Issue

Component

Description

Issue

Component

Description

Issue

Component

Description

Previously, RUN_INDEX_TOOL analyzed all projections, and executed on only one thread. In addition to general performance improvements, the meta-function now supports two new parameters that can significantly improve Index tool performance:

• Narrow the scope of the operation to one or more projections.
• Specify the number of threads used to execute the function.

Issue

Component

Description

Issue

Component

Description

Hive file formats partition pruning requires all partition directory names to be in lower-case. Previously, Export to Parquet sometimes generated partition directory names in mixed-case. This caused the partition pruning optimization to be skipped on a load.

This issue has been fixed.

License usage was showing 0% when Autopass license is used.

This issue has been fixed. Vertica server hotfix upgrade is also required for this fix to take effect (same release number).

In rare circumstances, sessions using the KafkaSource COPY command could hang during connection tear down. This issue has been fixed by upgrading the underlying Kafka client library, rdkafka, from 0.9 to 0.11.

This upgraded library contains protocol support for newer versions of Kafka. When communicating with a Kafka cluster of version 0.10 or newer, Vertica's Kafka interactions can take advantage of newer protocols and optimizations. To do so, set api.version.request=true in your rdkafka settings.

In addition, the message.max.bytes rdkafka parameter has changed from representing the maximum size of a single message to the maximum size of a batch of messages, for compatibility with newer versions of Kafka. Applications that set this parameter may need to increase the value to reflect the new protocol semantics.

Both parameters can be changed via the kafka_conf argument of the Kafka UDX functions, or via uds_kv_parameters load-spec option of vkconfig. For additional details, consult the documentation.

Issue

Component

Description

If vertica.conf set a parameter to a value that started with hash (#), Vertica ignored that parameter and logged a warning.

This issue has been fixed.

Adding a large number of nodes in a single operation could lead to an admintools error parsing output.

This issues has been fixed.

Having frequent create/insert/drop tables caused a memory leak in the catalog.

This issue has been fixed.

Issue

Component

Description

Very large expressions could run out of stack and crash the node.

This issue has been fixed.

The Database Designer wizard in Management Console did not correctly set the default value for design k-safety.

This issues has been fixed.

If you ran APPROXIMATE_COUNT_DISTINCT_SYNOPSIS on a database table that contained NULL values, the synopsis object that it returned sometimes was larger than the one it returned after the NULL values were removed.

This issue has been resolved.

In some regular expression scalar functions, Vertica would crash for certain long input strings.

This issue has been fixed.

Issue

Component

Description

Vertica placed an exclusive lock on the global catalog while it created a query plan for CREATE TABLE AS <query>.

In this case, very large queries could prolong this lock until it eventually timed out. On rare occasions, the prolonged lock caused an out-of-memory exception that shut down the cluster.

This issue has been fixed.

If a control node and one of its child nodes went down, attempts to restart the second (child) node sometimes failed.

This issue has been fixed.

Add Column Not Null gets transformed into two different statements: Alter Table Add Column and Alter Table Add Constraint.

Alter Table Add Constraint could fail and result in a new column in your table, but without the constraint.

It also produced a confusing rollback message that did not indicate that the column was added.

This issue has been fixed.

Heavy workloads may have caused the Tuple Mover to perform operations for an extended period of time,

This led to memory accumulation before the Tuple  Mover exited its session and released the memory .

This issue has been fixed.

When a projection was created with only the "PINNED" keyword, Vertica incorrectly considered it a segmented projection. This caused optimizer internal errors and incorrect results when loading data into tables with these projections.

This issued has been fixed.

IMPORTANT: The fix only applies to newly created pinned projections. Existing pinned projection in the catalog are still incorrect and need to be dropped and recreated manually.

The SKIP keyword of a COPY statement was not properly supported with the FIXEDWIDTH data format.

This issue has been fixed.

Previously, query label only applied to CTAS statements, but was missing in the implicit INSERT SELECT statement.

With this fix, query label will be applied to both statements.

Issue

Component

Description

Long queries were failing in the MC 'Query Execution' page.

This issue was resolved.

Issue

Component

Description

On some Debian 7 systems, the vertica agent process would not start during installation.

This issue has been fixed.

If a Parquet file metadata is very large, Vertica can consume more memory than reserved and crash when the system runs out of memory.

This issue has been fixed.

Occasionally the EXPORT TO PARQUET command will write data to incorrect partition directories of types timestamp, date, and time.

This issue has been fixed.

Issue

Component

Description

Occasionally, a copy or external table query could crash a node.

This has been fixed.

The error dialog was displayed on the Load page 'Instance' tab when a longer time interval was selected.

This issue has been fixed.

Issue

Component

Description

Optimized MERGE was sometimes not used with NULL constants in varchar or numeric columns.

This issue has been fixed.

An Internal Error would sometimes occur in queries with subqueries containing a mix of outer and inner joins.

This issue has been fixed.

In some workloads, the memory consumed by a long-running session could grow over time.

This issue has been fixed.

Including certain functions like nvl in a grouping function argument resulted in a "Grouping function arguments need to be group by expressions" error.

This issue has been fixed.

Issue

Component

Description

Scrutinize was collecting the logs only on the node where it was executed, not on any other nodes.

This issue has been fixed.

COPY statements using the default parser could erroneously reject rows if the input data contained overflowing floating point values in one column and the maximum 64-bit integer value in another column.

This issue has been fixed.

Applying an access policy on some columns in a table significantly impacted execution of updates on that table, even where the query excluded the restricted columns. Further investigation determined that the query planner materialized unused columns, which adversely affected overall performance.

This issue has been fixed.

Some versions of Vertica allowed projections to be defined with a subquery that referenced a view or another projection. These malformed projections occasionally caused the database to shut down.

Vertica now detects these errors when it parses the DDL and returns with a message that projection FROM clauses must only reference tables.

Loads and copies from HDFS occasionally failed if Vertica was reading data through libhdfs++ when the NameNode abruptly disconnected from the network. libhdfs++ was being compiled with a debug flag on, which caused network errors to be caught in debug assertions rather than normal error handling and retry logic.

This issue has been fixed.

The ORC Parser skipped rows when the predicate "IS NULL" was applied on a column containing all NULLs.

This issue has been fixed.

INSERT DIRECT failed when the configuration parameter ReuseDataConnections was set to 0 (disabled). When this parameter was disabled, no buffer was allocated to accept inserted data.

Now, a buffer is always allocated to accept data regardless of how ReuseDataConnections is set.

Object replication and restore left behind temporary files in the database Snapshot folder.

This fix ensures such files are properly cleaned-up when the operation completes.

Sometimes, adding many nodes in a single operation would fail with a message about invalid control characters.

This issue has been fixed.

Issue

Component

Description

The way null values were detected for float data type columns led to inconsistent results for some queries with median and percentile functions.

This issue has been fixed.

Using an ML function, which accepts a model as a parameter in the definition of a view, and then feeding that view into another ML function as its input relation caused failures in some special cases.

This issue has been fixed.

Attempts to swap partitions between flattened tables that were created in different versions of Vertica failed, due to minor differences in how source and target SET USING columns were internally defined.

This issue has been fixed.

Dependent index tables did not have their reference indices updated when the base table's columns were altered but not removed. This caused an internal error on reference.

This issue has been fixed so the indices are updated when you alter the base table's columns.

Issue

Component

Description

Some instances of type casting on a datetime column that contained a value of 'infinity' triggered segmentation faults that caused the database to fail.

This issue has been fixed

If you tried to insert data that was too wide for a VARCHAR or VARBINARY column, the error message did not specify the column name.

This error message now includes the column name.

Export to Parquet using the local disk non-deterministically failed in some situations with a 'No such file or directory' error message.

This issue has been fixed.

Issue

Component

Description

Vertica did not previously release the catalog memory for objects that were no longer visible by any current or subsequent transactions.

The Vertica garbage collector algorithm has been improved.

Issue

Component

Description

Running a large query with several analytic functions returned the wrong results.

This issue has been fixed.

For lengthy string values, the hash value computed by the JDBC implementation differed from the hash() function on the Vertica server.

This issue has been fixed.

A non-DBADMIN user could not launch the Kafka scheduler.

This issue has been fixed.

Scrutinize was collecting the logs only on the node where it was executed, not on any other nodes.

This issue has been fixed.

When using ALTER NODE to change the MaxClientSessions parameter, the node's state changes from Standby or Ephemeral to Permanent.

This issue has been fixed.

Issue

Component

Description

An issue with VStream in the Vertica client resulted in a Socket connection error handling error message and caused the client to crash. This issue has been fixed.

Backing up on S3 failed when both of the following occurred:

1. when you backup to the root of the S3 bucket, AND

2. when the backup location reaches the restorePointLimit.

This issue has been fixed.

Issue

Component

Description

Upgrading a cluster from 8.1.1-4 to 9.0.1-2 no errors were reported, but the database would not start after the upgrade.

The Compute Cluster upgrade change task kept looping and rolled back with an Invalid model name error.

This issue has been fixed.

Issue

Component

Description

The default value of MaxParsedQuerySizeMB has changed. The original default was 512MB. This only bound a certain amount of "used" parse memory. The default is now 1024MB. This results in bounding all parse memory.

There may be some queries that used to be able to run successfully, that will now encounter the Request size too big. Please try to simplify the query error. This is not a regression. To successfully run the query, increase the value of the knob and reset the session.

After renaming a table and changing the schema to which table belongs, the TABLE_NAME field in TABLE_CONSTRAINTS system table was not updated. This issue has been fixed.

Issue

Component

Description

After a user connected to HDFS using the Vertica realm, users from other realms could not connect to HDFS. This behavior has been corrected.

There was an internal issue with AnalyzeRowCount causing a node to crash.

To work around this issue, disable AnalyzeRowCount using the following command:

select disable_service('TM','AnalyzeRowCount');

Dropping an Identity column resulted in loss of all the data in the table.

This issue has been fixed.

Issue

Component

Description

At times Vertica crashed when running queries with both of the following:

• multiple percentile_cont, percentile_disc, or median functions with similar partition by clauses AND
• a limit clause

This issue has been fixed.

The functions APPROXIMATE_COUNT_DISTINCT_OF_SYNOPSIS and APPROXIMATE_COUNT_DISTINCT_SYNOPSIS are not back compatible with 7.x.

We provide a workaround to make APPROXIMATE_COUNT_DISTINCT_OF_SYNOPSIS and APPROXIMATE_COUNT_DISTINCT_SYNOPSIS perform the same as 7.x as a choice.

Workaround: Make ACD perform the same as 7.x in 9.0SP1:

1. Drop the ACD library by running /opt/vertica/packages/approximate/ddl/uninstall.sql.

2. Install the deprecated ACD library by running approximate/ddl/install_deprecated.sql.

Switch back to using the new ACD in 9.0SP1:

1. Drop the deprecated ACD library by running approximate/ddl/uninstall.sql.

2. Install the new ACD library by running approximate/ddl/install.sql.

Issue

Component

Description

Previously, querying the projections_column table sometimes incorrectly displayed the most recent analyze event as FULL if the rowcount of the table in question changed.

Querying the projections_column table now correctly displays the most recent analyze event as ROWCOUNT after a change in table rowcount.

Previously, the network port connection used by the Management Console to communicate with nodes accepted the older DES and 3DES ciphers, which have a known vulnerability. This connection no longer accepts connections that use these ciphers.

Issue

Component

Description

After being granted usage and select privileges on a schema and table in a specific resource pool, the user was denied permission when trying to access the schema and table. This was because Vertica was checking permissions on secondary resource pools.

This issue is resolved and Vertica no longer check permissions on secondary resource pools.

Modifying roles in Vertica, for example SET ROLE DEFAULT, generated an error. This issue has been resolved.

Upgrading from Vertica 7.2 to 8.1.1-2 caused a loss of historical data due to changes in data format of synopsis objects.

Use AGGREGATE FUNCTION "public.ApproxCountDistinctOfSynopsisDeprecated" to process existing synopsis objects.

Issue

Component

Description

CREATE VIEW statements sometimes hung and caused the server to fail if the view was selecting from HDFS sources. This issue has been resolved.

This fix adds SSL support for HCatalogConnector with HiveServer2.

Issue

Component

Description

When using HCatalog connector, non-superusers could not read from an HCatalog schema despite having usage granted.

In rare circumstances, if multiple users attempted to concurrently authenticate with HDFS, an unnecessary assert caused core dump.

Issue

Component

Description

Upgrading from 7.2.3 to Vertica 8.0 or above can cause naming conflicts due to new UDx names introduced in for Machine Learning in Vertica 8.0.

Two types of conflicts can occur:

a. If a 7.2.3 UDx name conflicts with UDx's introduced in the Vertica 8.0 Machine Learning Package, the Machine Learning Package installation fails when you start the database for the first time after upgrading Vertica. In this case, you can find the reason for failure and name conflict details in the AdminTools log file.

b. If a 7.2.3 UDx name conflicts with Vertica Machine Learning functions introduced in Vertica 8.0, no failure occurs. However, after upgrading, the Vertica 7.2.3 UDx's are not available for execution due to the naming conflict. Vertica 8.0 introduced the following functions:

• KMEANS

• LINEAR_REG

• LOGISTIC_REG

• BALANCE

• DELETE_MODEL

• SUMMARIZE_MODEL

• RENAME_MODEL

• NORMALIZE

Workaround: In both cases, rename the UDx causing the conflict and retry the installation.

After installing the Vertica ODBC version 8.1.1 client on a Windows 7 pre-SP1 operating system, an attempt to connect using the ODBC DSN fails with the message, “The specified module could not be found. (Vertica, C:\Program Files\Vertica Systems\ODBC64\lib\vertica_8.1_odbc_3.5.dll)”.

Workaround: To resolve this issue, Vertica recommends you upgrade to Windows 7 Service Pack 1.

Be aware that if Vertica cancels a query that generated an error, Vertica sometimes additionally generates the error "Send: Connection not open" during the cancel, even though that is not the cause of the original error.

After a user authenticates on a node using the Vertica realm, users from other realms cannot log in on that node. Other nodes continue to work until a user in the Vertica realm authenticates on them.

Workaround: To get a clean state, restart the database.

The HCatalog Connector produces an error if users kinit using a realm other than the default_realm as defined in the system krb5.conf.

Workaround:

Set the "default_realm" inside krb5.conf to be the realm of the connecting user. Take care in changing this default, as it affects all of Vertica and not just the HCatalog Connector.

After a user authenticates on a node using the Vertica realm, users from other realms cannot log in on that node. Other nodes continue to work until a user in the Vertica realm authenticates on them.

Workaround:

To get a clean state, restart the database.

If all partitions are moved from a table while a node is down and then a new projection is created on that table, the epoch for the new projection will be more recent than the epoch for the MOVE PARTITION event. In that rare case, the down node fails to recover. This problem also occurs if, when the node is down, Vertica swaps out all partitions, but no partitions have been swapped in.

Workaround: Find the newly created projection in the PROJECTION_CHECKPOINT_EPOCHS system table. Drop that projection and restart the down node.

After a user authenticates on a node using the Vertica realm, users from other realms cannot log in on that node. Other nodes continue to work until a user in the Vertica realm authenticates on them.

Workaround: To get a clean state, restart the database.