Known Issues for Adaptive Server Cluster Edition

Bootstrapping COTS JVM may result in a SIGSEGV.

When Adaptive Server attempts to bootstrap COTS JVM, you may encounter a signal 11 followed by a stacktrace in the error log which includes the module jvmh_utl_printn. This will result in Java operations not being able to execute.

Workaround: To avoid the issue, change the global java.policy file by adding the following rule and changing the first string starting after file: with your absolute path.

grant codeBase "file:/absolute_path_to/sybase/ASE-15_0/lib/pca/sybasert.jar" { permission java.lang.reflect.ReflectPermission "suppressAccessChecks"; };

For example,

grant codeBase "file:/opt/sybase/ASE-15_0/lib/pca/sybasert.jar" { permission java.lang.reflect.Reflect-Permission "suppressAccessChecks"; };

For each platform, the java.policy file is located here:

• IBM - $SYBASE/shared/JRE-6_0_SR9FP1_64BIT/lib/security/java.policy
• HP - $SYBASE/shared/JRE-6_0_10/lib/security/java.policy
• Linux - $SYBASE/shared/JRE-6_0_24_64BIT/lib/security/java.policy
• Solaris - $SYBASE/shared/JRE-6_0_24/lib/security/java.policy

Segmentation violation when decreasing number of engines (thread count) using the alter thread pool command.

Under certain circumstances, Adapter Server may encounter a segmentation violation followed by a stacktrace involving the function uppwakeup() when the number of engines is decreased via the alter thread pool command.

Workaround: To avoid the segmentation violation, do not decrease the number of engines (thread count).

In some Cluster Edition configurations, Adaptive Server 624 and 69x errors may occur in rare circumstances during database replication of a database or replication of tables within a database.

In active-active or active-passive cluster configurations, you can configure RepAgent to run on only one node of the cluster. These errors may occur in rare circumstances when RepAgent reads an older image of database log pages that is on disk while the latest image is on another node.

Workaround: Sybase recommends that for:

• Active-passive configuration – run RepAgent on the active node to ensure that data and log pages are on the same node as RepAgent.
• Active-active configuration – start Adaptive Server with the 16872 trace flag to prevent log page corruption and 69x errors. However, using this flag degrades the server performance.

This message is reported by Open Client applications connecting to Adaptive Server Enterprise:

ct_send(): network packet layer: internal Client Library error: State error: trying to write when connection is expecting a read.

This message is reported in some cases when Kerberos message confidentiality service is turned on.

Workaround: Reconnect to the server with confidentiality service turned off.

The following message is reported in the error log regarding thread pools. "kernel Expected object passed to kmuxtskGetThreadpool is not of type Multiplexed Task, 0x4d858884 != 0x52544354 (TCTR)"

Workaround: You can safely ignore this message.

The sybcluster localize utility does not function properly and exceptions are raised.

If the $SYBASE/SYBASE_ASE/locales directory contains files or directories that are not server-related locales files, the sybcluster localize utility does not function properly. If unrelated files are in the locales directory, the localize command aborts and raises an exception at the beginning phase. If an unrelated directory is in the locales directory, the charset selection menu is shown incorrectly and localize may abort during further processing.

Workaround: Do not place files or directories unrelated to localization in the $SYBASE/SYBASE_ASE/locales directory.

Adaptive Server 624 or 69x errors may occur in a rare circumstances when RepAgent reads an older image of the database log pages from disk, while the latest image resides in memory of the remote cluster node.

You may encounter errors in these scenarios:

Active - Active setup: There is more than one node on the cluster which sends insert/update/delete operations on the database. This results in logs generated for a database on more than one node.

Active - Passive setup: There is only one node on the cluster which sends insert/update/delete operations on the database. This node is considered an active node. If RepAgent is configured on a passive node, the passive node will only have read operations.

Workaround: Active - Active setup: In order to configure RepAgent and prevent 624 or 69x errors, use traceflag 16872. Using this traceflag will degrade the performance of the servers in the active-active scenario.

Active - Passive: Configure RepAgent on the active node.

The LogicalReads columns in the MDA tables is listed as 0.

Workaround: Turn on statement statistics active to use LogicalReads for the above MDA tables.

Using I/O fencing via SCSI-3 PGR when using tbe AIX DKPRES API with disk reserve_policy set to PR_shared, may allow any remote application to open the disk which automatically creates and removes registrations on the device.

AIX supports I/O fencing via SCSI-3 PGR in two ways: AIX DKPRES API or raw SCSI-3 PGR API.

When the disk reserve_policy is set to no_reserve, Adaptive Server uses raw SCSI-3 PGR API. In this configuration, remote applications cannot inadvertently create and remove registrations on disks.

When the disk reserve_policy is set to PR_shared, Adaptive Server uses AIX DKPRES API.The PR_shared setting may allow any remote application, for example dd, to open the disk, which automatically creates and removes registrations on the device. This can make the database devices susceptible to data corruption. syb_cluster/UAF are known to open database disks and cause this problem, and therefore should not be left running when using AIX DKPRES.

Workaround: The preferred method is raw SCSI-3 PGR API, however, some IBM storage devices like DS5800 do not support raw SCSI-3 PGR. First try setting the reserve_policy to no_reserve. If the ASE instance fails to start up due to I/O fencing issues, then the device does not support raw SCSI-3 PGR API. In such cases, configured the devices to use DKPRES API, so set the reserve_policy to PR_shared.

The select for update semantics introduced in version 15.7 do not support datapages-locked tables.

select for update support exists only for datarows-locked pages.

Workaround: None.

If enable DTM is set, the distributed recovery feature is disabled and all databases are recovered only on the cluster coordinator, rather than on different instances of the cluster.

Workaround: None.

There are two known issues with transfer table:

• When creating a unique index that is using the ignore_dup_key property, importing a row with a key that is already present in the table leads to an error, and the import is aborted. This differs from when a duplicate key is inserted with bcp or with a regular insert statement, since in those cases, the row is discarded, and the transaction continues.
• When an insert trigger exists, the trigger is not fired when data is inserted through transfer table...from.

Workaround: None.

Incorrect cluster failover may occur during diagnostic shared-memory dump.

If you start an instance while the rest of the cluster is performing a diagnostic shared memory dump, the instance that is starting may incorrectly perform a cluster takeover. This occurs only if automatic cluster takeover is set to 1, or if the --cluster_takeover option is passed to the data server. In environments without i/o fencing enabled, this may lead to data corruption.

Workaround: Do not start an instance during a diagnostic shared-memory dump. Configure automatic cluster takeover to 0.

Configuring Job Scheduler Self Management for Simplified Chinese.

When you configure the Job Scheduler Self Management feature for an Adaptive Server localized for Simplified Chinese, you may receive an error saying that a varchar datatype cannot be converted to a date. The Job Scheduler configuration process stops at this point.

Workaround: Do not configure the Self Management feature for Simplified Chinese.

Creating Job Scheduler schedules for Adaptive Server with Simplified Chinese.

When you attempt to create a Job Scheduler schedule using the Adaptive Server plug-in for an Adaptive Server that is localized for Simplified Chinese, you may receive an error indicating that the varchar datatype cannot be converted to a date.

Workaround: Localize the Adaptive Server for Simplified Chinese before configuring Job Scheduler.