Welcome to Greenplum Database

Transcription

1 Rev: A01 Updated: October 10, 2014 Welcome to Greenplum Database Greenplum Database is a massively parallel processing (MPP) database server that supports next generation data warehousing and large-scale analytics processing. By automatically partitioning data and running parallel queries, it allows a cluster of servers to operate as a single database supercomputer performing tens or hundreds times faster than a traditional database. It supports SQL, MapReduce parallel processing, and data volumes ranging from hundreds of gigabytes, to hundreds of terabytes. Note: This document contains pertinent release information about Greenplum Database For previous versions of the release notes for Greenplum Database, go to Pivotal Documentation. About Pivotal, Inc. Greenplum is currently transitioning to a new corporate identity (Pivotal Software, Inc.). During this transition, there will be some legacy instances of our former corporate identity (Greenplum) appearing in our products and documentation. If you have any questions or concerns, please do not hesitate to contact us through our web site: About Greenplum Database Greenplum Database is a patch release that includes Greenplum Database external tables support for MapR, performance and stability enhancements, and fixes known issues. Please refer to the following sections for more information about this release. Product Enhancement Changed and Deprecated Features Downloading Greenplum Database Supported Platforms Resolved Issues in Greenplum Database x Known Issues in Greenplum Database Upgrading to Greenplum Database 4.2.x.x Greenplum Database Tools Compatibility Greenplum Database Extensions Compatibility Hadoop Distribution Compatibility Greenplum Database 4.2 Documentation Welcome to Greenplum Database

2 Product Enhancement Greenplum Database supports MapR with Greenplum Database external tables that use the gphdfs protocol to access HDFS data. For information about MapR support, see Hadoop Distribution Compatibility. For information about external tables, see Loading and Unloading Data in the Greenplum Database Database Administrator Guide. Changed and Deprecated Features The following are Greenplum Database changed and deprecated features. Changed Feature To enhance the performance of the Greenplum Command Center database gpperfmon, the distribution key of the log_alert_* tables has been changed to the logtime column from the logsegment column. The first time you start Greenplum Database after an upgrade, the distribution key is changed and the table data is redistributed. For information about migrating the gpperfmon database see Upgrading from 4.2.x.x to x. Deprecated Features The gpsnmpd agent is deprecated and will be removed in a future release The gpdetective utility is deprecated and will be removed in a future release. Connecting to IBM Cognos with ODBC drivers is deprecated will be removed in a future release. Greenplum Database supports Connecting to IBM Cognos with JDBC drivers. See Supported Platforms for more information. Note: Send questions or comments about deprecated items to gpdb@pivotal.io. Downloading Greenplum Database You can download Greenplum Database software and documentation from these locations: Greenplum Database software is available from Pivotal Network. Current release Greenplum Database documentation is available from the Pivotal Documentation site. Previous release Greenplum Database 4.2 software can be downloaded from the EMC Download Center. Previous release versions of Greenplum Database documentation, as well as other Greenplum Database documents, are available from the EMC Support Zone Product Enhancement 2

3 Supported Platforms Greenplum Database runs on the following platforms: Red Hat Enterprise Linux 64-bit 5.x and 6.x SUSE Linux Enterprise Server 64-bit 10 SP4 11 SP1, 11 SP2, and 11SP3 Solaris x86 64-bit v10 U7, U8, U9, U10 Note: The Solaris operating system is deprecated in the Greenplum Database release and will not be supported in a future release. Oracle Unbreakable Linux 64-bit 5.5 CentOS 64-bit 5.x and 6.x Greenplum Database supports Data Domain Boost on Red Hat Enterprise Linux. This table lists the versions of Data Domain Boost SDK and DDOS supported by Greenplum Database x, x, and x Table 1 Data Domain Boost Compatibility Greenplum Database Data Domain Boost DDOS , 5.3, and , 5.3, and , 5.1, and x , 5.1, and 5.2 GPDB was certified against the following DCA versions: DCA v1 - ISO_VERSION=" " DCA v2 - ISO_VERSION=" " Important: In the next major release of Greenplum Database, connecting to IBM Cognos software with an ODBC driver will not be supported. Greenplum Database supports connecting to IBM Cognos software with a JDBC driver. Pivotal recommends that user migrate to a version of IBM Cognos software that support connecting Greenplum Database with an JDBC driver. Send questions or comments about deprecated items to gpdb@pivotal.io. Supported Platforms 3

4 Resolved Issues in Greenplum Database x The table below lists issues that are now resolved in Greenplum Database x. For issues resolved in prior x releases, refer to the corresponding release notes available from Support Zone. Table 2 Resolved Issues in x Issue Number Category Resolved in Description DDL and Utility Statements When creating a temporary table with the ON COMMIT DELETE ROWS clause in a heavy workload environment, the prepared transaction that created the temporary table failed in some cases Monitoring: Commander Center Alerting In the Greenplum Command Center database gpperfmon, performance issues were caused by data skew in the log_alert_history table. The distribution key for the table has been changed to resolve the issue. See Changed Feature Management Scripts: gpsd Storage Access Methods The Greenplum Database support utility gpsd has been has been enhanced to be more memory efficient in order to prevent failures when used on a database that contains large schemas In some cases, queries that were run against the gp_toolkit view gp_size_of_table_and_indexes_disk returned an error when the queries are run concurrently with DDL statements Backup and Restore A table could not be restored (with the gpdbrestore -T option) from a back up on a Data Domain Boost system that was created with the gpcrondump --ddboost options GPHDFS Greenplum Database external tables support accessing HDFS data with the gphdfs protocol and MapR configured with multiple clusters. See Product Enhancement DDL and Utility Statements Monitoring: gpperfmon server Catalog and Metadata Functions and Languages Query Execution, Storage Access Methods Temporary tables were not properly cleaned up in the following situation. A user defined function (UDF) was created with a security definer and that included statements to create the temporary table. The UDF was executed by a regular user who was given EXECUTE permission on the function. This caused the temporary table to stay in the database after the session was disconnected In some cases, a memory leak causes the gpmmon process to consume a large amount of memory and CPU resources Some operations created duplicate entries in system catalog table gp_distribution_policy. The duplicate entries caused the Greenplum Database utility gpcrondump to fail In some cases, Greenplum Database did not handle data of type date properly and caused a segmentation fault If either a non-partitioned append-only table or an individual append-only part of a partitioned table had more than 127 million rows on a segment, a query that uses an index to access the table data could return duplicate rows. This issue has been fixed. Resolved Issues in Greenplum Database x 4

5 Table 2 Resolved Issues in x Issue Number Category Resolved in Description Security Greenplum Database software has been updated to use OpenSSL 0.9.8zb in response to the OpenSSL Security Advisory [6 Aug 2014]. For information about the advisory, see GPHDFS The Greenplum Database external table protocol gphdfs supports the Cloudera 4.x and 5.x HDFS distributions. See Hadoop Distribution Compatibility Query Execution In some cases, a segmentation fault occurred when a DELCARE CURSOR WITH HOLD command was run by an ODBC driver Loaders: gpfdist The Greenplum Database gpfdist utility failed with a SIGSEGV error when the utility received a empty request with two consecutive return characters \n\n GPHDFS The Greenplum Database external table protocol gphdfs supports the Pivotal 2.0 distribution. See Hadoop Distribution Compatibility PostGIS When using PostGIS, In some cases a closed curved polygon that was converted to a linear polygon was not closed due to a linear approximation precision issue with PostGIS Loaders: gpfdist, Loaders: gpload Catalog and Metadata, Vacuum In some cases when network load was heavy, the Greenplum Database utility gpfdist intermittently failed with this error: gpfdist closed connection to server In some cases, when a VACUUM FULL command was cancelled, incorrect handling of Greenplum Database transaction log caused a PANIC signal to be issued and prevented Greenplum Database from performing a crash recovery of a segment mirror Backup and Restore During a backup operation, the Greenplum Database utility gpcrondump held an EXCLUSIVE lock on the catalog table pg_class longer than required Management Scripts In some cases after expanding a Greenplum Database system, running gpinitstandby -n failed to resynchronize the data between the primary and standby master host Replication: Segment Mirroring In some rare cases, if a restart occurred while the gprecoverseg utility was running, some tables and a persistent table were detected having less data on a mirror segment that corresponds to a primary segment Query Execution Greenplum Database did not manage temporary workfiles (spill files) properly. In some cases, this caused a query that required workfiles to fail with a message that stated that a Greenplum Database segment had reached the maximum configured workfile usage limit Query Execution In some cases, the clean up of an aborted transaction was not handled correctly and caused a PANIC signal to be issued Languages: PL/Java In some cases, Greenplum Database did not handle concurrent shared memory operations properly from PL/Java routines. This caused a PANIC signal to be issued. Resolved Issues in Greenplum Database x 5

6 Table 2 Resolved Issues in x Issue Number Category Resolved in Description Backup and Restore, Management Scripts: master mirroring When configuring a Greenplum Database system with a standby master, the gpinitstandby utility did not correctly update the pg_hba.conf file on Greenplum Database segment hosts Vacuum Performing a VACUUM operation on a partitioned append-optimized table did not correctly reduce the age of the parent table and child tables Replication: Segment Mirroring In some rare cases, the crash recovery of a segment mirror failed due to an inconsistent LSN Interconnect In some cases when a Greenplum Database process was cancelled on the Greenplum Database master, corresponding processes remained running on Greenplum Database segment instances Client Access Methods and Tools Installer, Languages: R PL/R When Greenplum Database received a SIGSEGV when running the COPY command, Greenplum Database hung and continuously logged this warning message: copy: unexpected response (3) The R command line interface could not be started in the R environment that was installed with the Greenplum Database PL/R extension package Query Execution FOR loops in PL/pgSQL did not close the sequence generator if further access was still required Authentication An issue in Greenplum Database prevented LDAPS (LDAP over SSL) from functioning on the standard secure port Replication: Segment Mirroring Replication: Segment Mirroring A transaction lock did not block reader processes from proceeding when a writer process was holding the same lock. In some cases this caused a race condition to occur. Now, Greenplum Database blocks reader processes when a writer process holds the same lock to prevent race conditions from occurring In some cases, Greenplum Database continuously logged this message when sending file replication process statistics to Greenplum Database perfmon process: Error when sending file rep stats to perfmon Security Greenplum Database software has been updated to use OpenSSL 0.9.8za in response to the OpenSSL Security Advisory [05 Jun 2014]. For information about the advisory, see Backup and Restore In some cases, running the Greenplum Database gpdbrestore utility with the -T or --table-file option failed with this error. ValueError: need more than 1 value to unpack Locking, Signals, Processes Transaction Management In some cases, concurrent CREATE TABLE and DROP TABLE operations caused Greenplum Database to hang due to incorrect lock handling Some queries against an append-only table with compression enabled that contains a column with an unknown data type caused a Greenplum Database SIGSEGV error. Resolved Issues in Greenplum Database x 6

7 Table 2 Resolved Issues in x Issue Number Category Resolved in Description Query Execution In some cases, when the disk space for temporary spill files was exhausted, the affected Greenplum Database segment instance would encounter a SIGSEGV Client Access Methods and Tools For Greenplum Database with Kerberos enabled, the database role attribute Valid Until was ignored. This issue has been resolved Query Execution In some cases, when a Greenplum Database segment fault occurred during the execution of a PL/R function, PL/R hung and continuously returned the same error message Management Scripts: expansion Build, Monitoring: gpperfmon server During the process of adding new hosts, the Greenplum Database expand utility gpexpand did not update the pg_hba.conf files on Greenplum Database hosts with the correct host information In some cases, the Greenplum Database gpmmon process failed. The gpmmon process is used for Greenplum Database performance monitoring Languages: R, PLR In Greenplum Database, a signal handling issue in the R programming language caused a potential for postgres processes to hang when running P/LR functions Languages: R, PLR When a NULL value in an array was passed to a PL/R function, a Greenplum Database segment failure occurred. This issue has been resolved Replication: Segment Mirroring Loaders: Copy/Externa Tables The gprecoverseg utility failed to recover a Greenplum Database segment that was marked as down when the data directory location for the segment was a symbolic link, and a postgres process was running with the same PID as the PID associated with the down segment In some cases, when an INSERT FROM SELECT command was run that selected from readable external able and inserted into writable external table, this warning was generated: WARNING select failed on curl_multi_fdset (maxfd 10) (4 - Interrupted system call) Query Execution When executing some SQL queries, Greenplum Database became unresponsive because the query generated a large number of spill files (more than 100,000 files). Now, Greenplum Database limits the maximum number of spill files that can be created by a query Query Execution When a query was run that contained a polymorphic, user-defined aggregate function, and Greenplum Database was required to create spill files on disk, the query failed with this error. ERROR: could not determine actual argument type for polymorphic function This issue has been fixed Dispatch In some cases when temporary tables were used, Greenplum Database did not perform the clean up of temporary namespaces properly after a transaction completed and caused a SIGSEGV. Resolved Issues in Greenplum Database x 7

8 Table 2 Resolved Issues in x Issue Number Category Resolved in Description Query Planner For some queries that contained subqueries, the query planner produced an incorrect plan, which caused errors in the query dispatcher and a postmaster reset Management Scripts: expansion, Management Scripts: master mirroring Storage Access Methods Management Scripts: master mirroring Locking, Signals, Processes The Greenplum Database utilities gpactivatestandby and gpexpand used SSH to connect to localhost (the Greenplum Database host where the utility was run). Using SSH was redundant as the command was already on the local host and has been eliminated After a database expansion, some tables created with APPENDONLY=TRUE and compression enabled consumed much more disk space than before the expansion. To reduce disk space in this situation, the Greenplum Database gpreload utility reloads table data with column data sorted The Greenplum Database gpinitstandby utility completed successfully but returned an error when the $GPHOME/share directory was not writable. Now, the utility returns this warning: Please run gppkg --clean after successful standby initialization Greenplum Database hung due to incorrect lock handling that caused a race condition. The lock handling issue was caused by a compiler optimization Backup and Restore When the Greenplum Database gpdbrestore utility could not find files on the Greenplum Database master segment that are used to perform a restore operation, the utility did not return the correct error message Query Execution In some cases, a user-defined function that executed the EXPLAIN command caused a Greenplum Database PANIC Query Planner In some cases, an SQL query that contains the following returned incorrect results: a combination of a median function with other aggregates where the GROUP BY columns are a subset of the table's distribution columns Replication: Segment Mirroring The Greenplum Database standby master crashed if the transaction logs sent from the primary master contained btree split records with a checkpoint between the split records Management Scripts When a Greenplum Database extension package was updated with the Greenplum Database gppkg utility option -u, gppkg did not warn the user that updating a package includes removing all previous versions of the system objects related to the package. Now, the gppkg utility warns the user and lets the user cancel the operation Dispatch Greenplum Database encountered a SIGSEGV due to improper memory management between the Greenplum Database query dispatcher and query executor Query Planner In certain queries that contain the median function and a GROUP BY clause, the query planner produced an incorrect plan in which some necessary columns were not projected in the operator nodes. This caused an error when trying to look up the missing columns. Resolved Issues in Greenplum Database x 8

9 Table 2 Resolved Issues in x Issue Number Category Resolved in Description Backup and Restore When restoring sets of tables with the Greenplum Database gpdbrestore utility, primary keys, indexes, and GRANT privileges were not restored for filtered tables. For tables with non-english characters (for example, table name and records contained Korean or Chinese characters) primary keys, indexes and GRANT privileges were not restored. For tables with English characters, GRANT privileges were not restored Replication: Segment Mirroring In some cases, running the Greenplum Database command gprecoverseg -r to rebalance segment instances failed due to an inconsistent system catalog Management Scripts The Greenplum Database utility gptransfer copies database objects from databases in a source Greenplum Database system to databases in a destination Greenplum Database system OS Abstraction Improved handing of situations where Greenplum Database encounters segment violation errors Interconnect In some cases when the Greenplum Database query dispatcher encountered connection errors, a postmaster reset occured DDL and Utility Statements The performance of Greenplum Database truncate operations degraded between restarts of Greenplum Database Query Execution In Greenplum Database, support of subnormal double-precision (float8) numbers differed between Red Hat Enterprise Linux 5 and Red Hat Enterprise Linux 6. For example, the value 5e-309 was not handled consistently by Greenplum Database on RHEL 5 and RHEL 6. This issue has been resolved Query Planner For some SQL queries that contained a subquery, this error message was returned. ERROR: no parameter found for initplan subquery Transaction Management, Vacuum In Greenplum Database, the transaction age of the database template0 increased with no activity occurring in the database Backup and Restore If a Greenplum database was backed up and the database name contained upper-case characters, the Greenplum Database gpdbrestore utility did not restore the database with the correct name Backup and Restore In some cases, a Greenplum database could not be restored when the database was backed up with Greenplum Database gpcrondump utility that specified the back up location with the -u option. The -u option specifies the absolute path where the backup files will be placed on each Greenplum Database host. The database could not be restored if the permission of backup location was changed to read only. The Greenplum Database gpdbrestore operation failed because it could not write restore report files to the backup location. Now, gpdbrestore supports the option --report-status-dir to specify a location on each Greenplum Database host where gpdbestore writes the report files. Resolved Issues in Greenplum Database x 9

10 Table 2 Resolved Issues in x Issue Number Category Resolved in Description Backup and Restore When performing a selective restore of a partitioned table from a full backup with the Greenplum Database utility gpdbrestore, the data from leaf partitions are now restored. Previously, when performing a selective restore of a partitioned table, you needed to specify all the individual leaf partitions Loaders: Copy/External Tables Functions and Languages, Transaction Management A COPY command cancel request (Ctrl+c) followed by another COPY command and a cancel request caused the Greenplum Database session to hang. When cancel request was attempted again, a SIGSEGV error occured In PL/PGSQL functions, exception blocks were not handled properly. Depending on where the exception is encountered during function execution, the improper block handling resulted in either the creation of catalog inconsistency between master and segment, or Greenplum Database issuing the following message: The distributed transaction 'Prepare' broadcast failed to one or more segments Backup and Restore The database objects in the gp_toolkit schema were not restored after a database was re-created and then restored with the Greenplum Database gpdbrestore utility. The gp_toolkit objects are now restored when a database is re-created and restored. Resolved Issues in Greenplum Database x 10

11 Known Issues in Greenplum Database This section lists the known issues in Greenplum Database A workaround is provided where applicable. For known issues discovered in previous releases, including patch releases to Greenplum Database 4.1 or 4.0.x, see the corresponding release notes, available from EMC Support Zone: Table 3 All Known Issues in Issue Category Description gphdfs If a readable external table is created with FORMAT 'CSV' and uses the gphdfs protocol, reading a record fails if the record spans multiple lines and the record is stored in multiple HDFS blocks. Workaround: Remove line separators from within the record so that the record does not span multiple lines Client Access Using ODBC to connect to Greenplum Database on a Solaris platform when debugging is enabled in the ODBC.ini file might cause a fatal error. Workaround: Disable ODBC debugging. For example, comment out these debugging entries in the ODBC.ini file: debug=1 CommLog=1 Debug=On Loaders: External Tables Resource Management DDL and Utility Statements When a query joins an external table that uses the gpfdist protocol with a heap table, the planner might choose an incorrect plan that returns no results. Workaround: This can be avoided by running ANALYZE on the query before running the query. In Greenplum Database and later, the priority of some running queries, cannot be dynamically adjusted with the gp_adjust_priority() function. The attempt to execute this request might silently fail. The return value of the gp_adjust_priority() call indicates success or failure. If 1 is returned, the request was not successfully executed. If a number greater than 1 is returned, the request was successful. If the request fails, the priority of all running queries are unchanged, they remain as they were before the gp_adjust_priority() call. OLAP queries with a HAVING clause that filters for NULL on dimension columns incorrectly return empty results Query Planner For SQL queries of either of the following forms: SELECT columns FROM table WHERE table.column NOT IN subquery; SELECT columns FROM table WHERE table.column = ALL subquery; tuples that satisfy both of the following conditions are not included in the result set: table.column is NULL. subquery returns the empty result Query Planner Greenplum Database executes an SQL query in two stages if a scalar subquery is involved. The output of the first stage plan is fed into the second stage plan as a external parameter. If the first stage plan generates zero tuples and directly contributes to the output of the second stage plan, incorrect results might be returned. Known Issues in Greenplum Database

12 Table 3 All Known Issues in Issue Category Description Backup and Restore DDL and Utility Statements Backup and Restore DDL and Utility Statements Replication: Resync, Transaction Management When restoring sets of tables with the Greenplum Database utility gpdbrestore, the table schemas must be defined in the database. If a table s schema is not defined in the database, the table is not restored. When performing a full restore, the database schemas are created when the tables are restored. Workaround: Before restoring a set of tables, create the schemas for the tables in the database. SSL is only supported on the master host. It is not supported on segment hosts. Special characters such as!, $, #, cannot be used in the password for the Data Domain Boost user when specifying the Data Domain Boost credentials with the gpcrondump options --ddboost-host and --ddboost-user. TRUNCATE command does not remove rows from a sub-table of a partitioned table. If you specify a sub-table of a partitioned table with the TRUNCATE command, the command does not remove rows from the sub-table and its child tables. Workaround: Use the ALTER TABLE command with the TRUNCATE PARTITION clause to remove rows from the sub-table and its child tables. In some rare circumstances, performing a full recovery with gprecoverseg fails due to inconsistent LSN. Workaround: Stop and restart the database. Then perform a full recovery with gprecoverseg Interconnect After installing Greenplum Database 4.2.4, downgrading the Greenplum Database installation to a previous minor version causes a crash. Workaround: Before downgrading Greenplum Database to a previous minor version, change the value for the Greenplum Database parameter gp_interconnect_type from UDPIFC to a supported value such as UDP. The parameter value UDPIFC was introduced in Greenplum Database and is not valid in previous versions Loaders: gpload gpload fails on Windows XP with Python 2.6. Workaround: Install Python 2.5 on the system where gpload is installed Backup and Restore The gpcrondump and gpdbrestore utilities do not handle errors returned by DD Boost or Data Domain correctly. These are two examples: If invalid Data Domain credentials are specified when setting the Data Domain Boost credentials with the gpcrondump utility, the error message does not indicate that invalid credentials were specified. Restoring a Greenplum database from a Data Domain system with gpdbrestore and the --ddboost option indicates success even though segment failures occured during the restore. Workaround: The errors are logged in the master and segment server backup or restore status and report files. Scan the status and report files to check for error messages. Known Issues in Greenplum Database

13 Table 3 All Known Issues in Issue Category Description Backup and Restore Greenplum Database s implementation of RSA lock box for Data Domain Boost changes backup and restore requirements for customers running SUSE. The current implementation of the RSA lock box for Data Domain Boost login credential encryption only supports customers running on Red Hat Enterprise Linux. Workaround: If you run Greenplum Database on SUSE, use NFS as your backup solution. See the Greenplum Database System Administrator Guide for information on setting up a NFS backup Backup and Restore Backup and Restore Data Domain Boost credentials cannot be set up in some environments due to the absence of certain libraries (for example, libstdc++) expected to reside on the platform. Workaround: Install the missing libraries manually on the system. When performing a data-only restore of a particular table, it is possible to introduce data into Greenplum Database that contradicts the distribution policy of that table. In such cases, subsequent queries may return unexpected and incorrect results. To avoid this scenario, we suggest you carefully consider the table schema when performing a restore Loaders External web tables that use IPv6 addresses must include a port number Catalog and Metadata Management Scripts Suite Drop language plpgsql cascade results in a loss of gp_toolkit functionality. Workaround: Reinstall gp_toolkit. Greenplum Management utilities cannot parse IPv6 IP addresses. Workaround: Always specify IPv6 hostnames rather than IP addresses Loaders The bytenum field (byte offset in the load file where the error occurred) in the error log when using gpfdist with data in text format errors is not populated, making it difficult to find the location of an error in the source file Management Scripts Suite gpexpand --rollback fails if an error occurs during expansion such that it leaves the database down gpstart also fails as it detects that expansion is in progress and suggests to run gpexpand --rollback which will not work because the database is down. Workaround: Run gpstart -m to start the master and then run rollback, Loaders Running gpload with the --ssl option and the relative path of the source file results in an error that states the source file is missing. Workaround: Provide the full path in the yaml file or add the loaded data file to the certificate folder Loaders Unable to define external tables with fixed width format and empty line delimiter when file size is larger than gpfdist chunk (by default, 32K) Backup and Restore Backup and Restore gpdbrestore outputting incorrect non-zero error message. When performing single table restore, gpdbrestore gives warning messages about non-zero tables however prints out zero rows. NFS backup with gpcrondump -c can fail. In circumstances where you haven't backed up to a local disk before, backups to NFS using gpcrondump with the -c option can fail. On fresh systems where a backup has not been previously invoked there are no dump files to cleanup and the -c flag will have no effect. Workaround: Do not run gpcrondump with the -c option the first time a backup is invoked from a system. Known Issues in Greenplum Database

14 Table 3 All Known Issues in Issue Category Description Upgrade/ Downgrade Management Scripts Suite DDL/DML: Partitioning Management Scripts Suite Major version upgrades internally depend on the gp_toolkit system schema. The alteration or absence of this schema may cause upgrades to error out during preliminary checks. Workaround: To enable the upgrade process to proceed, you need to reinstall the gp_toolkit schema in all affected databases by applying the SQL file found here: $GPHOME/share/postgresql/gp_toolkit.sql. Running more than one gpfilespace command concurrently with itself to move either temporary files (--movetempfilespace) or transaction files (--movetransfilespace) to a new filespace can in some circumstances cause OID inconsistencies. Workaround: Do not run more than one gpfilespace command concurrently with itself. If an OID inconsistency is introduced gpfilespace --movetempfilespace or gpfilespace --movetransfilespace can be used to revert to the default filespace. ALTER TABLE ADD PARTITION inheritance issue When performing an ALTER TABLE ADD PARTITION operation, the resulting parts may not correctly inherit the storage properties of the parent table in cases such as adding a default partition or more complex subpartitioning. This issue can be avoided by explicitly dictating the storage properties during the ADD PARTITION invocation. For leaf partitions that are already afflicted, the issue can be rectified through use of EXCHANGE PARTITION. Under some circumstances, gppkg on SUSE is unable to correctly interpret error messages returned by rpm. On SUSE, gppkg is unable to operate correctly under circumstances that require a non-trivial interpretation of underlying rpm commands. This includes scenarios that result from overlapping packages, partial installs, and partial uninstalls Security A Red Hat Enterprise Linux (RHEL) 6.x security configuration file limits the number of processes that can run on gpadmin. RHEL 6.x contains a security file (/etc/security/limits.d/90-nproc.conf) that limits available processes running on gpadmin to Workaround: Remove this file or increase the processes to Installer When you run gppkg -q -info<some gppkg>, the system shows the GPDB version as main build dev Management Scripts Suite Resource Management Management Scripts Suite You may see warning messages that interfere with the operation of management scripts when logging in. Greenplum recommends that you edit the /etc/motd file and add the warning message to it. This will send the messages to are redirected to stdout and not stderr. You must encode these warning messages in UTF-8 format. Resource queue deadlocks may be encountered if a cursor is associated with a query invoking a function within another function. Filespaces are inconsistent when the Greenplum database is down. Filespaces become inconsistent in case of a network failure. Greenplum recommends that processes such as moving a filespace be done in an environment with an uninterrupted power supply Loaders: gpfdist gpfdist shows the error Address already in use after successfully binding to socket IPv6. Greenplum supports IPv4 and IPv6. However, gpfdist fails to bind to socket IPv4, and shows the message Address already in use, but binds successfully to socket IPv6. Known Issues in Greenplum Database

15 Table 3 All Known Issues in Issue Category Description Backup and Restore Backup and Restore Restoring a compressed dump with the --ddboost option displays incorrect dump parameter information. When using gpdbrestore --ddboost to restore a compressed dump, the restore parameters incorrectly show Restore compressed dump = Off. This error occurs even if gpdbrestore passes the --gp-c option to use gunzip for in-line de-compression. When running gpdbrestore with the list (-L) option, external tables do not appear; this has no functional impact on the restore job. Known Issues in Greenplum Database

16 Upgrading to Greenplum Database 4.2.x.x The upgrade path supported for this release is Greenplum Database 4.1.x.x to Greenplum Database 4.2.x.x. The minimum recommended upgrade path for this release is from Greenplum Database version If you have an earlier major version of the database, you must first upgrade to version 4.1.x. Prerequisites Before starting the upgrade process, Pivotal recommends the following: Verify the health of the Greenplum Database host hardware, and that you verify that the hosts meet the requirements for running Greenplum Database. The Greenplum Database gpcheckperf utility can assist you in confirming the host requirements. Run the gpcheckcat utility to check for Greenplum Database catalog inconsistencies. The utility is in $GPHOME/bin/lib. Pivotal recommends that Greenplum Database be in restricted mode when you run the gpcheckcat utility. See the Greenplum Database Utility Guide for information about the gpcheckcat utility. If gpcheckcat reports catalog inconsistencies, you can run gpcheckcat with the -g option to generate SQL scripts to fix the inconsistencies. After you run the SQL scripts, run gpcheckcat again. You might need to repeat the process of running gpcheckcat and creating SQL scripts to ensure that there are no inconsistencies. Pivotal recommends that the SQL scripts generated by gpcheckcat be run on a quiescent system. The utility might report false alerts if there is activity on the system. Important: If the gpcheckcat utility reports errors, but does not generate a SQL script to fix the errors, contact Pivotal support. Information for contacting Pivotal Support is at Greenplum Database Upgrade Procedures For detailed upgrade procedures and information, see the following sections: Upgrading from 4.2.x.x to x Upgrading from 4.1.x.x to 4.2.x.x For Users Running Greenplum Database 4.0.x.x For Users Running Greenplum Database 3.3.x.x Troubleshooting a Failed Upgrade Upgrading from 4.2.x.x to x An upgrade from 4.2.x.x to x involves stopping Greenplum Database, updating the Greenplum Database software binaries, and restarting Greenplum Database. Note: If the Greenplum Command Center database gpperfmon is installed in your Greenplum Database system, the migration process changes the distribution key of the Greenplum Database log_alert_* tables to the logtime column. The redistribution of the table data might take some time the first time you start Greenplum Database after migration. The change occurs only the first time you start Greenplum Database after a migration. Upgrading to Greenplum Database 4.2.x.x 16

17 1. Log in to your Greenplum Database master host as the Greenplum administrative user: $ su - gpadmin 2. Uninstall the Greenplum Database the gnet extension package if it is installed. The gnet extension package contains the software for the gphdfs protocol. For Greenplum Database and later releases, the extension is bundled with Greenplum Database. The files for gphdfs are installed in $GPHOME/lib/hadoop. 3. Perform a smart shutdown of your current Greenplum Database 4.2.x.x system (there can be no active connections to the database): $ gpstop 4. Run the installer for on the Greenplum Database master host. When prompted, choose an installation location in the same base directory as your current installation. For example: /usr/local/greenplum-db x 5. Edit the environment of the Greenplum Database superuser (gpadmin) and make sure you are sourcing the greenplum_path.sh file for the new installation. For example, change the following line in.bashrc or your chosen profile file: source /usr/local/greenplum-db-4.2.x.x/greenplum_path.sh to: source /usr/local/greenplum-db /greenplum_path.sh Or if you are sourcing a symbolic link (/usr/local/greenplum-db) in your profile files, update the link to point to the newly installed version. For example: $ rm /usr/local/greenplum-db $ ln -s /usr/local/greenplum-db /usr/local/greenplum-db 6. Source the environment file you just edited. For example: $ source ~/.bashrc 7. Run the gpseginstall utility to install the binaries on all the segment hosts specified in the hostfile. For example: $ gpseginstall -f hostfile 8. After all segment hosts have been upgraded, you can log in as the gpadmin user and restart your Greenplum Database system: $ su - gpadmin $ gpstart 9. If you are utilizing Data Domain Boost, you have to re-enter your DD Boost credentials after upgrading from Greenplum Database to as follows: gpcrondump --ddboost-host ddboost_hostname --ddboost-user ddboost_user Upgrading to Greenplum Database 4.2.x.x 17

18 Note that if you do not reenter your login credentials after an upgrade, your backup will never start because the Greenplum Database cannot connect to the Data Domain system. You will receive an error advising you to check your login credentials. Upgrading from 4.1.x.x to 4.2.x.x This section describes how you can upgrade from Greenplum Database 4.1.x.x or later to Greenplum Database 4.2.x.x. For users running versions prior to 4.1.x.x of Greenplum Database, see the following: For Users Running Greenplum Database 4.0.x.x For Users Running Greenplum Database 3.3.x.x Planning Your Upgrade Before you begin your upgrade, make sure the master and all segments (data directories and filespace) have at least 2GB of free space. Prior to upgrading your database, Greenplum recommends that you run a pre-upgrade check to verify your database is healthy. You can perform a pre-upgrade check by executing the gpmigrator (_mirror) utility with the --check-only option. For example: source $new_gphome/greenplum_path.sh; gpmigrator_mirror --check-only $old_gphome $new_gphome Note: Performing a pre-upgrade check of your database with the gpmigrator (_mirror) utility should done during a database maintenance period. When the utility checks the database catalog, users cannot access the database. Some of the rules for partitioned tables are different in 4.2 than in previous releases. gpmigrator detects partitioned tables that violate these new rules and aborts the upgrade. In most cases, gpmigrator will create a repair script you can run to bring your 4.1 Greenplum Database into line with the new rules in 4.2. See Upgrading Partitioned Tables with Constraints for more details. Upgrading Partitioned Tables with Constraints Partition tables with CHECK, PRIMARY KEY, or UNIQUE constraints must be updated prior to upgrading: Regular CHECK, PRIMARY KEY, or UNIQUE constraints added by database users usually appear on every sub-table of the partitioned table. Their names may need to be adjusted. If necessary, gpmigrator creates a repair script to do this. PRIMARY KEY and UNIQUE constraints on partitioned tables that do not include all the columns of the partition key need to be removed. If needed, gpmigrator creates a repair script to do this. Note that the unique index underlying the constraint remains and provides the same protection against duplicate keys as did the constraint. As was previously the case, it is possible for different parts of the partitioned table to contain the same key value. Upgrading to Greenplum Database 4.2.x.x 18

19 If the added constraints appear on some but not all of the sub-tables that make up a partitioned table, they cannot be updated automatically. In this case, you need to either drop the irregular constraints or add the missing constraints. Other causes of irregularity may exist but are rare. Greenplum recommends that you contact support if you encounter any issues with partitioned tables that cannot be resolved automatically with gpmigrator. Upgrade Procedure This section divides the upgrade into the following phases: pre-upgrade preparation, software installation, upgrade execution, and post-upgrade tasks. We have also provided you with an Upgrade Checklist that summarizes this procedure. Important: Carefully evaluate each section and perform all required and conditional steps. Failing to perform any of these steps can result in an aborted upgrade, placing your system in an unusable or even unrecoverable state. Pre-Upgrade Preparation (on your 4.1.x system) Perform these steps on your current 4.1.x Greenplum Database system. This procedure is performed from your Greenplum master host and should be executed by the Greenplum superuser (gpadmin). 1. Log in to the Greenplum Database master as the gpadmin user: $ su - gpadmin 2. (optional) Vacuum all databases prior to upgrade. For example: $ vacuumdb database_name 3. (optional) Clean out old server log files from your master and segment data directories. For example, to remove log files from 2011 from your segment hosts: $ gpssh -f seg_host_file -e 'rm /gpdata/*/gp*/pg_log/gpdb-2011-*.csv' Note: Running Vacuum and cleaning out old logs files is not required, but it will reduce the size of Greenplum Database files to be backed up and migrated. 4. Run gpstate to check for failed segments. $ gpstate 5. If you have failed segments, you must recover them using gprecoverseg before you can upgrade. $ gprecoverseg Note: It might be necessary to restart the database if the preferred role does not match the current role; for example, if a primary segment is acting as a mirror segment or a mirror segment is acting as a primary segment. 6. Copy or preserve any additional folders or files (such as backup folders) that you have added in the Greenplum data directories or $GPHOME directory. Only files or folders strictly related to Greenplum Database operations are preserved by the migration utility. Upgrading to Greenplum Database 4.2.x.x 19

20 Install the Greenplum Software Binaries 1. Download or copy the installer file to the Greenplum Database master host. 2. Unzip the installer file. For example: # unzip greenplum-db-4.2.x.x-platform.zip 3. Launch the installer using bash. For example: # /bin/bash greenplum-db-4.2.x.x-platform.bin 4. The installer will prompt you to accept the Greenplum Database license agreement. Type yes to accept the license agreement. 5. The installer will prompt you to provide an installation path. Press ENTER to accept the default install path (for example: /usr/local/greenplum-db-4.2.x.x), or enter an absolute path to an install location. You must have write permissions to the location you specify. 6. The installer will install the Greenplum software and create a greenplum-db symbolic link one directory level above your version-specific Greenplum installation directory. The symbolic link is used to facilitate patch maintenance and upgrades between versions. The installed location is referred to as $GPHOME. 7. Source the path file from your new 4.2.x.x installation. For example: $ source /usr/local/greenplum-db-4.2.x.x/greenplum_path.sh 8. Run the gpseginstall utility to install the x binaries on all the segment hosts specified in the hostfile. For example: $ gpseginstall -f hostfile Upgrade Execution During upgrade, all client connections to the master will be locked out. Inform all database users of the upgrade and lockout time frame. From this point onward, users should not be allowed on the system until the upgrade is complete. 9. Source the path file from your old 4.1.x.x installation. For example: $ source /usr/local/greenplum-db /greenplum_path.sh 10. (optional but strongly recommended) Back up all databases in your Greenplum Database system using gpcrondump (or zfs snapshots on Solaris systems). See the Greenplum Database Administrator Guide for more information on how to do backups using gpcrondump. Make sure to secure your backup files in a location outside of your Greenplum data directories. 11. If your system has a standby master host configured, remove the standby master from your system configuration. For example: $ gpinitstandby -r 12. Perform a clean shutdown of your current Greenplum system. For example: $ gpstop 13. Source the path file from your new 4.2.x.x installation. For example: Upgrading to Greenplum Database 4.2.x.x 20

21 $ source /usr/home/greenplum-db-4.2.x.x/greenplum_path.sh 14. Update your environment so it is sourcing your new 4.2.x.x installation. a. For example, update the greenplum-db symbolic link on the master and standby master to point to the new 4.2.x.x installation directory. For example (as root): # rm -rf /usr/local/greenplum-db # ln -s /usr/local/greenplum-db-4.2.x.x /usr/local/greenplum-db # chown -R gpadmin /usr/local/greenplum-db b. Using gpssh, also update the greenplum-db symbolic link on all of your segment hosts. For example (as root): # gpssh -f segment_hosts_file => rm -rf /usr/local/greenplum-db => ln -s /usr/local/greenplum-db-4.2.x.x /usr/local/greenplum-db => chown -R gpadmin /usr/local/greenplum-db => exit 15. (optional but recommended) Prior to running the migration, perform a pre-upgrade check to verify that your database is healthy by executing the 4.2.x.x version of the gpmigrator utility with the --check-only option. For example: # gpmigrator_mirror --check-only /usr/local/greenplum-db /usr/local/greenplum-db-4.2.x.x 16. As gpadmin, run the 4.2.x.x version of the migration utility specifying your old and new GPHOME locations. If your system has mirrors, use gpmigrator_mirror. If your system does not have mirrors, use gpmigrator. For example on a system with mirrors: $ su - gpadmin $ gpmigrator_mirror /usr/local/greenplum-db /usr/local/greenplum-db-4.2.x.x Note: If the migration does not complete successfully, contact Customer Support (see Troubleshooting a Failed Upgrade on page 24). 17. The migration can take a while to complete. After the migration utility has completed successfully, the Greenplum Database 4.2.x.x system will be running and accepting connections. Note: After the migration utility has completed, the resynchronization of the mirror segments with the primary segments continues. Even though the system is running, the mirrors are not active until the resynchronization is complete. Post-Upgrade (on your 4.2.x.x system) 18. If your system had a standby master host configured, reinitialize your standby master using gpinitstandby: $ gpinitstandby -s standby_hostname Upgrading to Greenplum Database 4.2.x.x 21