Oracle PCA is delivered as an appliance: a complete and controlled system composed of selected hardware and software components. If you install additional software packages on the pre-configured appliance components, be it a compute node, management node or storage component, you introduce new variables that potentially disrupt the operation of the appliance as a whole. Unless otherwise instructed, Oracle advises against the installation of additional packages, either from a third party or from Oracle's own software channels like the Oracle Linux YUM repositories.

Workaround: Do not install additional software on any internal Oracle PCA system components. If your internal processes require certain additional tools, contact your Oracle representative to discuss these requirements.

In the Oracle PCA Dashboard, the Network Setup tab becomes available when the first compute node has been provisioned successfully. However, when installing and provisioning a new system, you must wait until all nodes have completed the provisioning process before changing the network configuration. Also, when provisioning new nodes at a later time, or when upgrading the environment, do not apply a new network configuration before all operations have completed. Failure to follow these guidelines is likely to leave your environment in an indeterminate state.

Workaround: Before reconfiguring the system network settings, make sure that no provisioning or upgrade processes are running.

Bug 17475738

External time synchronization, based on ntpd , is left in default configuration at the factory. As a result, NTP does not work when you first power on the Oracle PCA, and you may find messages in system logs similar to these:

Oct  1 11:20:33 ovcamn06r1 kernel: o2dlm: Joining domain ovca ( 0 1 ) 2 nodes
Oct  1 11:20:53 ovcamn06r1 ntpd_initres[3478]: host name not found:0.rhel.pool.ntp.org
Oct  1 11:20:58 ovcamn06r1 ntpd_initres[3478]: host name not found:1.rhel.pool.ntp.org
Oct  1 11:21:03 ovcamn06r1 ntpd_initres[3478]: host name not found:2.rhel.pool.ntp.org

Workaround: Apply the appropriate network configuration for your data center environment, as described in the section Network Setup in the Oracle Private Cloud Appliance Administrator's Guide. When the data center network configuration is applied successfully, the default values for NTP configuration are overwritten and components will synchronize their clocks with the source you entered.

Bug 17548941

Towards the end of the management node install.log file, the following warnings appear:

> WARNING:
> /lib/modules/2.6.39-300.32.1.el6uek.x86_64/kernel/drivers/infiniband/ \
> hw/ipath/ib_ipath.ko needs unknown symbol ib_wq
> WARNING:
> /lib/modules/2.6.39-300.32.1.el6uek.x86_64/kernel/drivers/infiniband/ \
> hw/qib/ib_qib.ko needs unknown symbol ib_wq
> WARNING:
> /lib/modules/2.6.39-300.32.1.el6uek.x86_64/kernel/drivers/infiniband/ \
> ulp/srp/ib_srp.ko needs unknown symbol ib_wq
> *** FINISHED INSTALLING PACKAGES ***

These warnings have no adverse effects and may be disregarded.

Bug 16946511

The role of the Node Manager database is to track the various states a compute node goes through during provisioning. After successful provisioning the database continues to list a node as running, even if it is shut down. For nodes that are fully operational, the server status is tracked by Oracle VM Manager. However, the Oracle PCA Dashboard displays status information from the Node Manager. This may lead to inconsistent information between the Dashboard and Oracle VM Manager, but it is not considered a bug.

Workaround: To verify the status of operational compute nodes, use the Oracle VM Manager user interface.

Bug 17456373

The purpose of a lock of the type provisioning or all_provisioning is to prevent all compute nodes from starting or continuing a provisioning process. However, when you attempt to reprovision a running compute node from the Oracle PCA CLI while an active lock is in place, the compute node state changes to "reprovision_only" and it is marked as "DEAD". Provisioning of the compute node continues as normal when the provisioning lock is deactivated.

Bug 22151616

Compute node provisioning can take up to several hours to complete. However, those nodes are added to the Oracle VM server pool early on in the process, but they are not placed in maintenance mode. In theory the discovered servers are available for use in Oracle VM Manager, but you must not attempt to alter their configuration in any way before the Oracle PCA Dashboard indicates that provisioning has completed.

Workaround: Wait for compute node provisioning to finish. Do not modify the compute nodes or server pool in any way in Oracle VM Manager.

Bug 22159111

Reprovisioning or upgrading a compute node that hosts virtual machines (VMs) is considered bad practice. Good practice is to migrate all VMs away from the compute node before starting a reprovisioning operation or software update. At the start of the reprovisioning, the removal of the compute node from its server pool could fail partially, due to the presence of configured VMs that are either running or powered off. When the compute node returns to normal operation after reprovisioning, it could report failures related to server pool configuration and storage layer operations. As a result, both the compute node and its remaining VMs could end up in an error state. There is no straightforward recovery procedure.

Workaround: Avoid upgrade and reprovisioning issues due to existing VM configurations by migrating all VMs away from their host first.

Bug 23563071

Using the Oracle PCA CLI it is possible to force the reprovisioning of a compute node even if it is hosting running virtual machines. The compute node is not placed in maintenance mode when running Oracle VM Server 3.2.11. Consequently, the active virtual machines are not shut down or migrated to another compute node. Instead these VMs remain in running status and Oracle VM Manager reports their host compute node as "N/A".

Workaround: In this particular condition the VMs can no longer be migrated. They must be killed and restarted. After a successful restart they return to normal operation on a different host compute node in accordance with start policy defined for the server pool.

Bug 22018046

As the Oracle VM environment grows and contains more and more virtual machines and many different VLANs connecting them, the number of management operations and registered events increases rapidly. In a system with this much activity the provisioning of a compute node takes significantly longer, because the provisioning tasks run through the same management node where Oracle VM Manager is active. There is no impact on functionality, but the provisioning tasks can take several hours to complete.

There is no workaround to speed up the provisioning of a compute node when the entire system is under heavy load. It is recommended to perform compute node provisioning at a time when system activity is at its lowest.

Bug 22159038 and 22085580

The host network is a custom network type that enables connectivity between the physical Oracle PCA hosts and external network resources. As part of the host network configuration, a static route is configured on each server participating in the network. However, the required static route can only be configured if the server in question has been upgraded to the version of Oracle VM Server included in Release 2.2.1 of the Oracle Private Cloud Appliance Controller Software. If a host is running a previous version its routing table is not updated.

Workaround: If you intend to use a host network in your environment, make sure that the compute nodes are running the correct version of Oracle VM Server, as included in the Release 2.2.1 ISO image of the Oracle PCA Controller Software.

Bug 23182978 and 23233700

When you create a custom network, it is technically possible – though not supported – to alter the VLAN tag in Oracle VM Manager. However, when you attempt to add a compute node, the system creates the network bond on the server but fails to enable the modified VLAN configuration. At this point the custom network is stuck in a failed state: neither the network nor the vNIC bond can be deleted, and the VLAN configuration can no longer be changed back to the original tag.

Workaround: Do not modify appliance-level networking in Oracle VM Manager. There are no documented workarounds and any recovery operation is likely to require significant downtime of the Oracle PCA environment.

Bug 23250544

Compute nodes currently support a maximum of 36 vNICs, of which 6 are used by the default network configuration. In theory, this allows for 15 more custom network bonds of 2 vNICs each to be created. However, the maximum allowed is 3 internal custom networks and 7 external custom networks, which is equivalent to 10 network bonds. You should not configure any vNICs beyond these maximums, even if the system allows you to.

Workaround: When configuring custom networking, always adhere to the limitations set forth in Chapter 3, Configuration Maximums.

Bug 24407432

The Oracle PCA Dashboard cannot be used to perform an update of the software stack.

Workaround: Use the command line tool pca-updater to update the software stack of your Oracle PCA. For details, refer to the section Oracle Private Cloud Appliance Software Update in the Oracle Private Cloud Appliance Administrator's Guide. For step-by-step instructions, refer to the section Update. You can use SSH to log in to each management node and check /etc/pca-info for log entries indicating restarted services and new software revisions.

Bug 17476010, 17475976 and 17475845

The first step of the software update process is to download an image file, which is unpacked in a particular location on the ZFS storage appliance. When the download is interrupted, the file system is not cleaned up or rolled back to a previous state. As a result, contents from different versions of the software image may end up in the source location from where the installation files are loaded. In addition, the downloaded *.iso file remains stored in /tmp and is not unmounted. If downloads are frequently started and stopped, this could cause the system to run out of free loop devices to mount the *.iso files, or even to run out of free space.

Workaround: The files left behind by previous downloads do not prevent you from running the update procedure again and restarting the download. Download a new software update image. When it completes successfully you can install the new version of the software, as described in the section Update in the Oracle Private Cloud Appliance Administrator's Guide.

Bug 18352512

When the Oracle PCA Controller Software update includes a firmware upgrade of the internal ZFS storage appliance, the automated upgrade process requires full exclusive control of the storage appliance ILOM. If another user is logged onto the ILOM, the firmware upgrade fails. As a result, the Controller Software update cannot continue.

Workaround: Make sure that no other user is logged onto the storage appliance ILOM and restart the software update with the pca-updater command. Proceed as follows:

Bug 23149946

Several iSCSI LUNs, including the essential server pool file system, are mapped on each compute node. When you update the appliance software, it may occur that one or more LUNs are missing on certain compute nodes. In addition, there may be problems with the configuration of the clustered server pool, preventing the existing compute nodes from joining the pool and resuming correct operation after the software update.

Workaround: To avoid these software update issues, upgrade all previously provisioned compute nodes by following the procedure described in the section Upgrading Existing Compute Node Configuration from Release 1.0.2 in the Oracle Private Cloud Appliance Administrator's Guide.

Bugs 17922555, 18459090, 18433922 and 18397780

When adding LUNs on the Oracle PCA internal ZFS Storage Appliance you must add them under the "OVM" target group, because this target groups exists by default and there can be only one. However, if you continue to use default settings the new LUNs are mapped to the "All Initiators" group. This means that the LUNs are mapped to all nodes in the system, and this causes several problems inside the appliance rack. Instead, LUNs must be associated with a different initiator group so that the appliance software can map them correctly at initial setup or during a software update.

Workaround: When creating additional LUNs on the internal ZFS Storage Appliance, do not map to the default All Initiators group but use a separate one instead.

Bugs 22309236 and 18155778

When a failover occurs between the storage heads of the Oracle PCA internal ZFS storage appliance, or an externally connected ZFS storage appliance, the file systems used by virtual machines may become read-only, preventing normal VM operation. Compute nodes may also hang or crash as a result.

Workaround: There is no documented workaround to prevent the issue. Once the storage head failover has completed, you can reboot the virtual machines to bring them back online in read-write mode.

Bugs 19324312 and 19670873

During the Oracle PCA software update from Release 1.0.2 to Release 1.1.x, it may occur that the specific tuning settings for Oracle VM Manager are not applied correctly, and that default settings are used instead.

Workaround: Verify the Oracle VM Manager tuning settings and re-apply them if necessary. Follow the instructions in the section Verifying and Re-applying Oracle VM Manager Tuning after Software Update in the Oracle Private Cloud Appliance Administrator's Guide.

Bug 18477228

If you have changed the password for Oracle VM Manager or its related components Oracle WebLogic Server and Oracle MySQL database, and you need to restore the Oracle VM Manager from a backup that was made prior to the password change, the passwords will be out of sync. As a result of this password mismatch, Oracle VM Manager cannot connect to its database and cannot be started.

Workaround: Follow the instructions in the section Restoring a Backup After a Password Change in the Oracle Private Cloud Appliance Administrator's Guide.

Bug 19333583

When logging in to the active management node using SSH, you typically use the virtual IP address shared between both management nodes. However, since they are separate physical hosts, they have a different host key. If the host key is stored in the SSH client, and a failover to the secondary management node occurs, the next attempt to create an SSH connection through the virtual IP address results in a host key verification failure.

Workaround: Do not store the host key in the SSH client. If the key has been stored, remove it from the client's file system; typically inside the user directory in .ssh/known_hosts.

Bug 22915408

Particularly in environments with a large number of virtual machines, and when many virtual machine operations – such as start, stop, save, restore or migrate – occur in a short time, the Java processes of Oracle VM may consume a lot of CPU and memory capacity on the master management node. Users will notice the browser and command line interfaces becoming very slow or unresponsive. This behavior is likely caused by a memory leak in the Oracle VM CLI.

Workaround: A possible remedy is to restart the Oracle VM CLI from the Oracle Linux shell on the master management node.

# /u01/app/oracle/ovm-manager-3/ovm_cli/bin/stopCLIMain.sh
# nohup /u01/app/oracle/ovm-manager-3/ovm_cli/bin/startCLIMain.sh&

Bug 18965916

The default compute node configuration does not allow connectivity to additional storage resources in the data center network. Compute nodes are connected to the data center subnet to enable public connectivity for the virtual machines they host, but the compute nodes' physical network interfaces have no IP address in that subnet. Consequently, SAN or file server discovery will fail.

Bug 17508885

Fibre Channel LUNs should only be presented to compute nodes. Presenting the LUNs to the management nodes can cause their kernel to panic. Use proper (soft) zoning on the FC switch to prevent the management nodes from accessing the LUNs. For details, refer to the support note with Doc ID 2148589.1.

Bug 22983265

When network throughput is very high, certain conditions, like a large number of MTU 9000 streams, have been known to cause a kernel panic in a compute node. In that case, /var/log/messages on the affected compute node contains entries like "Task Python:xxxxx blocked for more than 120 seconds". As a result, HA virtual machines may not have been migrated in time to another compute node. Usually compute nodes return to their normal operation automatically.

Workaround: If HA virtual machines have not been live-migrated off the affected compute node, log into Oracle VM Manager and restart the virtual machines manually. If an affected compute node does not return to normal operation, restart it from Oracle VM Manager.

Bugs 20981004 and 21841578

Before the product name change from Oracle Virtual Compute Appliance to Oracle Private Cloud Appliance, the Oracle PCA Dashboard could be accessed at https://<manager-vip>:7002/ovca. As of Release 2.0.5, the URL ends in /dashboard instead. However, there is no redirect from /ovca to /dashboard.

Workaround: Enter the correct URL: https://<manager-vip>:7002/dashboard.

Bug 21199163

When you activate the Screen Reader, through the Accessibility Options at login or in the Settings toolbar, the labels on the network ports of the I/O modules in the Network View tab of the Oracle PCA Dashboard are no longer correctly aligned with the background image.

There is no workaround available.

Bug 23099040

Oracle PCA Release 2.2.1 uses the Oracle Application Development Framework (ADF) version 11.1.1.2.0 for both the Dashboard and the Oracle VM Manager user interface. This version of ADF does not support Microsoft Internet Explorer 10 or 11.

Workaround: Use Internet Explorer 9 or a different web browser; for example Mozilla Firefox.

Bug 18791952

Both the Oracle PCA Dashboard and the Oracle VM Manager user interface run on an architecture based on Oracle WebLogic Server, Oracle Application Development Framework (ADF) and Oracle JDK 6. The cryptographic protocols supported on this architecture are SSLv3 and TLSv1.0. Mozilla Firefox version 38.2.0 or later no longer supports SSLv3 connections with a self-signed certificate. As a result, an error message might appear when you try to open the user interface login page.

In Oracle PCA Release 2.1.1 – with Oracle VM Release 3.2.10 – a server-side fix eliminates these secure connection failures. If secure connection failures occur with future versions of Mozilla Firefox, the workaround below might resolve them.

Workaround: Override the default Mozilla Firefox security protocol as follows:

Bug 21622475 and 21803485

In environments with a large number of virtual machines and frequent connections through the VM console of Oracle VM Manager, the browser UI login to Oracle VM Manager may fail with an "unexpected error during login". A restart of the ovmm service is required.

Workaround: From the Oracle Linux shell of the master management node, restart the ovmm service by entering the command service ovmm restart. You should now be able to log into Oracle VM Manager again.

Bug 19562053

During the upgrade to Oracle PCA Software Release 2.0.4 a new version of the Xen hypervisor is installed on the compute nodes. While the upgrade is in progress, entries may appear in the ovs-agent.log files on the compute nodes indicating that xen commands are not executed properly ("Error getting VM stats"). This is a benign and temporary condition resolved by the compute node reboot at the end of the upgrade process. No workaround is required.

Bug 20901778

The compute nodes in an Oracle PCA are all placed in a single clustered server pool during provisioning. A clustered server pool is created as part of the provisioning process. One of the configuration parameters is the cluster time-out: the time a server is allowed to be unavailable before failover events are triggered. To avoid false positives, and thus unwanted failovers, the Oracle PCA server pool time-out is set to 300 seconds. As a consequence, a virtual machine configured with high availability (HA VM) can be unavailable for 5 minutes when its host fails. After the cluster time-out has passed, the HA VM is automatically restarted on another compute node in the server pool.

This behavior is as designed; it is not a bug. The server pool cluster configuration causes the delay in restarting VMs after a failover has occurred.

The Hardware Management daemon, which runs as the process named hwmgmtd, can sometimes consume a large amount of CPU capacity. This tends to become worse over time and eventually reach 100 percent. As a direct result, the system becomes less responsive over time.

Workaround: If you find that CPU load on a compute node is high, log in to its Oracle Linux shell and use the top command to check if hwmgmtd is consuming a lot of CPU capacity. If so, restart the daemon by entering the command /sbin/service hwmgmtd restart.

Bug 23174421

To simplify task management in the CLI the task identifiers (UUIDs) have been shortened. After an upgrade from a Release 2.0.x the task list may still contain entries from before the upgrade, resulting in misaligned entries due to the longer UUID. The command output then looks similar to this example:

PCA> list task
Task_ID         Status  Progress Start_Time           Task_Name
-------         ------  -------- ----------           ---------
3327cc9b1414e2  RUNNING None     08-18-2015 11:45:54  update_download_image
9df321d37eed4bfea74221d22c26bfce SUCCESS      100 08-18-2015 09:59:08
update_run_ovmm_upgrader
8bcdcdf785ac4dfe96406284f1689802 SUCCESS      100 08-18-2015 08:46:11
update_download_image
f1e6e60351174870b853b24f8eb7b429 SUCCESS      100 08-18-2015 04:00:01  backup
e2e00c638b9e43808623c25ffd4dd42b SUCCESS      100 08-17-2015 16:00:01  backup
d34325e2ff544598bd6dcf786af8bf30 SUCCESS      100 08-17-2015 10:47:20
update_download_image
dd9d1f3b5c6f4bf187298ed9dcffe8f6 SUCCESS      100 08-17-2015 04:00:01  backup
a48f438fe02d4b9baa91912b34532601 SUCCESS      100 08-16-2015 16:00:01  backup
e03c442d27bb47d896ab6a8545482bdc SUCCESS      100 08-16-2015 04:00:01  backup
f1d2f637ad514dce9a3c389e5e7bbed5 SUCCESS      100 08-15-2015 16:00:02  backup
c4bf0d86c7a24a4fb656926954ee6cf2 SUCCESS      100 08-15-2015 04:00:01  backup
016acaf01d154095af4faa259297d942 SUCCESS      100 08-14-2015 16:00:01  backup
-----------------
12 rows displayed

Workaround: It is generally good practice to purge old jobs from time to time. You can remove the old tasks with the command delete task uuid. When all old tasks have been removed the task list is output with correct column alignment.

Bug 21650772

The CLI command list opus-ports lists ports for additional switches that are not present within your environment. These switches are labelled OPUS-3, OPUS-4, OPUS-5 and OPUS-6 and are listed as belonging to rack numbers that are not available in a current deployment. This is due to the design, which caters to the future expansion of an environment. These entries are currently displayed in the listing of port relationships between compute nodes and each Oracle Switch ES1-24, and can be safely ignored.

Bug 18904287

Oracle PCA uses a hosts file to resolve addresses on the internal management network. The host names of the lower InfiniBand switch in rack 2 and rack 3 have been misconfigured as ovcasw19r2 and ovcasw19r3 while they should be named ovcasw16r2 and ovcasw16r3 respectively.

Workaround: If you need to set up a direct SSH connection to the InfiniBand switches in an expansion rack, use the IP address instead of the host name.

For fresh installs the issue has been resolved in Release 2.2.2. These are the correct static entries:

ovcasw19r1 : 192.168.4.202
ovcasw20r1 : 192.168.4.203
ovcasw16r2 : 192.168.4.206
ovcasw20r2 : 192.168.4.207
ovcasw16r3 : 192.168.4.208
ovcasw20r3 : 192.168.4.209

Bug 24764346