release that includes new and changed features and resolves several issues.

Features

Greenplum Database 6.23.0 includes these new and changed features:

  • The CREATE TABLE command includes a new storage parameter – analyze_hll_non_part_table – which allows you to force collection of HLL statistics even if the table is not part of a partitioned table.

  • The gpmt gp_log_collector command now takes a new option – -with-gpcc – which collects Greenplum Command Center logs from a variety of locations, including:

  • $GPCC_HOME/logs/*

  • $GPCC_HOME/conf/app.conf
  • $HOME/gpmetrics/* (Greenplum Command Center 6.8.0+ on Greenplum Database 6.x or Greenplum Command Center 4.16.0+ on Greenplum Database 5.x)
  • $MASTER_DATA_DIRECTORY/gpmetrics (Greenplum Command Center 6.7.0- on Greenplum Database 6.x or Greenplum Command Center 4.15.0- on Greenplum Database 5.x)
  • The output of the gppkg -q --all command
  • The gpmt gp_log_collector command now takes a new option – with-gpss – which collects all logs relevant to Greenplum Streaming Server.
  • The CPUSET syntax for reserving CPU cores for a resource group has changed to require separate separate core specifications for the master host and segments hosts. When using CPUSET with CREATE RESOURCE GROUP or ALTER RESOURCE GROUP you must configure cores for the master host first and then segment hosts, using a semicolon to separate the two. For example, CPUSET='1;1-3' configures core 1 for the master host, and cores 1 through 3 for segment hosts.
  • The Greenplum Database cluster management utilities now use version 5.7.0 of the psutil Python library.
  • The embedded Python distribution is updated to version 2.7.18.
  • Greenplum Database 6.23 limits the number of supported CUBE grouping columns to a maximum of 12.
  • Improves analyzedb performance when it operates on a large number of tables.
  • The gpsd and minirepro diagnostic tools now include the number of segments in their command output.
  • The performance of the minirepro diagnostic tool is improved by using the Postgres Planner instead of the Pivotal Query Optimizer during minirepro restore.
  • Backports upstream PostgreSQL pg_dump performance enhancements.
  • Includes pgcrypto module changes to facilitate testing.
  • The license file for the Windows Client and Loader Tools Package is updated with new terms.

  • Greenplum Database 6.23.0 introduces three new modules:

  • ltree, which provides data types for representing labels of data stored in a hierarchical tree-like structure.

  • timestamp9, which provides an efficient nanosecond-precision timestamp data type.
  • uuid-ossp, which provides functions to generate universally unique identifiers (UUIDs).

  • Version 6.23.0 introduces the following new server configuration parameters:

  • gp_print_create_gang_time - directs Greenplum Database to log additional information about worker process (gang) creation during query execution.

  • gp_max_parallel_cursors - limits the number of active parallel retrieve cursors allowed on a Greenplum Database cluster.
  • optimizer_skew_factor - provides a method to configure GPORCA for skew-aware join costing.

  • The diskquota module was updated to version 2.1, which includes these changes:

  • The 10 database limit is removed, diskquota can now monitor up to 50 databases.

  • The default value of diskquota.max_active_tables is changed to 300 * 1024 to reduce default memory usage.
  • diskquota adds a new configuration parameter, diskquota.max_workers, that you can set to specify the maximum number of diskquota worker processes that may be running at one time.
  • diskquota also adds the diskquota.max_table_segments configuration parameter. You can set this parameter to specify the maximum number of table shards supported by diskquota in the Greenplum Database cluster, which gates the number of tables that it can support.

Refer to the diskquota module documentation for installation and upgrade instructions.

Resolved Issues

Server

32596

Resolved an issue where Greenplum Database dereferenced a null pointer and PANICed during execution of a query that included record-type parameters.

32567

Resolved an issue where a query could not be terminated with pg_terminate_backend() because Greenplum Database polled an invalid ( -1) file descriptor. Greenplum now removes invalid file descriptors from the polling list.

32557

Resolved an issue where Greenplum Database returned the error correlated subquery with skip-level correlations is not supported when it executed a query (that ran successfully in a previous version of Greenplum) that included a single-level correlated subquery.

32502

Added additional logging code for resource queue events.

14465

Resolved an issue where Greenplum Database returned the error no pre-assigned OID for relation ... when an index was created with a WHERE clause that invoked a function.

32573

Resolved an issue where Greenplum Database was throwing a “PANIC”,“XX000”,“Unexpected internal error Master process received signal SIGSEGV” error due to incorrect handling of errors in the explain analyze phase of execution.

14510

Resolved a problem where tables created using the WITH (appendonly, orientation=column) had no entries in pg_attribute_encoding, resulting in a table that could not be queried.

14446

Resolved a problem where GRANT and REVOKE statements that involved both partitioned and non-partitioned tables applied their changes only to partitioned tables, with no error or indication that non-partitioned tables were ignored. The code was modified to ensure that such statements are applied to both partitioned and non-partitioned tables.

122

Resolved problem where pg_dumpall, used by the VMware Greenplum upgrade utility, did not correctly dump tablespaces in VMware Greenplum version 5.x databases.

121

Resolved a problem where upgrade could fail with a pg_upgrade error similar to: could not open Append-Only segment file ""base/16410/20863.1665"": No such file or directory"

n/a

Resolved a problem that could cause the error, ERROR: external table <table> command ended with SHELL TERMINATED by signal SIGPIPE (13), with external table queries executed on the master segment when all tuples were not retrieved from a table scan. The problem was fixed by ensuring that squelching mechanics are applied after a SIGPIPE for queries that excecute on the master node, as they are applied for queries executed on segment nodes.

Query Processing

32481

Resolved an issue where Greenplum Database cancelled a query due to high virtual memory usage when object identifier duplication caused GPORCA to fall back to the Postgres Planner. GPORCA can now handle object identifiers that are not unique across the system.

32453

Resolves an issue where GPORCA failed to produce a plan when the query involved an update on a partitioned table by removing an unused DML column.

32291

Resolves an issue where GPORCA generated a poor plan because it did not identify operation skew. GPORCA introduces the new optimizer_skew_factor server configuration parameter that you can use to enable skew-aware join costing.

14155

Resolves an issue where a query failed with the error data is corrupt because GPORCA used an incorrect hashing function when it translated a constant value for a text-type domain.

N/A

Resolves an issue where a query that included a DynamicSeqScan with a subplan failed with the error Unexpected internal error.

Cluster Management

32190

Resolved an issue where gprecoverseg failed to detect than an instance of pg_basebackup was already running, which led to the corruption of the data directory and a PANIC error.

Extensions

N/A

Resolved an issue where diskquota failed to launch new background workers before a slots used by a previous worker was freed. Also changed the logging level for background worker events to less verbose (from INFO to DEBUG1).

N/A

Resolved an issue where, when running pg_basebackup, the last WAL segment file could have a checksum mismatch across the primary and mirror segments.

Other

14530

Resolved an issue with the open source distribution of Greenplum Database where the installation package was missing required Python libraries.

Release 6.22

Release 6.22.2

Release Date: 2022-11-18

VMware Tanzu Greenplum 6.22.2 is a maintenance release that resolves several issues.

Resolved Issues

Server

N/A

Resolved an issue where treating non-plan nodes as plan nodes was causing memory corruption and PANIC errors.

N/A

Resolved an issue where a CREATE INDEX command on an AO/CO table would hang if there was a pre-existing read-only transaction accessing the same table.

N/A

Resolved an issue where a key internal process was crashing when theree were too many simultaneous connections to the database.

N/A

Resolved an issue where memory corruption occurred after a VACUUM command due to an internal data structure not being correctly initialized.

N/A

Resolved an issue where DELETE FROM queries on AO/CO auxiliary tables were erroring out.

N/A

Resolved an issue with pg_dump where, during gpcopy operations, external leaf tables were copied as heap tables instead of external tables.

32523, 32314

Resolved an issue where Greenplum Database might drop packets and generate interconnect errors due to incorrect cache eviction logic on remote peer shutdown.

32367

Resolved an issue where Greenplum Database was consuming high amounts of memory when processing column oriented tables that had high numbers of partitions.

32367, 14122

Resolved memory leaks related to dynamic bitmap scan on column-oriented tables.

32495

Resolved a performance issue that occured when Greenplum Database planned a query that included a subselect containing a GROUP BY clause (as distinct).

32491

Resolved an issue where pg_upgrade failed when it restored an external table that had a dropped column(s).

32455

Resolved an issue where a PANIC error occurred after an attempt to split a partition into a new partition and a default partition on a table that did not have a default partition.

32458

Resolved an issue where running an internal function on a partition table caused the master Greenplum Database instance to crash.

32500

Resolved an issue introduced in Greenplum Database 6.21.0 where the error "PANIC","XX000","Unexpected internal error: Master process received signal SIGSEGV occurred after executing a query that included an unsupported feature.

14177

Resolved an issue where a snapshot could not be imported because the timestamp was missing, affecting snapshot visibility.

14107

Resolved an issue where Greenplum Database returned the error operator 37 is not a valid ordering operator because it incorrectly generated a sort operator and equality operator when the types of left and right operand differed.

14003

Resolved a compile warning by replacing a strncpy() call with memcpy().

Query Processing

32525

Resolved a problem where GPORCA would fall back to the planner for queries having an intersection of bit values if one of the bit vectors returned a size of 0. GPORCA now correctly removes the empty vector and produces a correct query plan.

32505

Resolved an issue where GPORCA did not generate a performant plan when the query included an N-ary join with only inner joins, and the predicate was false.

32480

Resolved a crash that could occur during query planning for certain UNION queries if the output column for the first child of the UNION was dropped in the projection, making statistics for the column unavailable. The problem was resolved by ensuring that none of the project eleents are dropped.

32450

Resolved a problem where GPORCA failed to do direct dispatch when filtering gp_segment_id on conjunction constraints.

32291

GPORCA now produces a more performant query plan for queries where it can propagate a predicate from scalar ANY subquery to the outer relation.

14297

Fixed a regression in PostGIS compatibility that caused an exception if the TVF constant value was NULL. GPORCA now correctly falls back to the planner in this case.

14273

Resolved an issue where GPORCA could crash when a query involved dynamic partition elimination with a dynamic partial scan.

14246

Resolved a dependency issue that occurred when minirepo was run on a query that accessed a materialized view.

Data Flow

32536

Resolved a pgbouncer problem where admin users logging into the pgbouncer database were logged in as the pgbouncer user instead of as the configured admin_user.

Release 6.22.1

Release Date: 2022-10-07

VMware Tanzu Greenplum 6.22.1 is a maintenance release that resolves several issues.

Resolved Issues

Server

32314-1

Resolved a problem where queries could hang while using an interconnect proxy if network issues occurred during common TCP interconnect teardown operations. A timeout mechanism was introduced to ensure that sender(s) have closed the peer end of the receiver’s connections.

32314-2

Resolved a problem where logs from ic_proxy did not consider the current setting of the gp_log_interconnect configuration parameter. The parameter value is now taken into consideration in the ic_proxy logging mechanism.

14189

Resolves an issue where a mirror failed to start archiving WAL after it was promoted when the mirror was configured with archive_mode=on.

14000

Resolves an issue where Greenplum Database wasted resources when it examined plan nodes that belong to other slices and initialized unnecessary structures. Greenplum now eliminates alien nodes before execution for entry db.

13997

The psql command \d+ now shows correct index information for append-only auxiliary tables.

13739

Resolves an issue where Greenplum Database returned a wrong result or silently errored out when a subplan contained a SharedScan and the subplan’s parent Plan included motion, by fixing cross-slice interaction detection for subplans.

11948

Resolved a crash that could occur because Greenplum incorrectly allowed REINDEX TABLE on a partitioned table from within a PL/SQL function. Because REINDEX TABLE on partitioned table expands the table and starts a new transaction to reindex each table, the operation cannot be rolled back and may crash if called within PL/SQL. The code was modified to prevent this operation from running inside of a function.

Query Processing

32467

Resolved a query failure that could occur if a predicate with a set-returning function was pushed down in cases where the context could not accept a set. The existing logic was updated to ensure that such predicates are not pushed down.

32465

Resolves an issue where Greenplum Database consumed a large amount of memory when GPORCA planned certain queries that generated a dynamic bitmap scan.

32307

Resolved a problem where GPORCA generated poor query plans because it did not permit aggregate functions from running on replicated slices. This meant that results were always broadcasted or gathered on a single segment in order to ensure consistency and correct results. The code was modified to ensure that commonly-used aggregate functions which are not sensitive to the order of data (such as sum, avg, count, min, and max) are safely executed. For cases where aggregate functions are sensitive to the order of data (such as array_agg and custom user-created functions) this optimization is avoided.

32233

Addressed a performance problem where outer join operations performed unnecessary data redistribution to create a hash distribution specification from the inner relation. The code was modified to derive a combined hash specification for outer join operations from both relations.

Cluster Management

32289

Resolves an issue where gpcheckcat incorrectly flagged default privilege objects as having missing dependencies.

11105

Resolved a problem where gpactivatestandby promotion could fail because the process did not wait long enough for the walreceiver process to shut down. The problem was fixed by synchronizing the standby promotion timeout value with the mirror promotion timeout value.

Release 6.22.0

Release Date: 2022-09-09

VMware Tanzu Greenplum 6.22.0 is a minor release that includes new and changed features and resolves several issues.

Features

Greenplum Database 6.22.0 includes these new and changed features:

  • Greenplum Database now includes a plugin that allows you to monitor Greenplum Text clusters via Greenplum Command Center. For more information, see Monitoring Greenplum Text.
  • A new extension, plpython3u, is provided to support developing Python 3.9 functions with PL/Python. See PL/Python Language. An additional Data Science Package for Python is provided to install common Python 3.9 modules for developing plypython3u functions. See Data Science Package for Python.
  • The Data Science Package for Python 2.7 was updated to add the regex package. The package versions for lxml, nltk, tenserflow, tenserflow-gpu, and keras were updated to address CVEs.
  • pg_dump performance was improved to reduce the time required to dump database schemas during the upgrade process. This especially benefits clusters that have a large number of append-optimized tables.
  • Greenplum Database now includes a view – gp_suboverflowed_backend – created by the new extension gp_subtransaction_overflow– to help users query for suboverflowed backends. Users can use segment id and process id information reported in the view to terminate the offending backends, thereby preventing degradation of performance. For more information, see Monitoring a Greenplum System.
  • Greenplum Database now includes an extension – gp_pitr – to support implementing Point-in-Time Recovery for Greenplum Database 6X. gp_pitr creates a new view – gp_stat_archiver – and two new internally used user-defined functions. For more information, see Additional Supplied Modules.
  • The pgcrypto extension was updated to support FIPS encryption via the pgcrypto.fips configuration parameter. See Configuring FIPS Encryption in the pgcrypto documentation.
  • The diskquota module was updated to version 2.0.1, and introduces a change to logging behavior. Now, if the diskquota module is enabled in a database that has existing data but you do not run the diskquota.init_table_size_table() function, a single error message is logged when the Greenplum cluster starts up. In earlier versions of diskquota, the error message was logged repeatedly (approximately every 2 seconds) and could flood the log monitor system with the same error.
  • Greenplum Database introduces a new server configuration parameter, optimizer_enable_replicated_table, that you can use to control GPORCA’s behavior when it encounters DML operations on replicated tables. By default, GPORCA attempts to plan and execute operations on replicated tables.
  • Greenplum Streaming Server (GPSS) version 1.8.0 is included, which includes new features, changes, and bug fixes. Refer to the Greenplum Streaming Server Documentation for more information about this release and for upgrade instructions.
  • The s3 protocol now supports obtaining S3 authentication credentials from the default AWS credential file, ~/.aws/credentials.
  • gpfdist now supports reading and writing files compressed with the Zstandard compression algorithm.
  • The file protocol now supports reading files compressed with the Zstandard compression algorithm.
  • Greenplum Database 6.22.0 introduces a new module, ip4r, that provides data types for operations on IPv4 and IPv6 IP addresses.
  • Greenplum Database 6.22.0 introduces a new module, postgresql-hll, that provides HyperLogLog data types for PostgreSQL and Greenplum Database.
  • The gp_percentile_agg module has been updated fix a bug and to generate more performant plans for ordered-set aggregates with skewed data. If you are currently using this module, you must upgrade the module after you upgrade Greenplum Database.
  • Greenplum Database 6.22.0 adds the f_smgr public interface for storage operations.
  • The performance of gpexpand computation of table size has been improved.
  • The PgBouncer distributed with Greenplum Database 6.22.0 adds support for native LDAP authentication between the psql client and the pgbouncer process. Refer to Configuring LDAP-based Authentication for PgBouncer for configuration information.
  • Greenplum Database supports configuring the ldapsearchfilter attribute in pg_hba.conf. You can optionally use this attribute to provide a search filter to use when doing search+bind authentication. See LDAP Authentication.
  • Greenplum Database improves pg_dump performance by gathering AO metadata in a single query.
  • The VMware Tanzu Greenplum Connector for Apache NiFi version 1.1.0 is available, which includes a change. Refer to the Greenplum Connector for Apache NiFi Documentation for more information about this release.
  • The gpmt gp_log_collector utility now includes the --with-gpbackup option to collect logs related to backup and restore operations. See gpmt gp_log_collector.
  • The DataDirect ODBC drivers were updated to version 7.1.6+7.16.445 to include the latest fixes and to no longer require acceptance of a separate licensing agreement. If you have installed a DataDirect ODBC driver version 7.1.6+7.16.389 or earlier, see Upgrading DataDirect ODBC Drivers. The new version resolves a compatibility issue with the Greenplum Direct Query Custom .pqx connector and Microsoft’s PowerBI Gateway. Both the ODBC driver and .pqx connector must be upgraded on Desktop and Gateway systems.

Resolved Issues

Server

32398

Fixes double freeing of a buffer during stream teardown that resulted in a PANIC.

32384

Resolves an issue where a CTE query that involved several columns caused Greenplum Database to PANIC.

32381

Resolves Postgres CVE-2022-2625, Extension scripts replace objects not belonging to the extension.

32271

Resolves an issue where SELECT or other read-only operations resulted in setting hint bits on pages, which generated Full Page Images (FPI) WAL records, and introduced a large replication lag. Greenplum Database now extends the use of the wait_for_replication_threshold server configuration parameter for this scenario.

32063

Resolves an issue where gpupgrade execute ran slow due to inefficient and repetitive pg_dump metadata queries.

31879, 32167

Subtransaction overflows that are causing performance problems can now be easily identified and terminated.

31298

Resolves an issue where a big lag between the primary and mirror during incremental recovery halted transactions on the primary.

14041

Resolves an issue where creating a bitmap index on a column, and using the bitmap index to scan with a NULL array condition on this column, failed.

13929

Resolves an occasional overflow state by adding a GUC to control the output of a suboverflow transaction SQL statement.

13902

Resolved an issue where the current client_encoding is applied as the default value, causing a problem when client encoding is different from server encoding. The problem is fixed by using GPDB server encoding by default, whether it is QD or QE, making the default encoding of external tables the same as the database encoding.

13888

Resolves an issue where a query on a replicated table that was created from INSERT INTO SELECT FROM another replicated table returned incorrect results when sequence generation was involved.

13883

Fixed potential memory leak by using delete to destroy objects allocated by new[].

13882

Fixes query time failures caused by escaped characters in S3 URLs.

13614

Improves performance by adding a GUC to control the output of suboverflow SQL statement transactions.

11848

Resolves an issue where the final query plan included unnecessary sort nodes when the query included grouping sets with distinct qualified aggregates.

Query Processing

32273

Adds server configuration parameter optimizer_enable_replicated_table, which is on by default, to control GPORCA planning and execution of operations on a replicated table. GPORCAfalls back to the Postgres Planner when this parameter is set to off.

Data Flow

13811, 8194

Removes unnecessary null pointer checks in gpcloud.

13882

Replaces escape characters with spaces in s3 protocol URLs to prevent query time failures.

31797

Resolves an issue with s3 protocol S3 authentication by adding support for obtaining credentials from the default AWS credential file.

32029

Resolves a pg_exttable catalog table encoding mismatch by defaulting to the server encoding.

32365

Resolves a gpfdist performance issue by adding support for Zstandard compression.

Cluster Management

181803382

Resolves an issue where gpinitsystem could throw an error when running with the debug ( -D) option.

32157

Resolves an issue where gpexpand would repeat the error Port <port-number> used multiple times on <host-name> if a segments were not in their preferred roles. gpexpand now correctly assigns port numbers in this case, but generates a warning message to indicate that segments are not operating in their preferred roles.

DataDirect ODBC

32225

The DataDirect ODBC driver version 7.1.6+7.16.445 resolves a compatibility issue with the Greenplum Direct Query Custom .pqx connector and Microsoft’s PowerBI Gateway. Both the ODBC driver and .pqx connector must be upgraded on Desktop and Gateway systems. See Upgrading DataDirect ODBC Drivers.

Release 6.21

Release 6.21.3

Release Date: 2022-10-28

Server

32495

Resolved a problem where the planner produced an inefficient query plan and poor performance if a subquery contained a GROUP BY clause (as DISTINCT), because the sublink could not be converted to a join.

Release 6.21.2

Release Date: 2022-08-25

Server

32358

Index creation is now allowed on partitioned tables containing external partitions. Creating a non-unique index on a partitioned table will ignore external partitions, but fail if any of the indexes is a constraint index.

32293

Updates the index VACUUM strategy for Append-Optimized tables to correctly determine when an index entry can be removed. An index entry is now removed only after ensuring that the entry points to a droppable segment, and is therefore already invisible.

31941

LDAP bind credentials that were leaked to the log after failed login attempts are no longer a security risk.

13860

The EPQ routine for AO/AOCS relations and DML node has been disallowed.

13699

Duplicate gp_fastsequence values ( ctid) no longer appear in AO tables.

Query Processing

32348

Non-optimal plans that caused queries to hang were fixed by setting the number of rebinds (rescans) correctly when statistics are copied.

Release 6.21.1

Release Date: 2022-08-03

Features

Greenplum Database 6.21.1 includes a new metrics collector extension that is compatible with Greenplum Command Center 6.8.

Resolved Issues

Server

13881

Resolves an issue in transaction processing where transactions were erroneously rolled back because gang processes were being deleted.

12690

Resolves an issue where, when a memory limit was hit, a query was throwing errors without calling the correct cleanup function.

13874

Resolves an error handling issue that caused PANIC errors during query execution.

32237

Fixes for PostgreSQL CVE-2022-1552 have been backported to secure Autovacuum, REINDEX, and other commands that activate relevant protections too late or not at all.

32211

A pg_upgrade check for parent partitions with seg entries was added to prevent upgrade failure, and logicalEof was added to errdetail in OpenAOSegmentFile.

32210

Primary now writes a special WAL at the place where the mirror is supposed to continue streaming the rest of the broken WAL.

31937

pg_resetxlog infrastructure is used to set the oldest xid instead of freezing the catalog, preventing CLOG truncation during a segment upgrade.

13706

Resolves an error, FATAL: Unexpected internal error (list.c:395), that could occur when creating a plan for a correlated subquery against a partitioned table. The problem occurred in some queries when an InitPlan could not be created, and the planner generated copies of nested subplans in the wrong order. The fix ensures that first level subplans and nested suplans are generated correctly in this situation.

13696

Improves planning performance for DML queries by avoiding unnecessary planning for join tree relations, especially with complex subqueries.

13694

A fix for the logic in pg_lock_status() that keeps it from missing some rows and counting others twice has been backported.

13678

Resolves a problem where unnecessary checks were performed when runaway_detector_activation_percent was deactivated (set to 100).

13675

An error is thrown when COPY FROM/INSERT will result in inserted data being invisible to user queries issued from master.

13669

Resolved an assertion failure, "FATAL" error "Unexpected internal error (vmem_tracker.c:436)","Process 399817 will wait for gp_debug_linger=120 seconds before termination, that could occur during initialization when resource groups were enabled.

11662

Appendoptimized materialized views now have invalid relfrozenxid.

6716

Resolves a problem where CREATE EXTENSION WITH SCHEMA failed because Tanzu Greenplum did not set up the search_path on both the Query Dispatcher and the Query Executor before executing the extension script.

Query Processing

32302

Prevents optimizer from crashing during query optimization due to lack of statistics.

32273

Adds a GUC that keeps queries on distributed replicated tables from hanging when using ORCA by controlling fallback for replicated tables.

32238

ORCA now requires an index type to properly cost bitmap scans before performing full table scans.

32233

Eliminates query performance difference between GPORCA and Postgres.

32125

The master logger process that was creating high load on master, causing segments to go down, was fixed by processing the string with two kinds of modification.

32061

ORCA now prevents creation of spill files by pushing cast operations below the join when safe.

13746

Fallback to Postgres optimizer when a target list in CTE producer is empty.

13683

Fixes incorrect addrefs that caused memory leaks.

Data Flow

32199

Resolves a problem in the fixed width formatting code that could cause the error The line delimiter specified in the Formatter arguments: < 3> is not located in the data file even though the text was formatted correctly. The problem occurred because the code did not handle the situation where the end of a buffer coincided with the end of the data (and therefore does not include an expected newline character).

32151

Resolves a gpfdist crash that could occur during session clean up if a session timed due to a disk full condition. Instead of crashing, gpfdist now logs an error for this condition.

Tanzu Greenplum on vSphere

1102

Resolved an issue where users were seeing WARNINGS in gpinitsystem logs when there was a mismatch between what the hostname utility reported and the actual segment host name.

1042

Resolved an issue where “command not found” warnings were seen during deployment of Greenplum.

1034

Resolved an issue where deployment would fail if the segment count was equal to 1 or odd numbers for Mirrored Greenplum deployments.

844

Resolved an issue where a manual shutdown of the Guest OS would show a ‘failed’ state for the Greenplum Virtual Service.

Release 6.21.0

Release Date: 2022-06-10

VMware Tanzu Greenplum 6.21.0 is a minor release that includes new and changed features and resolves several issues.

Features

Greenplum Database 6.21.0 includes these new and changed features:

  • The LOCK TABLE SQL command now includes a MASTER ONLY option; when enabled, Greenplum Database locks tables on the master only, rather than on the master and all of the segments. This option is particularly useful for metadata-only operations.
  • The gp_parallel_retrieve_cursor contrib module is no longer Beta, it is promoted to a supported feature.
  • The greenplum_fdw contrib module is no longer Beta, it is promoted to a supported feature.
  • Greenplum Database now supports the SET TRANSACTION SNAPSHOT command.
  • Greenplum Database 6.21.0 adds support for both the SCRAM-SHA-256 password hashing algorithm and the scram-sha-256 client authentication method. Refer to Protecting Passwords in Greenplum Database for more information.
  • Greenplum Database now includes a configuration parameter — gp_interconnect_address_type — to specify the type of address binding strategy Greenplum Database uses for communication between segment host sockets. This allows users, if their configuration permits, to specify that Greenplum use a unicast address — rather than a wildcard address — for interconnect sockets. This reduces port usage on segment hosts and prevents interconnect traffic from being routed through unintended (and possibly slower) network interfaces.
  • Log messages that are generated for connection checks between the Query Executor and Query Dispatcher now include the error number that was generated by the receiver.
  • Greenplum Database introduces a new server configuration parameter, gp_log_endpoints, that you can use to enable logging of parallel retrieve cursor endpoint details.
  • Greenplum Database introduces the new resource group server configuration parameter gp_resource_group_enable_recalculate_query_mem. The default value of this parameter is false; Greenplum Database calculates the maximum per-query memory allotment for each segment host based on the memory configuration and the number of primary segments on the master host. If the memory configuration of your Greenplum Database master and segment hosts differ, set the parameter to true so that Greenplum calculates the maximum per-transaction memory allotment for a segment host based on the configuration of that host.
  • Greenplum Database 6.21.0 introduces a new module, gp_percentile_agg, that you can use to improve GPORCA plan generation performance for queries that include ordered-set aggregate functions such as percentile_cont(), percentile_disc(), and median().
  • Greenplum Database 6.21.0 includes version 2.0 of the diskquota module. This version introduces support for schema and role tablespace disk quotas as well as per-segment tablespace quotas. diskquota can also now enforce quota hard limits; if a query exceeds a disk quota during execution, diskquota terminates the query. Refer to the module documentation for more information and for upgrade instructions.
  • The VMware Tanzu Greenplum Connector for Apache NiFi version 1.0.1 is available, which includes a change. See the Connector 1.0 Release Notes for more information.
  • The Greenplum Magic Tool (gpmt) family of tools now includes the gp_storage_rca_collector tool, which collects storage-related artifacts. See the storage_rca_collector topic for details.
  • The gpmt gp_log_collector tool now collects PXF logs and gpupgrade logs. In addition, this tool has been enhanced to collect system initialization timestamps, /var/log/dmesg content, and output from top and sar commands.
  • The PgBouncer distributed with Greenplum Database has been updated to version 1.16.1.
  • Greenplum Streaming Server (GPSS) version 1.7.2 is included, which includes changes and bug fixes. Refer to the Greenplum Streaming Server Documentation for more information about this release and for upgrade instructions.

  • For customers deploying Greenplum on vSphere, the following new features and changes have been introduced in this release:

  • The Greenplum Virtual Machine now provides a service running on the master VM that monitors the postmaster process and restarts the process if it fails. This improves the reliability and availability of the postmaster process. It is enabled by default.

  • The Greenplum Virtual Machine now provides a service running on each segment VM that monitors the segment process and restarts the process if it fails. This improves the reliability and availability of the segment process. It is enabled by default.
  • Greenplum on vSphere is now supported on vSphere 6.7 and higher.
  • Greenplum on vSphere now supports mirrorless Greenplum deployments.
  • The VxRail reference architecture has been updated to recommend 100GbE networking instead of 25GbE. Increasing the available network throughput significantly improves storage system performance and provides a more balanced system across resources.
  • The PostGIS spatial database extension version 2.5.4+pivotal.7.build.1 is included, which contains updated GEOS and EXPAT dependencies. Because the GEOS geometry library is updated to version 3.10.2; PostGIS now includes support for the ST_Subdivide, ST_ClipByBox2D, ST_VoronoiLines, and ST_VoronoiPolygons functions.

Resolved Issues

Server

32247, 13532

Prevents panic in Planner by not bringing a SegmentGeneral path that refs outer Params to SingleQE.

32229

Resolves a problem with TCP/UDP connections communicating across different subnets by adding a new GUC, gp_interconnect_address_type, which allows users to switch from the wildcard default to a unicast IP address to support motion communication. (Resolved by PR-13344.)

32220

Moves incrementing txns_using_rel to follow ERROR reports to fix leaks in AppendOnlyHash entries. (Resolved by PR-13570.)

32213, 13549

Resolves wrong date behavior by rejecting ambiguous 5 digit date strings in non-standard formats.

32173

Resolves slow intermittent query response by using hostname instead of an IP address to compute the host segments.

32067, 32141

Resolves an issue where OOM errors have occurred during queries by adding a new GUC, gp_resource_group_enable_recalculate_query_mem.

32066

Prevents gp_tablespace_segment_location() from executing on entrydb QE. (Resolved by PR-13287.)

31957

Enhances OLAP performance by ensuring that both sides of the NOT-IN sublink testexpr are non-nullable. (Resolved by PR-13420.)

13446

Fixes incorrect scan position during bitmap index words scans. (Resolved by PR-13522.)

32205

Falls back to Planner from ORCA if a target list is empty.

13467

Resolves a cache lookup failure for attribute 0 of relation xxx from child inheritance/partition table to parent.

182378648

Enables recursion for utility mode connections to propagate ALTER OWNER commands

（本文超出站点单文档体积上限，已截断）