|Oracle8i Parallel Server Concepts and Administration
This chapter describes:
This section covers aspects of database creation specific to OPS:
Database creation tasks specific to OPS can be summarized as follows:
Certain initialization parameters critical at database creation or that affect certain database operations must have the same value for every instance in OPS. Be sure these are identical across all instances before creating a database for a multi-instance environment.
To enable the ARCH process while creating a database, set the initialization parameter LOG_ARCHIVE_START to TRUE. Then change the mode to ARCHIVELOG with the ALTER DATABASE statement before starting up the instance that creates the database.
Alternatively, you can reduce overhead by creating the database in NOARCHIVELOG mode. This is the default. Then change to ARHIVELOG mode.
You cannot use the STARTUP command to change the database archiving mode. After creating a database, use the following Server Manager commands to change to archiving mode and reopen the database with parallel server enabled:
Use the standard procedure to create a database.
This section describes CREATE DATABASE options specific to OPS.
The MAXINSTANCES option of CREATE DATABASE limits the number of instances that can access a database concurrently. MAXINSTANCES defaults to the maximum value specific to your operating system; on most systems the default is two.
For OPS, set MAXINSTANCES to a value greater than the maximum number of instances you expect to run concurrently. This way, if instance A fails and is being recovered by instance B, you will be able to start instance C before instance A is fully recovered.
The MAXLOGFILES option of CREATE DATABASE specifies the maximum number of redo log groups that can be created for the database, and the MAXLOGMEMBERS option specifies the maximum number of members or copies per group.
For OPS, set MAXLOGFILES to the maximum number of threads possible, multiplied by the maximum anticipated number of groups per thread.
The MAXLOGHISTORY option of CREATE DATABASE specifies the maximum number of redo log files that can be recorded in the log history of the control file. The log history is used for automatic media recovery of OPS.
For OPS, you should set MAXLOGHISTORY to a large value, such as 1000. The control files can then only store information about this number of redo log files. When the log history exceeds this limit, old history entries are overwritten. The default for MAXLOGHISTORY is zero, which disables the log history.
The MAXDATAFILES option is generic, but OPS tends to have more data files and log files than standard systems. On your platform the default value of this option may be too low.
Oracle8i SQL Reference for complete descriptions of the SQL statements CREATE DATABASE and ALTER DATABASE. Also see your Oracle operating system-specific documentation for information on default values of CREATE DATABASE options. For more information about redo log groups and members, see "Redo Log Files" . Also see "Redo Log History in the Control File" for more information on MAXLOGHISTORY.
To prepare a new database for OPS, create and configure the following additional database objects.
You must create at least one rollback segment for each instance of a parallel server. To avoid contention, create these rollback segments in a tablespace other than the SYSTEM tablespace.
You must create and bring online one additional rollback segment in the SYSTEM tablespace before creating rollback segments in other tablespaces. The instance that creates the database can create this additional rollback segment and new tablespaces, but it cannot create database objects in non-SYSTEM tablespaces until you bring the additional rollback segment online.
To allocate a private rollback segment to one instance, follow these steps:
A private rollback segment should be specified in only one parameter file so that it is associated with only one instance. If an instance attempts to acquire a private rollback segment that another instance has already acquired, Oracle generates a message and prevents the instance from starting up.
Any instance can create a public rollback segment that can then be claimed by any instance when it starts up. Once a rollback segment has been claimed, it is only used by the instance that claimed it until the instance shuts down, releasing the rollback segment for use by another instance.
To create a public rollback segment, use the SQL statement CREATE PUBLIC ROLLBACK SEGMENT.
Typically, the parameter file for any particular instance does not specify public rollback segments because they are assumed to be available to any instance needing them. However, if another instance is not already using it, you can name a public rollback segment as a value of the ROLLBACK_SEGMENTS parameter.
Public rollback segments are identified as having the owner PUBLIC in the data dictionary view DBA_ROLLBACK_SEGS.
If the parameter file omits the ROLLBACK_SEGMENTS initialization parameter, the instance uses public rollback segments by default.
A public rollback segment is brought online when an instance requiring public rollback segments starts up and acquires it. However, starting an instance using public rollback segments does not ensure that any particular public rollback segment comes online unless the instance acquires all available public rollback segments. Once acquired, a public rollback segment is used exclusively by the acquiring instance.
Bringing online, taking offline, creating, and dropping rollback segments, whether private or public, is the same whether OPS is enabled or disabled.
Private rollback segments stay offline until brought online or until the owning instance restarts. A public rollback segment stays offline until brought online for a specific instance or until an instance requiring a public rollback segment starts up and acquires it.
If you need to keep a public rollback segment offline and do not want to drop it and re-create it, you must prevent other instances that require public rollback segments from starting up.
You can use the Server Manager command MONITOR ROLLBACK to display information about the status of the rollback segments that the current instance uses.
Alternatively, you can query the dynamic performance views V$ROLLNAME and V$ROLLSTAT for information about the current instance's rollback segments.
Use the Server Manager command CONNECT @instance-path to change the current instance before using the MONITOR command or querying the V$ views. You must have Net8 installed to use the CONNECT command for an instance on a remote node.
You can also query the data dictionary views DBA_ROLLBACK_SEGS and DBA_SEGMENTS for information about the current status of all rollback segments in your database.
For example, to list the current rollback segments, you can query DBA_ROLLBACK_SEGS with the following statement:
This query displays the rollback segment's name, ID number, owner, and whether it is in use, as shown in the following example:
SEGMENT_NAME SEGMENT_ID OWNER STATUS ------------------------ ---------- ------ ------------ SYSTEM 0 SYS ONLINE PUBLIC_RS 1 PUBLIC ONLINE USERS1_RS 2 SYS ONLINE USERS2_RS 3 SYS OFFLINE USERS3_RS 4 SYS ONLINE USERS4_RS 5 SYS ONLINE PUBLIC2_RS 6 PUBLIC OFFLINE
In this example, rollback segments identified as owned by user SYS are private rollback segments; the rollback segments identified as owned by user PUBLIC are public rollback segments.
The view DBA_ROLLBACK_SEGS also includes information (not shown) about the tablespace containing the rollback segment, the datafile containing the segment header, and the extent sizes. The view DBA_SEGMENTS includes additional information about the number of extents in each rollback segment and the segment size.
The Oracle8i Administrator's Guide for more information about rollback segments, and about connecting to a database. For the format of the connect string in instance-path see Net8 Administrator's Guide and your Oracle system-specific documentation. For a description of DBA_ROLLBACK_SEGS and DBA_SEGMENTS, and other dynamic performance views please see the Oracle8i Reference.
Each database instance has its own "thread" of online redo, consisting of its own online redo log groups. When running OPS, two or more instances concurrently access a single database and each instance must have its own thread. This section explains how to configure these online redo threads for multiple instances with OPS.
You must create each thread with at least two redo log files (or multiplexed groups), and you must enable the thread before an instance can use it.
The CREATE DATABASE statement creates thread number 1 as a public thread and enables it automatically. You must use the ALTER DATABASE statement to create and enable subsequent threads.
Threads can be either public or private. The initialization parameter THREAD assigns a unique thread number to the instance. If THREAD is zero, which is the default, the instance acquires an available public thread.
Each thread must be created with at least two redo log files or multiplexed groups. You must also enable each thread before an instance can use it.
The CREATE DATABASE statement creates thread number 1 as a public thread and enables it automatically. Subsequent threads must be created and enabled with the ALTER DATABASE statement. For example, the following statements create thread 2 with two groups of three members each, as shown in Figure 6-1 :
ALTER DATABASE ADD LOGFILE THREAD 2 GROUP 4 (disk1_file4, disk2_file4, disk3_file4) SIZE 1M REUSE GROUP 5 (disk1_file5, disk2_file5, disk3_file5) SIZE 1M REUSE; ALTER DATABASE ENABLE PUBLIC THREAD 2;
If you omit the keyword PUBLIC when you enable the thread, it will be a private thread that cannot be acquired by default. Only one thread number may be specified in the ALTER DATABASE ADD LOGFILE statement, and the THREAD clause must be specified if the thread number of the current instance was chosen by default.
You can disable a public or private thread with the statement ALTER DATABASE DISABLE THREAD. You cannot disable a thread if an instance using the thread has the database mounted. To change a thread from public to private, or vice versa, you must disable the thread and then enable it again. An instance cannot disable its own thread. The database must be open when you disable or enable a thread.
When you disable a thread, Oracle marks its current redo log file as needing to be archived. If you want to drop that file, you might need to first archive it manually.
An error or failure while a thread is being enabled can result in a thread that has a current set of log files but is not enabled. These log files cannot be dropped or archived. In this case, you should disable the thread, even though it is already disabled, then enable it.
The mode of using the redo log, ARCHIVELOG or NOARCHIVELOG, is set at database creation. Although rarely necessary, the archive mode can be changed by the SQL statement ALTER DATABASE. When archiving is enabled, online redo log files cannot be reused until they are archived. To switch archiving modes, the database must be mounted with OPS disabled, but the database cannot be open.
The redo log mode is associated with the database rather than with individual instances. For most purposes, all instances should use the same archiving method, either automatic or manual, if the redo log is being used in ARCHIVELOG mode.
You can change the configuration of the redo log, such as adding, dropping, or renaming a log file or log file member, while the database is mounted with OPS either enabled or disabled. The only restrictions are that you cannot drop or rename a log file or log file member currently in use by any thread. Moreover, you cannot drop a log file if that would reduce the number of log groups to less than two for the thread it is in.
Any instance can add or rename redo log files, or members, of any group for any other instance. As long as there are more than two groups for an instance, a redo log group can be dropped from that instance by any other instance. Changes to redo log files and log members take effect on the next log switch.
If datafiles are added while OPS is running, evaluate whether enough locks are available to cover the new files.
Added datafiles use the unassigned locks created when the value for GC_FILES_TO_LOCKS was set. If the remaining locks are not adequate to protect the new files and avoid contention, provide more locks by adjusting the GC* parameters. Performance problems are likely if you neglect to make these adjustments.
In a read-only database, extra locks are not necessary even if you added many new datafiles. In a database heavily used for inserts, however, you might need to provide more locks.
If you determine you need more locks, do the following:
You can use the CREATE CONTROLFILE statement to change the value of the following database parameters for an existing database: