Digital Clusters for Windows NT Version 1.0 With Service Pack 1 Release Information Date: October 1996 Product Version: 1.0 with Service Pack 1 How to Use These Release Notes This document provides general information and workarounds for known issues related to the Digital Clusters for Windows NT with Service Pack 1 release. It is intended as additional information over and above that presented in the Digital Clusters for Windows NT Version 1.0 documentation. Clusters Service Pack 1 is an upgrade for Digital Clusters for Windows NT Version 1.0. It supports Windows NT Version 3.51 with Service Pack 4 or 5. General information about Digital Clusters for Windows NT is located at the website http://www.windows.digital.com. These release notes are intended to be read, and the steps performed, in sequential order. Before performing the procedures contained in this document, be sure to read each group of steps thoroughly to be sure they apply to you. Please retain these instructions for future use. Table of Contents Restriction for Service Pack 1 Upgrade Installation...............2 Language Requirement for Clustered Servers........................2 List of Service Pack 1 Updated Files..............................2 Location of HSZ Drivers and Release Notes on the Service Pack CD-ROM............................................................3 Microsoft SQL Administrator Passwords Must Match..................3 Bug Fixes and Enhancements to Version 1.0 of Digital Clusters for Windows NT....................................................4 Updated Cluster Name Client Driver.............................4 New Documentation for Configuring Database Software for Failover.......................................................4 Manual Failover of SQL Databases with Active Connections.......4 SQL Database Errors Fixed......................................4 Oracle Failover Object Bug Fixes...............................5 Mylex Controllers Bug Fix......................................5 Cluster Administrator Bug Fixes................................5 Cluster Administrator Functions Correctly After Adapter Changes.....................................................5 Cluster Administrator Correctly Marks Groups as Off Line ...5 Cluster Administrator Displays Disk Signature...............6 Cluster Administrator Prevents Selection of Local and Cluster Buses............................................6 Cluster Administrator Functionality Enhancements...............6 Cluster Drive Letters Are Now Seen in Microsoft Office 95......6 Large Numbers of Shares Supported..............................6 Clusters Service Pack 1 Supports Hidden Cluster Shares.........7 Cluster Administrator User Interface Enhancement..................7 Dual HSZ40 RAID Controllers Supported.............................7 Restriction on Using AlphaServer 1000A as a Cluster Server........8 Configuring the Cluster Hardware..................................8 ECU Configuration Files........................................8 Configuring the Adaptec AHA-2944W..............................8 Diskperf Driver Incompatible with Clusters Software...............9 No Support Available for IDE Disks................................9 Changing the Adapter Configuration................................9 Assigning Fixed Drive Letters....................................10 Documentation Errata.............................................11 Configuration and Installation Guide..........................11 New Clusters Utilities Available.................................11 Utility for Manual Failover from Batch File...................11 Status Display Utility (FMSTAT) Is Available..................12 Alpha Version of SCSI Select Utility Available................13 Using an Extended LAN............................................13 Restrictions on LANs Using Bridges............................13 Restrictions on LANs Using Routers............................13 Installing the Cluster Software..................................14 No Support for Checked Versions of Windows NT.................15 Upgrading to Windows NT Version 3.51 Service Pack 5 and Digital Clusters for Windows NT Version 1.0 Service Pack 1..15 Cluster Names Must Be Unique..................................15 Logon Rights Occasionally Granted Improperly..................16 Effects of Canceling a Clusters Installation..................16 Using Cluster Administrator......................................18 Defining Share Names..........................................18 Display Unsynchronized with Cluster Failover Manager Database.18 Lost Delayed Write Pop-Up Boxes..................................19 Disk Administrator Shows Orphaned Disks..........................20 Potential Problem When Using Microsoft Word Version 6.0a.........20 Cluster Client Software..........................................21 Windows NT Clients............................................21 Windows for Workgroups Clients................................21 Windows 95 Clients............................................22 All Supported Clients.........................................23 Windows NT Macintosh Clients..................................23 Copyright and Trademark Information..............................24 Restriction for Service Pack 1 Upgrade Installation When upgrading to Service Pack 1, do not upgrade while running applications such as the Cluster Administrator or SQL Server. If you do try to install the Clusters Service Pack 1 while running an application, the upgrade will not complete. Language Requirement for Clustered Servers Within a cluster, the two cluster servers as well as the domain controller for the cluster must be running English versions of Windows NT. List of Service Pack 1 Updated Files The files that are updated when you install Service Pack 1 include the following: 1) In the user's target directory, which is typically: C:\Program Files\Digital\Clusters: cfmdsrv.exe cluivp.exe clustat.exe cluxfer.exe cns.exe fmcore.exe fmdisk.dll fmoracle.dll fmscript.dll fmstat.exe fmsql.dll logwatch.exe ntcluster.exe sp_fallback_DEC_perm_srv_db.sql 2) In the user's Windows NT 3.51 System32 directory: cfmd.dll cfms.dll clucis.dll clumgmt.dll clunsapi.dll fmlib.dll 3) In the user's Windows NT 3.51 System32\drivers directory: cfs.sys cluport.sys aic78xx.sys Location of HSZ Drivers and Release Notes on the Service Pack CD-ROM The HSZ drivers are in the following locations on the Service Pack CD-ROM: o The registry programs called by installation: \HSZDrvrs\HSZ_V2.5\Intel\reggie.exe or \HSZDrvrs\HSZ_V2.5\Alpha\reggie.exe o The installation program: \HSZINSTL.BAT o The HSZ driver: \HSZDisk.sys The Release Notes for the drivers are in the following location: \HSZDrvrs\HSZ_V2.5\RENOTSP1.DOC Microsoft SQL Administrator Passwords Must Match Ensure that SQL administrator passwords match between the SQL account and the SQL failover object. Mismatched passwords can cause the Failover Manager to freeze. Bug Fixes and Enhancements to Version 1.0 of Digital Clusters for Windows NT The Digital Clusters for Windows NT with Service Pack 1 release provides fixes to a number of functionality or user interface errors found in Digital Clusters for Windows NT, Version 1.0. The fixes are described in this section. Updated Cluster Name Client Driver The Windows NT Cluster Name client driver (cfs.sys) has been fixed so that if you update your system network bindings, the driver uses the updated values the next time the system is rebooted. In Version 1.0, the driver ignored the new binding values. New Documentation for Configuring Database Software for Failover A new chapter is available of the Digital Clusters for Windows NT Administrator's Guide. This chapter includes revised instructions for configuring Microsoft SQL Server Version 6.5 and Oracle7 Workgroup Server Versions 7.1 and 7.2 with Digital Clusters for Windows NT Version 1.0 with Service Pack 1. The new chapter is shipping as a printed addendum with the Service Pack 1 upgrade kit. Manual Failover of SQL Databases with Active Connections When running Version 1.0, a manual failover of a group would fail if it contained an SQL database object with active connections, and the SQL Server process would be unreliable. This problem is fixed in Service Pack 1. New code in this release first stops all active SQL client connections to the databases being failed over before executing the failover. SQL Database Errors Fixed The following SQL database errors were discovered in Version 1.0 and are fixed in Service Pack 1: o Error when database spans multiple disk fixed o Error when Cluster Failover Manager Hangs during System Shutdown When Version 1.0 was running and SQL databases were active on shared cluster disks, the Cluster Failover Manager process could hang when the system was shutting down. This problem is fixed in Service Pack 1. Oracle Failover Object Bug Fixes Service Pack 1 includes two Oracle Failover Object code fixes: o In Version 1.0, if you changed the sid attribute of an Oracle failover object, the Cluster Administrator appeared to complete the operation but did not propagate the change to the Cluster Failover Manager Database (CFMD). This required you to work around the bug by deleting and recreating the Oracle Failover object. This bug has been fixed in Service Pack 1. o In Version 1.0, when creating or modifying an Oracle Failover Object, the Cluster Administrator issued an error message if you did not provide an Oracle Username, Oracle Password, Oracle sid parameter file, or an Oracle Pipe name. In Service Pack 1, the Cluster Administrator interface requires that you enter values in these fields. Mylex Controllers Bug Fix In Version 1.0, the Cluster Administrator would sometimes display an incorrect error message on systems with Mylex controllers. The following message was displayed: "Attempting to insert duplicate SCSI disk". This bug is fixed in Service Pack 1. Cluster Administrator Bug Fixes Cluster Administrator Functions Correctly After Adapter Changes In Version 1.0, a bug existed that could cause the Cluster Administrator to shutdown on startup instead of displaying the Reconfigure Cluster Adapter dialog box. This bug is fixed in the Service Pack 1. Note, however, that after you change an adapter or modify an IRQ or I/O address, the Cluster Administratorstill issues several low-level error messages, indicating that the log disk is unavailable. You can ignore these error messages; after the messages are displayed, the dialog box displays correctly. At that point, verify or correct the information displayed in the dialog box, and select OK. Cluster Administrator Correctly Marks Groups as Off Line In Version 1.0, the Cluster Administrator sometimes marked failover groups as off line although they were actually on line on the remote server. This bug is fixed in Service Pack 1. Cluster Administrator Displays Disk Signature The Manage Adapter Configuration dialog box now displays the disk signature. Cluster Administrator Prevents Selection of Local and Cluster Buses In Version 1.0, the Manage Adapter Configuration or Reconfigure Adapter dialog box permitted you to select a local bus and a cluster bus simultaneously, which made it unclear which bus was associated with the specific adapter information being displayed by the dialog box. This bug has been fixed in Service Pack 1. Cluster Administrator Functionality Enhancements The following functionality enhancements to the Cluster Administrator are included in Service Pack 1: o In this release, only one instance of the Cluster Administrator is allowed to run. If the Cluster Administrator is already running and you try to start Clusters again, the Cluster Administrator main window simply pops to the foreground. Note, however, that it is possible to start another instance of the Cluster Administrator when it is first initializing. o In Version 1.0, if you modified a failover group and removed all of its failover objects, the Cluster Administrator failed when you selected OK. If you modify a failover group and remove all of its failover objects, the Cluster Administrator now issues an error message that a group must contain at least one object. o The Manage Manual Failover dialog box displays on line/off line status information. This dialog box lists each failover group as either off line or on line. For groups that are on line, the nodename for that group is displayed, also. Cluster Drive Letters Are Now Seen in Microsoft Office 95 When running Version 1.0, occasionally an application (particularly Microsoft Office 95) was unable to display drive letters for cluster disks in its Open... dialog box. This problem is fixed in Service Pack 1. Large Numbers of Shares Supported When running Version 1.0, an infinite loop could occur in the process running CNS.EXE if a large number of cluster shares were defined. This problem is fixed in Service Pack 1. Clusters Service Pack 1 Supports Hidden Cluster Shares When Version 1.0 was running and you created a share on a cluster disk with a share name ending with the dollar sign ($) character, the share was not marked as a cluster share. Therefore, it could not participate in any cluster failover and was inaccessible using the cluster alias. This problem is fixed in Service Pack 1. In addition, when the cluster software places a disk on line and assigns it a drive letter, the software now creates a hidden share based on the drive letter. For example, if disk X: comes on line, the cluster software creates a hidden share called X$. When a disk is placed off line, the hidden share name is removed. This behavior is consistent with Microsoft LAN Manager. Cluster Administrator User Interface Enhancements The following Cluster Administrator User Interface fixes are included in Service Pack 1: o The Cluster Administrator Tools pull-down menu includes a new option, Cluster Monitor, that you can click to access the FMSTAT utility. See the section in these Release Notes titled "Status Display Utility (FMSTAT) Is Available" for information about this utility. o In the Manage Failover dialog box, the Cancel button has been renamed to Closed. o When you choose Manage Log Disk, the Cluster Administrator now lists only those shared disks to which you may successfully assign ownership of the log disk. This differs from Version 1.0, in which all shared disks were listed, even those that were not on line to the same server as the current log disk, and therefore were unavailable to own the log disk. Dual HSZ40 RAID Controllers Supported Digital Clusters for Windows NT Service Pack 1 includes support for dual-redundant HSZ40 RAID controllers in the StorageWorks RAID 410 subsystem, or in any HSZ40-based RAID subsystem. NOTE For use with Service Pack 1, the correct version of the HSZDISK.SYS class driver is Version 2.5. The correct firmware revision for the HSZ40 is V3.0Z-1. RAID 410 controllers must be configured to work with Clusters. To configure a RAID 410 controller: 1. Install the RAID Manager with the serial cable. For information about installing the serial cable, refer to the installation instructions supplied in the application notes of the RAID CD-ROM. 2. Set the Host Functionality to D. 3. Set the Cache flush time to 10. 4. Create your storage sets. 5. On both systems, install the class driver, HSZDISK.SYS (Version 2.5). This class driver enables and supports fault tolerant capabilities. 6. Use Disk Administrator to partition and format the storage sets (NTFS). 7. Use Disk Administrator to assign drive letters. Note that HSZDISK reorders the existing drive letters. 8. Upgrade the HSZ40 controller to Version 3.0 and then install the Version 3.01 patch. Restriction on Using AlphaServer 1000A as a Cluster Server When using an AlphaServer 1000A as a cluster server with KZPSA SCSI adapter cards, the KZPSA cards may be installed only in PCI slots 0, 1, or 2. A software interaction between the KZPSA card and the AlphaServer prevents the KZPSA controller from functioning correctly in a bridged PCI slot (slot 3, 4, or 5). Configuring the Cluster Hardware ECU Configuration Files An ECU configuration file that includes a call to an overlay file will not execute on an AlphaServer system. Configuring the Adaptec AHA-2944W Adaptec AHA-2944W SCSI adapters can be manufactured using either passive or active SCSI termination. (Note, however: to date, no AHA-2944W adapters have shipped with active termination.) Use the following guidelines to configure the two types of adapters: o Adapters with Passive Termination To disable termination, remove resistor SIP packs RN1, RN2, RN3, RN4, RN5, RN6, RN7, RN8, RN9, RN10, RN11, and RN12. Jumper locations J4 and J5 have no effect and should be left open. Note that the SCSI Select function for enabling or disabling termination is inoperative. If on-card termination is to be used on a shared SCSI bus instead of external termination (via Y cables), the terminator SIPs will be powered from TERMPWR on the bus when the power to the board is turned off. o Adapters with Active Termination Use the SCSI Select enable/disable termination function. If on-card termination is to be used on a shared SCSI bus instead of external termination (via Y cables), jumper J4 must be installed so the active termination ICs will be powered from TERMPWR on the bus when the power to the board is turned off. Diskperf Driver Incompatible with Clusters Software The diskperf driver is used to collect performance data. The driver gets installed into the SCSI driver chain. However, the disk counters are suppressed when Digital Clusters for Windows NT is installed. No Support Available for IDE Disks Digital Clusters for Windows NT, Version 1.0, does not support systems that include IDE disks. The ATDISK.SYS driver used for IDE disks in incompatible with the cluster drivers. Using ATDISK.SYS in a cluster causes the system to fail when booted. Changing the Adapter Configuration If you have moved your SCSI bus adapter to another I/O slot, added or removed bus adapters, or installed a new version of the bus adapter driver, the cluster software will not be able to access your shared disks. The cluster software flags this situation by doing the following: o The cluster driver displays the message "Adapter configuration has changed" on the blue screen while the system is rebooting. o A message is written to the event log indicating the cluster software is unable to control the disks because there is no shared bus selected. o The Cluster Administrator displays a warning that it cannot obtain information on the shared disks. Note that under this condition, the shared disks are accessible simultaneously from both servers, and you should be careful not to access the file system through both servers. To correct the situation, use the Manage Adapter Configuration function of the Cluster Administrator to respecify the cluster adapter. You should then immediately reboot your server. Assigning Fixed Drive Letters Once the Windows NT operating system has been installed, it is imperative that the drive letter assigned to the system partition (typically C: or D:) remain constant, because hard-coded references to this drive letter are recorded in the Registry. The following events can cause the assignment of floating drive letters to change: o When additional disks are added to the system after Windows NT has been installed, such as during the cluster server software installation. o When the system has more than one SCSI bus adapter and the adapters are rearranged in the I/O slots or when adapters are added or removed. o When a new revision of an adapter driver is installed, changing the starting order of the drivers. o If you have installed the HSZDISK.SYS class driver with the RAID-410 storage device. This driver starts before the default SCSI class driver, causing the operating system to discover disks in the storage array before discovering the rest of the devices on the system, including the system disk. Under Windows NT, partition drive letters are either fixed or floating. Fixed drive letter assignments remain constant each time the system is rebooted. Floating assignments, on the other hand, are reassigned each time the system is rebooted and are assigned according to the order in which the disks are discovered and the type of partition. All primary partitions are assigned first, followed by all secondary partitions. If the drive letter of your system partition changes under any of the circumstances listed above, you might be able to remedy the situation by assigning it a fixed drive letter using the Tools menu, Drive Letter function in Disk Administrator. Note, however, you cannot assign a fixed drive letter to a secondary partition. (Although Disk Administrator appears to accept such an assignment, the drive letter does not, in fact, remain fixed.) If your system partition is a secondary partition (for example, D:), you should assign fixed drive letters to all your primary partitions, beginning with a letter high enough (perhaps F:) to leave "space" at the beginning of the alphabet so that your secondary partitions are always assigned to the same drive letter. NOTE Disk Administrator might need to reboot your system for the fixed drive letter to take effect. Documentation Errata Configuration and Installation Guide The following error was discovered in the Digital Clusters for Windows NT Version 1.0 Configuration and Installation Guide after the manual was printed: In Appendix A, Uninstalling the Software, Saving the Cluster Configuration: The procedure described in this section will not work except under very stringent conditions. Because the configuration restoration, as described in the Guide, is performed by the Registry editor instead of the Cluster Administrator, no way exists for restoration on one server to be replicated automatically on the second server. Furthermore, an attempt to perform this process manually can lead to inconsistencies in the dynamic state information that describes the cluster. Digital recommends that you do not attempt to save and restore your cluster configuration using the method described in the Configuration and Installation Guide. Instead, after reinstalling the cluster server software, you should reconfigure your cluster using the methods described in the Administrator's Guide. New Clusters Utilities Available Utility for Manual Failover from Batch File Service Pack 1 includes a new command line utility titled "cluxfer". This utility allows you to cause a manual failover of a cluster failover group from within a batch file, or from an application. With this utility, you can transfer a cluster failover group to a specific cluster node. To invoke the utility, enter the following command at the command line: cluxfer Groupname [Nodename] where: - Groupname is the name of the group to transfer. - Nodename is an optional argument that specifies the node to which the group should be transferred. If you specify the Nodename argument, the group is placed on line on that node if possible. If you do not specify the Nodename, and the group is currently on line, the group is placed off line on one node and brought on line on the other node of the cluster. If the group is currently off line, it is brought on line on its primary node. Status Display Utility (FMSTAT) Is Available FMSTAT is a simple cluster status display program that can be run on any Windows NT system. It displays the status of all failover groups in a cluster as seen from both machines in the cluster. If you run FMSTAT on a machine that is part of a cluster, you can start it by typing fmstat on a command line (or by using the Run... menu). The account under which FMSTAT is run must be the same cluster account that was specified during installation. You can run FMSTAT from other systems than the cluster members. In addition, the Digital Clusters for Windows NT client software is not required on a system to run FMSTAT. To run the status display program from another system, just copy the fmstat.exe executable file to the destination system, and run it as previously described. The utility displays information about your system cluster. If you want to examine the status of another cluster, or you want to run FMSTAT on a noncluster machine, start it with the command: fmstat computername where computername is the name of one of the computers in the other cluster. When displaying cluster information, FMSTAT displays a row for each failover group and a column for each of the two servers in the cluster. Within this matrix, the current status for each group is shown, with FMSTAT assigning one of three states to each group. FMSTAT indicates that a group is either: o Down (shown in blue): Indicates that the server node is currently down, or is not responding to RPC requests. You may see a "Down" status for a short period of time after FMSTAT is started while it is gathering cluster information. o Off line (shown in red): Indicates that the group is off line on this server, and has never been on line on this server. o Date and time stamped (shown in one of three colors): - Red: Indicates that the group is currently off line on this server, and the date/time shows the last time it was on line. - Green: Indicates that the group is currently on line on this server, and this is the primary server for the group. The date/time is the time when the group came on line on this server. - Yellow: Indicates that the group is on line on this server, and this is the backup server for this group. The date/time is the time when the group came on line on this server. You can click the left mouse button on any of the status entries for more detailed information about the failover group as seen from the specified node. The display is normally refreshed every 5 seconds. You can change how often the program polls the servers by using the File menu's Refresh Rate dialog box. Alpha Version of SCSI Select Utility Available The SCSI Select Utility is used to configure Adaptec SCSI controllers. This utility is contained in the BIOS of most Adaptec SCSI controllers and can only be run on Intel systems. However, a version of this utility is available that can be run from the firmware menu on Alpha systems. This utility is located on the Digital Clusters for Windows NT distribution CD-ROM, in the \tools directory. To run this utility from the firmware, select Run a program... and type the following: cd:\tools\scsisel.exe The Alpha version of SCSI Select is similar to the Intel version (the one contained in the BIOS of the adapter). Note, however, that the Alpha version currently does not provide the ability to disable the SCSI bus reset feature; therefore, it cannot completely prepare the adapter for cluster use. For complete configuration, the adapter still must be placed in an Intel-based system (such as a Digital Prioris server). Using an Extended LAN The Cluster Name Server protocol is a LAN protocol. Therefore, to use clusters in an extended LAN, note the following restrictions: Restrictions on LANs Using Bridges In an extended LAN that uses bridges, the bridges must not filter NETBIOS multicast or broadcast messages. Filtering these messages will prevent bridged clients from communicating with the cluster services. Restrictions on LANs Using Routers In an extended LAN that uses routers, client access to the cluster might be restricted, depending on the type of LAN protocol being used: NetBEUI Protocol If the client is using the NetBEUI (NBF) protocol and the router is configured to bridge all NetBEUI traffic, no restrictions exist. IPX Protocol If the client is using the IPX protocol and the router is configured to bridge all IPX traffic, no restrictions exist. TCP/IP Protocol For systems using the TCP/IP protocol (NetBT), several restrictions exist: o The IP addresses of the two cluster servers must be together in the same subnet o Nodes in a cluster must be assigned static IP addresses. NOTE When configuring the TCP/IP protocol, do not use the DHCP option for assigning network addresses. DHCP does not guarantee that IP addresses are assigned in the same subnet. Also, DHCP reassigns addresses after a failover, and clients might have difficulty locating the reassigned services. o The IP addresses of the clients must be located in the same subnet as the cluster machines to allow the clients to access cluster name alias and browsing features. Regardless of the subnet, however, clients can access the cluster machines by identifying their host names. To enable a cluster naming operation in a routed environment, broadcasts travel in both directions. The client broadcasts to the cluster to resolve a cluster alias name, and the cluster broadcasts to clients its availability, which clients then use when browsing. Instead of enabling the passage of all broadcast traffic between routers, on some routers you can direct broadcasts that occur on one subnet to specific systems on another subnet. For example, a client can resolve a cluster name alias if the router between the client and cluster is configured to pass broadcasts occurring on the client subnet directly to the two cluster members, only. Installing the Cluster Software No Support for Checked Versions of Windows NT Digital Clusters for Windows NT is not supported on checked (debug) versions of Windows NT Version 3.51, such as the version available with the Windows NT Device Driver Kit (DDK). Digital Clusters for Windows NT must be installed on a release build (retail kit) of Windows NT Version 3.51 with Service Pack 4 or Service Pack 5. Service Pack 5 is provided on the Digital Clusters for Windows NT distribution CD-ROM. NOTE You must install the Windows NT Service Pack 5 software BEFORE you install the cluster server software. Upgrading to Windows NT Version 3.51 Service Pack 5 and Digital Clusters for Windows NT Version 1.0 Service Pack 1 If you are currently running Digital Clusters for Windows NT Version 1.0 with Windows NT Version 3.51 Service Pack 4, you must follow these steps to upgrade to Windows NT Version 3.51 Service Pack 5 and Digital Clusters for Windows NT Version 1.0 Service Pack 1: 1. Stop all clusters services by using the Services applet on the Windows NT Control Pannel. The clusters services that must be stopped are: Cfmd Server Cluster Failover Manager Cluster Log Service Cluster Name Service 2. Install Windows NT Version 3.51 Service Pack 5. You may notice that the Service Pack 5 installation deletes some files used by your cluster software. These files will be replaced when you install the Digital Clusters for Windows NT Version 1.0 Service Pack 1 in the next step. 3. Install Digital Clusters for Windows NT Version 1.0 Service Pack 1. NOTE You must install Windows NT Version 3.51 Service Pack 5 BEFORE Digital Clusters for Windows NT Version 1.0 Service Pack 1 because Clusters Service Pack 1 updates files changed by Windows NT Version 3.51 Service Pack 5. Cluster Names Must Be Unique The names you assign to different clusters in the same LANmust be unique within the first eight characters. Otherwise,it is possible for one cluster to see resources owned byanother. For example, the names CLUSTER1 and CLUSTER2 are acceptable; CLUSTER1A and CLUSTER1B are not. Logon Rights Occasionally Granted Improperly On occasion the Windows NT software improperly grants the "logon as a service" right to a user account, even when the account was created correctly. If this happens, the cluster services will not start. The workaround for this problem is as follows: 1. Open the Services applet in Control Panel. 2. Select the CFMD server. 3. Click on Startup... then click on OK. The software displays the following message: The user is granted log in as a service. At this point, you can start the cluster services. Effects of Canceling a Clusters Installation If you exit prematurely from the Setup program when installing either the cluster server or cluster client software, application files and Registry entries can be left on your system. You must manually delete these files and Registry entries and reboot your system before reinstalling the cluster software. This section of the Release Notes describes the files and subdirectories that should be deleted, including: o Server Files and Registry Entries o Windows NT Client Files and Registry Entries o Windows 95 Client Files and Registry Entries o Windows for Workgroups Client Files Server Files and Registry Entries Delete the following server files and registry entries: o All files and subdirectories under the target directory. The target directory is typically: C:\Program Files\Digital\Cluster o Files copied to your Windows NT System32 directories, including those files located in: - The top level of the System32 directory, including: clumgt.dll cfms.dll fmlib.dll clucis.dll cfmd.dll - The System32\drivers subdirectory, including: cludisk.sys cluport.sys o The following keys, where root =HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services. Use the regedt32 tool to delete: root\ClusterFailoverManager root\Cfmd root\CluPort root\CluDisk root\ClusterNameServer root\Logwatch root\EventLog\System\CFS root\EventLog\System\CluPort root\EventLog\System\CluDisk root\EventLog\Application\Cluster Event Log Also delete the following key: HKEY_LOCAL_MACHINE\SOFTWARE\DigitalEquipmentCorporation \DigitalClustersforWindowsNT Windows NT Client Files and Registry Entries Delete the following client files and registry entries: o All files and subdirectories under the target directory. The target directory is typically: C:\Program Files\Digital\Cluster o Files copied to your Windows NT System32 directories, including those files located in: - The top level of the System32 directory: clunsapi.dll - The System32\drivers subdirectory: cfs.sys o The following keys, using the regedt32 tool: HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\CFS HKEY_LOCAL_MACHINE\SOFTWARE\DigitalEquipmentCorporation \DigitalClustersforWindowsNT Windows 95 Client Files and Registry Entries Delete the following Windows 95 client files and registry entries: o Files copied to the top level of your Windows 95 System32 directory, including: clunsapi.dll clusprop.dll dclnamc.vxd dclnamc.sym dclnamc.map dclnamc.exp o The following keys, using the regedit tool: HKEY_CLASSES_ROOT\CLSID\ 95F0CC20-8342-11CF-8F47-08002B2A957A HKEY_CLASSES_ROOT\Digital_Cluster HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\ Services\DigitalCluster HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\ Services\VxD\DCLNAMC o The data value "DigitalCluster" from the following key: HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\ NetworkProvider\Order Windows for Workgroups Client Files Delete the following Windows for Workgroup client files: o Files copied to the top level of your Windows System directory, including: clunsapi.dll clunsapi.pbd dclnamc.386 dclnamc.sym dclnamc.map dclnamc.exp o The following entry in the 386 Enhanced section of the SYSTEM.INI file: device = dclnamc.386 Using Cluster Administrator Defining Share Names The Digital Clusters for Windows NT software does not prevent the creation of two file shares with the same name if the shares are on different disks and the disks are on line to different cluster servers. Digital recommends that you do not create two shares with the same name on two servers in the same cluster. Display Unsynchronized with Cluster Failover Manager Database It is possible for the Cluster Administrator display to become unsynchronized with the cluster configuration database in the registry. This can happen if you have the Cluster Administrator running on both servers of the cluster and are making changes from both servers. The Cluster Administrator on one server may be unaware of changes propagated to the registry from the Cluster Administrator on the other server. Two workarounds exist for this problem: o If you suspect that the Cluster Administrator is out of synch, choose Refresh from the View menu. o Run the Cluster Administrator from one server at a time, only. Lost Delayed Write Pop-Up Boxes A failover group can transition from on line to off line while the system is still running. This can occur from a voluntary failover or involuntary failover. During a voluntary failover, a system puts a group off line in a controlled manner in response to a user request. An example of this is a group with failback enabled that is on line on the failover server when the primary server comes back up. The failover server does a voluntary failover to the primary server. During a voluntary failover the following happen: - The shares, if any, are deleted. - The file system caches are flushed to disk. - The file system is dismounted if possible. - The disk itself is put off line. Every opportunity is taken in this case to assure that cached file system data is returned to disk prior to the failover. During an involuntary failover, a disk goes off line immediately without a chance for any flushes to occur. This may happen in a network partition (communication lost between the two cluster members) where the offline member resumes ownership of a disk because it thought the other was down. This can occur although efforts are made to prevent it. When ownership of a disk is taken from the on line member, the disk becomes inaccessible immediately without the normal graceful offline behavior. As a result of an involuntary failover, the operating system receives errors when it attempts to write the data in the file system cache to the disk as part of normal background processing. The errors appear as pop-up boxes, indicating that a "delayed write" was unable to be written. The user can click OK to close these boxes. These warnings indicate that the disk failed over prematurely and that the user's file system I/Os were not written to the disk if they were issued just before, during, and after the disk went off line. Users should verify the state of their files and reissue the client operation when the group finishes failing over to the other server. Although the client file operation must be reissued, file system integrity is maintained. Disk Administrator Shows Orphaned Disks The cluster software uses either of two file system dismount mechanisms when a voluntary failover occurs (as described previously). The two mechanisms are called conventional and disconnected dismount. In the case of an involuntary failover, only the disconnected dismount is available. During a conventional dismount, the file system gracefully closes all existing connections to the disk partition on which the file system is mounted. In this case, all file system state is cleaned up and the file system is ready to remount the disk partition if the disk ever fails back to this server again. This is the preferred method and is tried first. However, the conventional dismount mechanism fails if it cannot lock the volume. (Locking the volume fails if another program, such as File Manager, has locked the volume, or if files remain open on the volume.) During a disconnected dismount, the disk data structure is disconnected from the underlying I/O system but leftphysically present in the system. When the disk is failed back to the system, it is reinstantiated. This guarantees that an old file system state associated with the old disk instance does not interfere with the new file system state on the new disk instance. However, as a side effect, the old disk instance remains visible in Disk Administrator, appearing as a gray box for the remainder of the system up time. This is normal behavior for the mechanism and is harmless. The gray box disappears when the system is rebooted. Potential Problem when Using Microsoft Word Version 6.0a If a document in a cluster share is being edited with Microsoft Word and the contents of the document are saved while the share is being failed over from one server to the other, Word can hang or crash. Once the failed server is rebooted, Word puts up one or more message boxes indicating an unrecoverable or serious disk failure. Clicking OK on these boxes eventually results in a protection failure in the Word application itself. If the application hangs, it can be terminated and the system will return to normal. This problem is not caused by Digital Clusters for Windows NT and has been reproduced using Microsoft Word Version 6.0a on systems without Digital Clusters for Windows NT installed. Cluster Client Software Windows NT Clients Network Drives May Not Connect Automatically The Digital Clusters for Windows NT client software affects the startup timing of some network-related services. If you have network devices connected to cluster shares that are marked to be automatically reconnected when you log in, you may find them visible but not fully connected if you log in to the system soon after it reboots. In this case, you can disconnect and then reconnect these network drives. To avoid this situation completely, wait a short time after the system reboots before logging in. The length of this waiting period depends on the speed of your system, the amount of network traffic in your environment, and other factors, but in general 2 or 3 minutes after the login box has appeared should be sufficient. Another way to avoid this situation is to ensure that, when you next restart and log into your system, you have no network drives connected to cluster shares that are marked to be automatically reconnected. Enumerating Cluster Objects The "net view \\clustername" command does not work properly on Windows NT clients. The information usually presented by this command can be obtained by enumerating a cluster's objects using File Manager. Determining Time from the Network The "net time \\clustername" command is not supported, as this is not a failover resource supported by clusters. Issuing the "net time \\servername" command to either server in the cluster does work, however. Windows for Workgroups Clients Installation Inconsistency The final screen of the software installation procedure for Windows for Workgroups clients allows two options: Restart the PC now or Don't restart now. Because of a problem in the product that provides installation services for Digital Clusters for Windows NT, it will permit you to check both these options. If you check both, the system restarts immediately. Enumerating Cluster Objects To enumerate clusters on a Windows for Workgroups client system, use the browser provided with File Manager. The name "DigitalClusters" appears as a domain name. Double-click on this name to display the set of known clusters. To enumerate the objects on a particular cluster, double-click on the cluster's icon. Windows 95 Clients Updated dclnamc.vxd File Available to Fix a Problem with Find Files/Folder On a Windows 95 client, if a user selects Find Files/Folders from the Start menu, the window can freeze and no further network browsing can occur. To fix this problem, an updated driver is provided with the Clusters Service Pack 1 release. The driver, dclnamc.vxd, is located in the following directory on the CD-ROM: \CluSP1\Client\W95 For your convenience, a batch file, CLW95SP1.BAT, is supplied that copies the new driver to your system. Follow these steps to use the batch file: 1. Start an MS-DOS dialog box by selecting Programs/Main/MS-DOS from the Start menu. 2. Change to the CD-ROM drive. For example, if D is mapped to your CD-ROM drive: > D: 3. Change to the directory CluSP1\Client\W95: > cd CluSP1\Client\W95 4. Run the batch file, passing it the location of your Windows 95 root directory. For example: > CLW95SP1 C:\WIN95 Wildcarding a Network Drive Letter Assignment Using the net use Command Wildcarding a network drive letter assignment using the following command string does not display correctly in My Computer and Windows Explorer on Windows 95 clients: net use * \\server-name\share-name Enumerating Cluster Objects The "net view \\clustername" command does not work properly on Windows 95 clients. The information usually presented by this command can be obtained by enumerating a cluster's objects using the Network Neighborhood. Double-click on the Network Neighborhood icon, and then double-click on the "Digital Clusters" icon. Clusters are denoted by domain icons. To enumerate the objects of a cluster, double-click on its icon. Determining Time from the Network The "net time \\clustername" command is not supported because this is not a failover resource supported by clusters. Issuing the "net time \\servername" command to either server in the cluster does work, however. All Supported Clients Order of Network Bindings For the Clusters software to detect system failures quickly, the network bindings under NETBIOS interface (see Control Panel, Network Applet) must be listed in the same order on both servers. This requirement applies to bindings under NETBIOS only. The Clusters software uses the bindings for its heartbeat traffic in the order you specify. You can choose the most advantageous order for your system; for example, if you use multiple LAN adapters, youmight list your private or least loaded LAN first so that the Clusters software uses it first. LANman Connections May Take Longer With the cluster software installed, connections to LANman servers may take longer than normal. Because LANman server names and cluster server names are indistinguishable, the cluster software first tries to establish a requested connection by locating a cluster server of the specified name. It is only after the attempted cluster connection times out that the software passes the request to the LANman redirector. Windows NT Macintosh Clients If Macintosh services are installed under Windows NT, a menu is added to File Manager providing management of Macintosh volumes. Because cluster shares are available devices, File Manager will then allow the user to create a Macintosh volume on a cluster share, and this volume would then be available to the Macintosh clients through the server that owns the share. However, following a failover, the client connection to this volume will be lost and clients will not be able to reconnect to this volume on the failover server. Do not create Macintosh volumes on cluster shares. Copyright and Trademark Information Restricted Rights: Use, duplication, or disclosure by the U.S. Government is subject to restrictions as set forth in subparagraph (c) (1) (ii) of the Rights in Technical Data and Computer Software clause at DFARS 252.227-7013. Copyright Digital Equipment Corporation 1996 All rights reserved. The information in this document is subject to change without notice and should not be construed as a commitment by Digital Equipment Corporation. Digital Equipment Corporation assumes no responsibility for any errors that may appear in this document. The software described in this document is furnished under a license and may be used or copied only in accordance with the terms of such license. No responsibility is assumed for the use or reliability of software on equipment that is not supplied by Digital or its affiliated companies. The following are trademarks of Digital Equipment Corporation: Alpha AXP, AlphaGeneration, AlphaServer, AlphaStation, Digital, Prioris, ServerWORKS, StorageWorks, and the DIGITAL logo. The following are third-party trademarks: Adaptec is a trademark of Adaptec Inc. Intel is a registered trademark of Intel Corporation. Macintosh is a registered trademark of Apple Computer, Inc. Microsoft, MS-DOS, Windows, and Windows 95 are registered trademarks and Windows NT is a trademark of Microsoft Corporation. NetBIOS is a trademark of Micro Computer Systems, Inc. NT is a trademark of Northern Telecom Limited. Oracle and SQL*Net are registered trademarks and Oracle7 is a trademark of Oracle Corporation. OS/2 is a registered trademark of International Business Machines Corporation. SQL Server is a trademark of Sybase, Inc. All other trademarks and registered trademarks are the property of their respective holders.