pointrelnotes.txt

vxworks下的一个嵌入式实时数据库,可以分析源码.

TPR#: 452062
LONG DESCRIPTION: Now command 'hsb switch primary' does not report error 14537 if 
    state ends to PRIMARY ACTIVE or PRIMARY ALONE. 
    E.g. role-switch was done even if connection is broken and there is possibility 
    to have two primary databases.

SHORT DESCRIPTION: RC field (of admin command) has '0' when netcopy failed.
TPR#: 87005
LONG DESCRIPTION: Documentation fixed to reflect the product behavior

SHORT DESCRIPTION: 'HSB status copy' returns a false alarm
TPR#: 870018
LONG DESCRIPTION: Documentation fixed to reflect the product behavior

SHORT DESCRIPTION: SQL_ATTR_QUERY_TIMEOUT is fully supported.
TPR#: 870020
LONG DESCRIPTION: Programmer Guide corrected.


OTHER FIXES
-----------

1. The value of the parameter HotStandby.2SafeAckPolicy=3 is now supported 
(2-Safe Durable).

2.In Accelerator, the server may be started repeatably, without hanging.

3. The re-execution of join queries is now significantly (multiple) faster.


PRODUCT ENHANCEMENTS
--------------------

SHORT DESCRIPTION: Enhancement request to enable monitoring on a diskless server 
    (soltrace.out, solmsg.out/solerror.out to a disk for debug purposes)
TPR#: 451738
LONG DESCRIPTION: A new option to SS_DEBUG utilities is added. Soltrace.out and 
    solerror.out output can be enabled in diskless by admin command 
        'ssd /DLLOG:<pathprefix for the output files>'
    Once diskless logging has been enabled all the output files will go to the 
    specified directory (<pathprefix for the output files>). 
    E.g. admin command 'mon on' starts logging messages to soltrace.out file, which 
    is written to <pathprefix for the output files>. Also ssdebug.out output can 
    be enabled in diskless if DLLOG path has been given. Tracing/logging in diskless 
    can be disabled by admin command 'ssd /NODLLOG'.

------------------------
New  commands
------------------------
1. ADMIN COMMAND 'shutdown force';
   is equivalent to the sequence:
   ADMIN COMMAND 'close';
   ADMIN COMMAND 'throwout all';
   ADMIN COMMAND 'shutdown';

2. ADMIN COMMAND 'errorcode all';
   lists all error codes that server may generate.

3. ADMIN COMMAND 'errorexit <number>';
   causes an immediate process exit with the given process exit code.

----------------------------------------
New or modified configuration parameters
----------------------------------------
Old parameter:
HotStandby.PingTimeout = 4000

The meaning has changed so that it really means timeout on each ping sent
by the server. After the time specified (in milliseconds) has passed the
server concludes that a connection is broken and changes the state
accordingly. Default is 4 sec.

New parameter:
HotStandby.PingInterval = 1000

The value is equal to the interval (in millisecobds) between two
consecutive pings sent by a server. Default is 1 s. With the default
settings, the server reacts to a connection break in 4-5 sec.



PRODUCT.....: BoostEngine 4.1
VERSION.....: 04.10.0134
PLATFORMS...: w32, l2x
RELEASE DATE: January 28, 2004

SHORT DESCRIPTION: Wrong index selection causes error in BLOb reference 
    count crashing the server.
TPR#: None
LONG DESCRIPTION: This is a very rare case: with only a few very large 
    BLObs (1GB+) in the database, with pagesize of 32K, the server 
    crashes because of a BLOb reference count error.



PRODUCT.....: BoostEngine 4.1
VERSION.....: 04.10.0133
PLATFORMS...: l2x, h1x, s9x, w32, hia64, s8x
RELEASE DATE: January 28, 2004

SHORT DESCRIPTION:  An occasional timing error caused the server process to start 
    consuming CPU in a busy-loop. The problem does not directly cause data loss or 
    crashes.
TPR#: 11433, 451960, 451961
LONG DESCRIPTION: The code resulting the loop was fixed.

SHORT DESCRIPTION:  Updates made inside a stored procedure are rolled-back at 
    disconnect when using ODBC-connection in auto-commit mode. 
TPR#: 11428
LONG DESCRIPTION: Now cursor counter is incremented when a procedure that does
    not return a resultset is called and commit is called when the statement 
    is closed.

SHORT DESCRIPTION: Irregular format in the solmsg.out file.
TPR#: 11491
LONG DESCRIPTION:  Now this is fixed.

SHORT DESCRIPTION:  ODBC call SQLPutData may go into an endless loop if data is 
    sent in chunks.
TPR#: 20161, 50146
LONG DESCRIPTION: Now SQLPutData correctly handles the data when its size
    is set with SQL_LEN_DATA_AT_EXEC(lenght) macro.

SHORT DESCRIPTION:  ODBC call SQLGetInfoA(hdbc, i, &strVal, maxLen, &recLen) 
    does not work if maxLen is larger than 1024*8-1 = 8191.
TPR#: 20162
LONG DESCRIPTION: When converting the string arguments into Unicode characters, 
    the length of the string must be smaller than mazimum SMALLINT. Now more
    checks is done during the conversion.

SHORT DESCRIPTION:  SQLGetTypeInfo() does not report correct precision for 
    TIMESTAMP.
TPR#: 20163
LONG DESCRIPTION: Now precision is correct: 29.

SHORT DESCRIPTION: Problem with ODBC function SQLGetData (Internal error in SQLGetData).
TPR#: 20164
LONG DESCRIPTION:  When a parameter is unbinded a row descriptor is removed,
     but the descriptor is not recreated when SQLFetch is called. Now this
     is fixed. 

SHORT DESCRIPTION:  Procedure not found (23005) error in returned incorrectly.
TPR#: 451931
LONG DESCRIPTION: After transaction timeout search routines for finding procedures
    did not return the correct error.

SHORT DESCRIPTION:  Problem with fetching decimal values through ODBC.
TPR#: 451937
LONG DESCRIPTION:  A bug with decimal conversions is now fixed.

PRODUCT ENHANCEMENTS
--------------------

TPR#: 451947
DESCRIPTION: Now the -xnointegrity option is supported in Solid Speed Loader.

TPR#: 451938
DESCRIPTION: Now cleanbgjobinfo admin command can be given as a part of the timed 
    command list (in the ini file).
    An example:
        At=00:00 backup, 02:45 cleanbgjobinfo


PRODUCT.....: BoostEngine 4.1
VERSION.....: 04.10.0131
PLATFORMS...: Linux
RELEASE DATE: January 08, 2004

SHORT DESCRIPTION:  An occasional timing error caused the server process to start consuming CPU in a
    busy-loop. The problem does not directly cause data loss or crashes.
TPR#: 11433, 451960, 451961
LONG DESCRIPTION: The code resulting the loop was fixed.



PRODUCT.....: BoostEngine 4.1
VERSION.....: 04.10.0129
PLATFORMS...: Linux
RELEASE DATE: January 08, 2004


Note: These release notes do not list all the bugs that are fixed.
Some of the bugs or limitations that were listed in 
the original 4.1 GA release notes have also been fixed. 

TPR#: 451937
Short description: Problem with fetching decimal values with ODBC.
Resolution: Decimal value has '.' at the end so it will cause 
     "String data, right truncated" error. 

     Now the truncated value is assigned to the binded variable 
     despite of the truncation.



PRODUCT.....: BoostEngine 4.1
VERSION.....: 04.10.0128
PLATFORMS...: l2x
RELEASE DATE: December 31, 2003


TPR#: 451923
Short description: After netcopy from PRIMARY the SECONDARY is not 
     sending notification events.
Resolution: When secondary starts receiving netcopy it does a 
     DB-shutdown, e.g. only netcopy request is served. Because of 
     this shutdown all registered callback information was lost.
     Now this information is maintained correctly after netcopy.

TPR#: 451931
Short description: Procedure not found (23005) error in solsql.
Resolution: After a transaction timeout procedure calls fail with 
     error 23005. The reason is that low level search routines return 
     DBE_ERR_TRXTIMEOUT but that is not propagated to higher level
     and a faulty error message is returned. Now timeout error is catched
     correctly.


PRODUCT.....: BoostEngine 4.1
VERSION.....: 04.10.0127
PLATFORMS...: l2x
RELEASE DATE: December 22, 2003


TPR#: 451695
Short description: If client application retrieves on a non-existent table, 
     the cursor is not properly released afterwards.
Resolution: Now if prepare fails a pending cursor drop is done in client.
     Actual cursor drop takes place in next SQLExecute.

TPR#: 451896
Short description: Old official delete being received in the replica for 
     an insert operation being propagated from the replica.
Resolution: Now there is a new bulletin board parameter OLD_OFFICIAL_UNIQUE_DELETE 
     for the case where an old official row is deleted when executing the 
     reply message in replica. This is currently possible if two incremental 
     subscriptions are processed at the same time in master.

TPR#: 451914
Short description: Enhancement request for reincluding the noconnect option 
     in the HSB switch primary command.
Resolution: Now command 'hsb set primary alone' is enabled when setrev is in 
     SECONDARY ACTIVE state. 
     This should have same effect as 'switch primary noconnect'.



PRODUCT.....: BoostEngine 4.1
VERSION.....: 04.10.0126
PLATFORMS...: All supported platforms
RELEASE DATE: December 8, 2003


TPR#: 11440
Short description: HSB switch primary returns: 14537 HotStandby 
     connection is broken.
Resolution: The behavior is made compatible with versions 3.7 and 4.0: 
    the return code is "0". 


KNOWN PROBLEMS
--------------

When the command 'hsb switch primary' is issued at the Secondary, when the
Primary is in the process of shutting down, the Secondary hangs. If the 
intention is to shut down one of the servers, it has to be made Secondary first.
If the intention is to simulate a failover, the Primary process should be 
killed. Also, in the case of a failover, the command 'hsb set primary alone' is 
preferable, at the Secondary.

Note that not all intuitively obvious migration scenarios (from 3.7 and 4.0) 
work correctly. Refer to the hot migration documentation in the previous point 
release package and the note on cold migration below (in DOCUMENTATION ERRATA).

ENHANCEMENTS
------------

The semantics of the command 'hsb set primary alone' has been extended so that 
it can be issued in the SECONDARY ACTIVE state. This can be used to force 
the Secondary to failover quickly, as described below.


------------------------------
Changing roles in Solid HSB 4.1
-------------------------------

0. Needs
---------

A role switch (from SECONDARY to PRIMARY) is required in the following
cases:

1) Failure of a the Primary server of node (failover).
2) Switching the roles for maintenance reasons (switchover).
3) Graceful shutdown of the Primary server

Also, the following termination scenarios are important:

4) Shutdown of the Secondary server
5) Shut down of the whole database system (both servers)


1. Failure of a the Primary server or node (failover)
-----------------------------------------------------
This scenario is crucial for maintaining high-grade HA. The fastest way
to perform failover is to issue, at Secondary:

admin command 'hsb set primary alone'

The speed of this command is based on the fact that no effort is made
to communicate with the Primary. The state transition is unconditional
and lasts from tens of milliseconds to several hundreds of
milliseconds, depending on the 2-Safe Ack Policy used and the number of
pending transactions.

Because the state of the Primary is not checked, the responsibility for
the correctness of the action is with the controlling software. If, for
some reason (like network partition), the Primary is still in service,
it will automatically move to the state PRIMARY UNCERTAIN after a time
resulting from the value of the parameter PingTimeout, typically
several seconds (average 7 s, by default). Consequently, it will be possible to
change the role of the original Primary to Secondary and resynchronize
the servers with 'hsb connect'.


2. Switching the roles for maintenance reasons (switchover)
-----------------------------------------------------------
If there is a need to switch the roles without shutting any server,
this command should be used:

admin command 'hsb switch primary' (at Secondary) or
admin command 'hsb switch secondary' (at Primary)

The execution involves synchronizing the servers so that no
transactions are lost and the service is unambiguously moved from one
server to another. The transition to the new state lasts between tens
and hundreds of milliseconds

The command 'hsb switch primary' can be also issued in the case of a
failover but the execution and the transition to the state PRIMARY
ALONE may last several seconds because of the involved timeout (the
server will make an effort to communicate with the failed Primary). By
default, the average timeout is ca. 7 seconds.

The command 'hsb switch primary' should never be used in connection
with the