Welcome to Patch 9708 for Advantage Ingres 2.6 on Windows. This patch may be installed over any Advantage Ingres 2.6 Windows releases 2.6/0201, 2.6/0207 and 2.6/0305. This Readme contains important information about patch 9708, including how to install and uninstall it on UNIX platforms, a list of all the problems that the patch addresses, and technical support contact information. You can connect to the Advantage Ingres Relational Database product page at http://ca.com/.
Please be sure to read the problem notes for INGAPI37, INGSRV1846, INGSRV2105, INGSRV2168, INGSRV2302, INGSRV2320 and INGSRV2461 as they require additional steps for remediation over and above applying the patch.
Please also read the problem note for INGSRV2279, as there is a change in the default optimizer configuration which may have a impact on performance.
Finally, please also note the minor behavioral changes introduced by the fixes for INGSRV1981 and INGSRV1997.
To install this patch on Windows platforms:
During installation the following options are available:
Providing the patch was installed with the 'Backup.....' option checked the patch may be uninstalled using the following procedure:
Answer 'Yes to all' when prompted.
This sequence of steps moves the backup files back into the %II_SYSTEM%
directory structure and overwrites those files that are already present.
If you have multiple backup directories from multiple patch installations care
should be taken when deciding which directory to use. For example, if patch 1
was installed, and then later patch 2 was installed, uninstalling using the
patch 1 backup will cause both patch 1 and 2 to be uninstalled.
The following reported problems have been resolved with this patch. For reference, each problem is listed with its published STARTRAK problem identifier, and the operating systems to which it applies. Please note, that in contrast to patches supplied for SP1 and before, problems are labelled with the STARTRAK problem identifier, rather than the internal bug number, in order to facilitate use of Computer Associates customer support systems.
Ensure Database Object Manager is in grayed out on the screen that results from right-clicking on a valid server in edbcnet.
Application in Web Logic hangs when selecting a BLOB over 32K through JDBC driver with Web Logic's connection pooling turned on.
When application updates date column through JDBC using escape sequence, JDBC Driver may send date one day less to the server. (EDJDBC73)
JDBC driver returns 0 when application querying for unicode data type NVARCHAR and LONG NVARCHAR.
IDMS restricts LIKE clause usage in queries against iicolumns catalog. Accessing schema information of IDMS database fails with IDMS ERROR -04: Table Procedure IDMSQNCO exception 38003 Column SCHEMA cannot be null.
getSchemas() returns empty ResultSet against IDMS database.
getcolumns() returns empty resultSet when queried against IDMS database as columnNamePattern is passed a value of % from DBvisulaizer tool.
Multithreaded OpenAPI application core dumps whilst trying to retieve an error message from the Ingres message files (.mnx). Typically this results in a SIGBUS in cer_sstr(). Statically linked applications must be rebuilt in order to include this solution.
Multithreaded OpenAPI application core dumps in GCF routines. Typically this is a SIGBUS in gcd_encode().
When used inside a block a .print command with a format specifier of (c0) will overwrite lines from a preceeding .print statement if the output was wrapped. The correct print position after a wrapped print statement which resides in a block should be the end of the 1st line of wrapped text.
When the .tmp file generated by an iiexport command which made use of the -allconstants option, is used to iimport an application into a database, subsequent executions of the application may SIGSRGV. Exiting . . . due to Segmentation Violation (SIGSEGV) @ PC ef125460 SP efffd370 PSR 1 165000 o0 0 Please contact your Technical Support representative.
Adding a tab to text when editing escape codes using the edit menuitem which in turn uses the system editor, causes all of the text entered to the right of a tab to disappear. This occurs when returning to the escape code editing screen from the system editor.
Upgradedb returns E_US18A5 CREATE RULE: You must be the owner of table ... in order to apply rules.
CBF: Scroll to 'Net Server' press Configure(1), then hit Key Pad 3. This causes a SIGSEGV.
Following an INGSTART, the II_SYSTEM disk cannot be dismounted.
INGSTART hangs after starting the RMCMD process, when executed from an account other than the Ingres System Administrator.
COPYDB/UNLOADDB does not allow the creation of a fully journalled copy.in script. This means that a checkpoint must be taken after copy.in or the copy.in has to be manually edited in order to avoid missing tables after a rollforwarddb through the copy.in transactions. A new "-journal" option has been added to allow the user to easily resolve this problem.
Issuing remote commands via Visual DBA against a 64-bit Server will appear not to work, even though in the background the commands execute successfully.
xmlimport fails with E_CO0032.
copydb/unloaddb: strings that contains "\t" (tab) are incorrectly generated in the copy.in file when the text is wrapped to the next line. Characters may be repeated at the end of one line and at the start of the next line. This leads to syntax errors during the copy.in
Attempts to start rmcmd fail with E_US0022 Either the flag format or one of the flags is incorrect, or the parameters are not in proper order.
A link time error of "unresolved external symbol _IIxaddfrm" is returned with the use of the embedded forms statement ADD FORM.
Data returned from a select statement of a long varchar column is not properly terminated.
An error of E_US1080 is returned when entering data via an embedded forms field.
Data returned from the embedded forms UNLOADTABLE statement fails when compared with data returned from an embedded forms PROMPT statement.
The value of a user declared string variable used in an embedded RETRIEVE statement is truncated.
Parameters passed in a list using the embedded forms statements APPEND and PARAM are truncated at the first parameter.
Select of nchar/nvarchar unicode data using dynamic sql does not return correct data length.
Select of nchar/nvarchar unicode data using dynamic sql does not return correct results for the output data.
In an embedded C program, insert interprets 'nvarchar' SQLLEN incorrectly. When using a program to insert "abc", the string is truncated and the "c" is lost.
Dynamic sql 'fetch' of data from nvarchar(10000) column using sqltype==IISQ_NVCHR_TYPE, returns corrupt value in sqldata. This happens regardless of how few bytes are actually in the column.
sqllen overflows when using long NVARCHARs in an ESQL program.
Comparisons used in the embedded SQL WHENEVER statement fail when pre-compiled with the -sqlcode flag.
Embeded Fortran programs would abend with a SEGV.
Embedded COBOL programs fail with SEGV abends and SIGBUS abends.
There is no support for compiling FORTRAN procedures from within ABF.
The IIdesc function, implemented to allow applications to pass an array of pointers to parameters declared in a QUEL PARAM clause, exceptions when run.
The returned string from a PROMPT embedded forms statement fails a comparison with a string constant.
A link time error of "unresolved external symbol _IIsqGd" is returned with SQLSTATE declared.
A null value returned from a BLOB data handler does not properly terminate the supplied buffer.
Unhandled exceptions within the icesvr process of Ingres/ICE are reported in the errlog.log file during running of reports. The icesvr process was unable to accept further requests until restarted.
The patch installer displays patch.html using notepad but should display it using an html browser.
During patch installation, instaux.bat reports 4 instances of the message: The name specified is not recognized as an internal or external command, operable program or batch file.
Ingres 2.6 for AMD64 Linux fails to install on RedHat AS 3.0 for AMD64.
IIGCC hangs when outbound max limit is reached.
Certain characters are not transliterated correctly between a system using the Windows 1250 code page and a system using the ISO88592 code page.
ODBC application gets syntax error on scalar function when using escape format {fn xxxx}.
When ODBC application calls SQLConnect with the invalid usr/pwd in a loop, ODBC driver returns the proper error for the first failed SQLConnect, and the following SQLConnect calls all failed with API error:"API function cannot be called in the current state".
ODBC 3.5 doesn't export the function SQLColAttribute().
When installing Ingres II using a separate account to the account designated as the Ingres System Administrator, the installation fails to initialize the RMCMD process information.
All the GBLSECTS used by Ingres are not released when Ingres is shutdown. Each time Ingres is started (INGSTART), and shutdown (INGSTOP), a few GBLSECTS are not released.
Error copying or modifying files which are >2GB. errlog contains: E_CL0608, E_DM9000, lseek() error 22, E_DM92CE, E_DM92E1, E_DM9259, E_DM9100, E_DM0091, E_QE0083. (107709)
If the dbms server parameter association_timeout is set to n seconds, the following E_GC0020 message is reported every n seconds even on an idle system: E_GC0020 Time_out expiration: service request incomplete
A syntax incorrect query of the form:
create table
Creating a database procedure containing parameters fails with error E_OP0890 Too many bases added to a QEN_ADF struct. The DBP contains a query of the form'delete...from...NOT IN (select...)'.
Compile times for large queries (involving large numbers of tables and/or potentially useful secondary indexes - typically more than 10) can be excessive, especially when one or more tables is very large. Ingres needs to compile such queries much faster.
XA global transaction may be rolled back if first updating statement of any participating transaction aborts, possibly resulting in lost updates.
DBMS server hangs after message "iiCLfdreq FATAL: fd nnnn surpasses max fd 8192" in errlog.log, where nnnn is a number greater than 8192. This can occur at sites which have configured the OS to allow more than 8192 file descriptors for a process.
E_DMA464_LGREAD_BADFORMAT or E_DM9054_BAD_LOG_RESERVE updating btree table having page size 64k.
With system_isolation being set to READ_COMMITTED, a simple select statement running against a readonly table with readlock equals to nolock fails with E_QE0082.
Incorrect results (extra/duplicate rows) from nested outer join query. Due to the use of a secondary index on the inner relation of an outer join which does not cover all the columns in the 'on' and 'where' clause.
Connecting to a specific server_class fails when the server_class in specified in lower case and appears in lower case in the config.dat. TM reports "E_GC0138 User provided a server class as part of the database name (dbname/class), but no servers of that class (for the specified database) are running in the target installation."
Query against database with custom collation sequence that treats uppercase and lowercase characters the same fails with incorrect results. The actual values from the rows are not returned, rather the collated values are displayed.
Massive 'IN' clauses can cause server to crash silently
QSF memory is gradually consumed until an "out of memory" failure occurs.
UPGRADEDB run-time is longer than previous releases due to
'chmod 600' operation on all data files.
To disable this operation set II_UPGRADEDB_CHMOD = FALSE
before running UPGRADEDB. The 'chmod 600' on all data files
in all data locations for each database must be run manually if
this is disabled at UPGRADEDB run time.
(110760)
In a VMS environment with a heavy load of IMA queries, the session(s) issuing the IMA queries may become unresponsive (while other DBMS sessions continue their work). In iimonitor, the hung session is shown in state CS_COMPUTABLE with CS_EDONE_MASK. Starting an IMA query from another session may cause the session to be in CS_CNDWAIT state.
Outer join queries with secondary indices on inner relations may (i) return incorrect results (ii) spend an excessive amount of time in the query optimizer or (iii) be executed without the use of an appropriate inner secondary index.
Upgradedb will drop select permits to public on all system catalog views.
auditdb crashes with: E_DM904A_FATAL_EXCEPTION A fatal error has occurred in the DMF Facility. When reporting on a table that has been dropped.
CTRL-C and CTRL-BREAK do not interrupt AUDITDB.
Spurious DMF errors followed by E_QE0083_ERROR_MODIFYING_TABLE when modifying a table from HIDATA compression to DATA compression.
Insert into blob table with long nvarchar columns along with other columns fails with E_AD9999_INTERNAL_ERROR.
Errlog.log reports a ULE_FORMAT error for error code 1106c. This is followed by E_DM905B_BAD_LOCK_ROW and E_DM9077_ERROR_LOCK_DOWNGRADE.
Ingres STAR query fails with: E_LQ0040 Protocol read-failure on internal data. Execution of this query is stopped. E_LC002B Read failure against pre-fetched data. The system encountered the end of the pre-fetched message stream. Expected to read 21 bytes but read only 8 bytes from a GCA_TUPLES message. Error occurs when using cursors in esqlc.
Any iimerge server (iidbms, dmfrcp, dmfacp etc.) crashes on startup if II_THREAD_TYPE is set to INTERNAL
Upgradedb may return E_US0042 errors unless you run upgradedb with the -d flag.
A join query may produce incorrect results (too many or too few rows). SET QEP will show the presence of a hash join.
ERslookup error received when Ingres attempts to display message E_DM9073_BAD_UNLOCK_ROW_BY_LKID or message E_DM9072_BAD_UNLOCK_ROW.
Subsequent select queries following a rollforwarddb or recovery REDO processing may fail with E_DM93A7, E_DM9206.
An 'is null' test on a nullable varchar variable inside a database procedure gives "E_OP0791_ADE_INSTRGEN" at procedure creation time. The errlog.log shows: E_AD2005_BAD_DTLEN, E_QS001E_ORPHANED_OBJ and E_QS0014_EXLOCK
rollforwarddb -table sometimes fails on VMS depending on when the previous ckpdb command was executed.
Create table statements for tables containing nvarchar or nchar columns sometimes fail with E_AD1014_BAD_VALUE_FOR_DT and E_PS04A8_UDT_LTXT_CONVERSION.
The function random(x,y) fails if abs(x) + abs(y) >= 2147483648 with "E_US1068 Integer overflow detected in query." randomf also fails.
Creating a database procedure using a nvarchar(15999) parameter fails with: E_OP0791 and E_AD2005
When "no" is used as a correlation name (e.g. "select * from t1 no ...") it is incorrectly flagged as a reserved word violation.
If a negative length parameter is coded in a substring function various symptoms occur, including server program exceptions. It should simply return an error.
If the "for
Upgradedb fails to set correct permissions on iiextended_relation table when upgrading a database.
Processing of certain long inserts causes ULM memory corruption in PSF which may later result in program exceptions in PSF.
When II_CHARSETXXX is set to KOREAN, some characters are not displayed correctly under windows platform when application uses JDBC driver.
The ODBC cannot connect with EDBC servers. The servers report an "invalid decimal character" error.
Setting default_page_size to a size for which the respective DMF cache has not been enabled causes the default page size to revert to 2K. This is correct. However, the user is not notified in VCBF when this is the case.
Pressing the Cache function key in cbf, or using the Cache page for a DBMS component in VCBF, when the cache_sharing parameter has been made derived in dbms.crs, causes an error.
Xmlimport fails if the data is a varchar column with only '' (empty string) in it or if it contains '&', Errors E_CO0040 and E_CO002a are given out.
Xmlimport fails if the data in the varchar column contains a '&' character. Error E_CO0040 and E_CO002a are given out.
Terminal monitor truncates long error messages.
When choosing, in IVM, to be alerted on messages that actually occured only earlier than what corresponds to IVM preferences (for example, choose to be alerted only on E_GCN0152_GCN_SHUTDOWN, and select (messages) "since last name server startup" in the preferences, the installation being started), a "You have unread Alert messages" message will incorrectly appear upon IVM startup or preference changes (assuming the "Message Box" checkbox has been left checked in the preferences)
On certain configurations, the precision and scale controls in the VDBA SQL/Test preferences are not wide enough, and don't allow to display the maximum value that can be entered (i.e. 255)
In some cases, VDBA fails to display any rules under the Rules Executing
Procedures branch of Procedures, when rules do in fact exist, and it also
fails to display the procedures associated with a rule under the
The Unicode Enabled check box is incorrectly blank on a unicode enabled database when you try to Alter the database in VDBA.
Some checkboxes are truncated in ivm panes.
When creating the transaction log files, VCBF reports an error when the total size is less than 16 Mb. The error should be if EACH file is less than 16Mb (the total being the file size multiplied by the number of partitions)
VCBF proposes, for the logging system, some parameters (such as log_file_1, log_file_name, log_file_parts) that are not proposed in CBF and should not be edited
Open a VDBA monitor window, press the "Log Information" or "Lock information" tree branch, and save the environment. Then Load the environment, and press the (since) "startup" button: VDBA crashes with a GPF
In VDBA SQL/Test windows, when selecting a database and then immediately loading a query script, the GO button is grayed. It becomes ungrayed if clicking in the query text area, but this should not be required.
When using vdba to copy a table from one database to another where the table has a multi-column key, the column order of the key in the target database can be altered
The "Raw" transaction log information is not available on UNIX versions of VCBF
When trying to import a CSV file where the first line ends with the separator sign (typically a comma), IIA crashes when pressing the "Next" button in Step 1 of 4
VCBF deletes the log directory when destroying the transaction log. It's then not possible to recreate it in cbf. VCBF should not delete the directory, in order to be consistent with CBF
Altering a Vnode attribute in VDBA or INGNET results in deleting the attribute instead of altering it.
Some Ingres variables, such as II_HALF_DUPLEX or II_STAR_LOG, are documented in the System Administrator's guide, but are not proposed nor visible in the IVM "Parameters" right pane tab available when the "Ingres Installation" branch is selected in the tree.
RMCMD (Remote Command Server) with fail to start, and will return error E_RE0018 when II_TEMPORARY is defined in terms of a JOB or PROCESS level logical (E.g. SYS$SCRATCH).
When stopping an Ingres installation through IVM (the "Ingres Installation" tree branch being selected), the "started" checkbox is not refreshed in the status right pane once the stop operation has completed
When altering a database in VDBA, VDBA crashes when scrolling to the end of the "extend" locations list (assuming there are enough "extend" locations to generate the scrolling), and clicking below the end of the list
When trying to copydb/unloaddb for database that has procedures with decimals or floats and II_DECIMAL=, the script created outputs errors when database is being recreated.
In IJA, select a database in the tree (for activating the right pane), then select the "nodes" branch, press the "refresh" button and select "Current and All sub-branches". Then select a database, press "view" in the right pane and try to redo a transaction; the (On) "node" combobox is empty (it should contain the vnode list) (The problem disappears when restarting the IJA executable)
In the VDBA Printer preferences dialog, the icon that represents the Portrait Orientation doesn't fit in the space provided . The top and bottom are slighly cut off.
Issuing remote commands via Visual DBA against a 64-bit Server will appear not to work, even though in the background the commands execute successfully. This has been fixed.
xmlimport fails with E_CO0032.
Syscheck fails with '-1 symbols not found' when running the 32 bit version on 64 bit Solaris.
ESQLCBL COBOL pre-compiler generates CALL "IILQSHSETHANDLER" statements with corrupt function names.
Precompiler output for the ADA language (from eqa and esqla) is not properly tailored to the needs of the GNAT Ada compiler available on Linux; the '-agnat' option needs to be supported.
The Ada precompiler for release 2.6 generates code like this: "while (IILQprsProcStatus(0) /= 0) loop"; the IILQprsProcStatus parameter is required for row producing procedures. However, the Ada function prototype in the eqdef headers was not adjusted to account for this new parameter, causing Ada compiler errors.
Embeded Fortran programs would abend with a SEGV.
Errors returned whilst running Ingres/ICE HTML variable dynamic SQL queries, when the SQL query contains HTML variables in the WHERE clause.
Fixed loading shared library error during Apache server startup.
Based on SP2, upgrade Ingres 2.6 SDK from version 1 to version 2.
During an upgrade, message "Create rmcmd objects failed:" may be displayed due to the current versions of the rmcmd objects being already installed.
Ingbuild failed with "gtar failed" message.
We now use the CORPQA license tar.Z file directly for installing licensing on UNIX platforms. (INTERNAL)
mkimage fails to create a new save set from existing Ingres save set with 'Can not determine structure within source save set'
Upgrade failure between 2.0 and 2.6/0207 caused by the mismatch of transaction log file sizes.
For double byte Ingres, if install the transaction log in a location that has more than one directory and has local character embedded in any of the directories, rcpconfig will fail. This has been fixed.
64-bit Tuxedo is not yet supported on Solaris, AIX, and HP-UX. It should be removed as a package from the install.
Upgrading a 2.0/0011 networked client installation to 2.6 fails with iisunet reporting the following errors: %DCL-W-ACTIMAGE, error activating image II_COMPATLIB -SYSTEM-F-SHRIDMISMAT, ident mismatch with shareable image
Attempting to install JDBC in a VMS client installation fails if done as part of the IVP or if iisujdbc.com is run right after running iisunet, without first doing an ingstart or logging off and on. The error is seen after the message "Generating default configuration..." and the VMS errors are %DCL-W-ACTIMAGE, error activating image II_COMPATLIB, and -CLI-E-IMAGEFNF, image file not found.
Terminal monitor SEGV's during iisudbms on RedHat 9 Linux.
After installing ONLY the 32-bit components from a 64-bit saveset, if one chooses to later on install the 64-bit components the process deletes the 32-bit libcompat.1.xx library. This has been fixed.
iilink no longer fails for DBMS Standalone Server installs
Bridge server on VMS shows tcp_ip as an available protocol. Only tcp_dec and tcp_wol should be shown. In addition, tcp_dec and tcp_wol should have distinct listen addresses.
When inserting a row containing a date, time, or timestamp against a gateway, if the statement was prepared, it fails with error "Numeric value out of range" when the statement is executed repeatedly.
ODBC driver does not connect when null arguments are supplied for the UID and PWD arguments in SQLConnect().
ODBC driver registers an error when SQLGetDiagRec() or SQLGetDiagField() are called with null arguments for the error message text, but valid arguments for the sqlstate and error buffer length.
ODBC reports "Error in assignment" error when fetching a decimal column that is coerced into SQL_C_LONG.
Committing a transaction to a replicated table whilst the distribution thread is active, may cause the transaction not to be replicated until the database has been closed and reopened.
II_ALLADT_BUILD.COM builds the IIUSERADTxx.EXE shared library, using the supplied SPATIAL object types, but doesn't include the Demo UDTs.
TRACE POINT DM420 sometimes displays negative values for FIX CALLS and other fields.
lockstat sometimes displays negative numbers in values listed under the section titled "Locking System Statistics".
SELECT dbmsinfo('transaction_state') always returns 1. Please note that after applying the fix for this bug, Ingres will not open a transaction for sigleton SELECT statements which only reference inquire_ingres() or dbmsinfo().
If optimizedb is used with the -zs# option on several tables whose names aren't unique in the first 16 characters, it may produce E_OP094C errors ("Could not create table for sample data").
The value for server_pid in ima_server_sessions gives the PID of the current server thread not the server itself.
Extreme min/max values in decimal sequences may cause unanticipated decimal overflow and server termination.
E_DM9043, E_DM934D, E_DM942B, E_DM9306, E_DM904A_FATAL_EXCEPTION doing select MAX(col) if the column is part of a btree index and there is a lock timeout or lock deadlock.
Virtual memory usage by the iidbms increases after modifying a table with persistent indexes, or if creating indexes using the parallel index syntax.
In cbf and vcbf, the user is unable to remove the value of a parameter which has a null default value
E_DMA434_LGK_VERSION_MISMATCH may occur during recovery process (RCP) startup if a concurrent non-server process (E.g. lockstat ) is simultaneously connecting to and creating the logging/locking shared memory segment.
mkrawlog.sh doesn't handle the parameter log_file_parts in config.dat correctly, after raw log file is deleted manually from $II_SYSTEM/ingres/log. After raw log file is deleted manually config.dat still contains information about it, but cbf doesn't think the raw log exists anymore.
E_QE001D General failure in QEF ... doing update where tid = ... or tid = ... The errlog.log also contains the error E_DM002A_BAD_PARAMETER.
Two sessions hang inserting into a table having a blob column. One session is waiting on a page lock in the blob extension table, the other has cs_state: CS_MUTEX, Mutex: TCB et (tablename).
On Linux ipcclean does not catches both versions of 'resource deleted' from ipcrm.
E_OP0901_UNKNOWN_EXCEPTION reported for a SIGSEGV in opj_traverse() when running a query containing many ANDs and ORs in the WHERE clause. Errlog.log also reports E_OP0082_UNEXPECTED_EX. Stack: opj_traverse listed repeatedly and then followed by opj_normalize, opj_jeqc, opj_joinop, ops_statement, ops_dbprocedure, ops_sequencer, . . .
QUEL & SQL command "SET TRACE OUTPUT 'file_spec';" allows a user to
overwrite an aritrary file on the sever as 'ingres'.
Remedy for this adds, and tests for, a distinct initial line within
'filespec'. Setting 'filespec' to an existing file without this
signature will return error "E_US1764 SET TRACE : Can't open file or
terminal 'filespec' for set trace command.", so trace logs created
prior to the application of this fix will need to be manually removed.
A select in a for-loop in a procedure in which an order by clause contains the same column more than once, and that column does NOT appear in the select-list, will cause an exception (SIGSEGV) in QEF. SEGV may also happen when order by in dbp select statement references column not in select list (or redefined by "AS" clause).
A query with more than one subselect in its where clause, one of which has an ifnull function with an aggregate function parameter in its select list, may drop valid rows from its result set.
E_OP08A2 errors may be seen in queries that have restrictions on histogrammed character (char, varchar) columns in a database with an alternate collating sequence.
A query with a CO join (possibly the result of a distinct operation) may return a QE0018 error.
Dbms server is doing unnecessary page locking on DMPE Temp table.
The following errors could occur when row locking: E_CL1001_LK_BUSY, E_DM9C01_FIXED_BUFFER_INFO, E_DM942B_CRASH_LOG, E_DM9289_BM_BAD_MUTEX_PROTOCOL or E_CL1001_LK_BUSY, E_DM93B2_BXSPLIT_ERROR, E_DM9264_DM1B_SEARCH, E_DM925E_DM1B_ALLOCATE These errors can then trigger the following errors in other sessions: E_DM93F6_BAD_REPOSITION, E_DM9261_DM1B_GET and/or E_DM968E_DMVE_BTREE_FIND, E_DM968F_DMVE_BID_CHECK, E_DM9509_DMXE_PASS_ABORT Another symptom is sessions hanging with one session holding a page buffer mutex while waiting to escalate from row to table locking.
Long_byte and long_varchar functions do not work correctly when udts are loaded into the server.
'GRANT idle_time_limit
The blob insert optimization added in SIR 97558 is not done when the table name in the INSERT query is in upper case, and the table name in iirelation is in lower case.
Attempting to checkpoint a database whose data files were created in an alternate location, and that alternate location contained embedded spaces, fails with E_DM1108_CPP_OPEN_DB. The errlog.log shows E_CL0610_DI_DIRNOTFOUND, E_DM9004_BAD_FILE_OPEN (whose displayed path is wrong), E_DM923A_CONFIG_OPEN_ERROR, E_DM9268_OPEN_DB, E_DM1108_CPP_OPEN_DB. This has been fixed.
When the -zs or -zss options of optimizedb are used to create a sampled histogram of more than 50%, a E_OP0972 error may be observed.
A complex query with 2 pairs of outer joins themselves joined together, and a where clause with an "is null" predicate OR'ed to a subselect may cause a SEGV in OPF.
An outer join query with a restriction predicate in the ON clause (an "is null" predicate in the problem query) and an unflattenable subselect in the where clause (a "not exists" query in the problem query), can cause a SEGV.
A row producing database procedure (RPP) with a UNION SELECT in which the INTO list includes NULL-able variables, and the corresponding SELECT list entries are not NULL can cause program exceptions in the DBMS server, the JDBC server, GCA or OpenAPI when a return row statement is issued with the local variables.
Auditdb -b flag does not print the transactions which are started before the begin time but have ended after the given begin time. Also Auditdb -all case, gives incorrect results, when -b flag is specified. All the records before the given time, except the BT and ET records are printed out.
E_DM9049_UNKNOWN_EXCEPTION, SEGV and stack dump doing recovery (rollforwarddb or rollback) of an insert to a table having a rtree secondary index.
Fixed SIGSEGV error when DBMS cache_sharing is selected in cbf.
Issuing a "select dbmsinfo('session_id')" on 64-bit platforms returns only 8 characters of the 16 character session ID. This has been fixed.
On 64 bit platforms execution of TRACE POINT DM101 fails with repeated exceptions (SIGSEGV). Similarly, internal errors which trigger a "dmd_check" will fail to dump diagnostic information.
SIGSEGV in MO method LK_lkb_llb_get() when dereferencing a stale lock (LKB) using IMA.
An outer join that uses the hash join technique may not return outer join rows (outer rows with no matching inners) when the inner join source is significantly larger than the outer source.
A join that uses the hash join technique and involves some use of disk overflow has a very small potential to return 0 result rows.
An outer join query with at least 5 tables plus applicable secondary indexes in which at least one of the outer joins can be satisfied using a secondary index in place of the inner base table, and which is compiled using the new 2.6 SP2 OPF enumeration heuristic, may produce the E_AD2004/E_OP0783 error pair.
E_DM9238_LOG_READ_EOF during transaction recovery, exacerbated when using a partitioned log file.
Copydb/unloaddb writes wrong syntax within alter key in copy.in file, when doublequotes are used in a long column name.
TID joins currently process the rows in whatever sequence the TIDs arrive from the subplan below. In most cases, this sequence bears no relationship to the physical order of the rows in the target table. For Tjoins on large numbers of rows, this means the rows are read randomly and the join may perform up to one disk read for each. This SIR proposes to presort the TID list into TID sequence to reduce the amount of disk access required for a Tjoin.
When upgrading a database with a large number of tables, 'upgradedb'
may echo an error similar to:
/bin/sh: /bin/chmod: arg list too long
Exact wording of the message may vary between UNIX operating systems.
A query may fail with error "E_OP0681 consistency check" if it contains: - more than one IFNULL with an aggregate operand OR - one IFNULL with an aggregate operand and another aggregate operand in the query
A "create table ... as select ..." query with some joins (or possibly aggregations) and executing in a high volume concurrent environment with "autocommit on" may result in SEGV or BUS errors in dmh_open.
Complex queries (5 or more base tables and 10 or more tables + indexes) may use the new heuristic for compilation. Instances of E_OP04C0 and E_OP0889 have been observed on a small number of such queries. Withhold error from user when CBF parm (or "set joinop newenum") is on. Windows fix for ifdef 0. Fix opf_new_enum CBF parm.
An aggregate query in which a Unicode column is in the group by clause may cause an endless loop generating the same result row.
A "select distinct" query in which the select list contains columns of an unsortable type (e.g., BLOB, UDT) may produce more result rows than expected (duplicates may NOT be removed).
Using "auditdb -t" to create an audit.trl file of a table owned by the user who is executing the auditdb command fails with errors which describe that the table does not excist. This has been fixed.
COPYDB SIGSEGVs when an individual command line element exceeds 64 characters.
Upgradedb gets E_US0014 Database not available.
The blob insert optimization added in SIR 97558 is not done when the statement is a prepared insert statement.
E_US125D Ambiguous replace error is being issued when updating a btree table with values from another table.
Sessions hang in LK_EVENT following a failed group read, including the session that encountered the read error.
When a User is created with own priviledges and profile, privileges granted by profile are ignored.
There is a table, based on the table, there are two views, and based on the two views, there is another view. If you add a new column to the table, you will get an error saying that table no longer exists when you drop the column. This has been fixed.
Selecting against IMA table 'ima_lgmo_lfb' incorrectly returns zero rows. This prevents the VDBA performance monitor from displaying information about the transaction log header and log file.
Copy Statements in fast load copy mode fails with E_US14E4 error if row level locking is set.
Errors in REDO recovery or in rollforwarddb when redoing updates to a compressed btree secondary index: E_DM9388,E_DM9390,E_DM93E3,E_DM968E,E_DM968F.
In CBF, the user is able to execute menu items that are not displayed on the menu bar of the Configure screen using a keypad.
Display of session ID values in IMA tables is inconsistent with Ingres
utilities which display session identifiers, in that IMA will display the
value as a decimal number, while ipm, and iimonitor will display this in
hexadecimal. The normal work around of converting the returned value to
an integer, and then displaying the integer's hex value (E.g. SELECT
HEX(INT4(session_id))) FROM ima_server_sessions) does not work on 64 bit
platforms since the session ID will overflow a four byte integer.
A new configuration parameter (hex_session_ids) has been provided to allow
columns in IMA tables which contain session identifiers to return these
values using hexadecimal notation. Columns effected by this parameter are:
System Component Table Name Column Name
DBMS Server ima_server_sessions session_ptr
DBMS Server ima_server_sessions session_id
DBMS Server ima_server_sessions_desc session_id
DBMS Server ima_server_sessions_extra session_id
DBMS Server ima_locklists locklist_session_id
DBMS Server ima_log_processes force_abort_session
DBMS Server ima_log_processes group_commit_session
DBMS Server ima_log_transactions tx_session_id
ICE Server ice_user_transactions trn_session
This parameter may be added to your config.dat by executing "iigenres
`iipmhost`". Value for this parameter can then changed using cbf or
vcbf with the configure option for the listed System Component.
The 2k page size cache cannot be turned on and off in VCBF
Issuing 'stop server' in iimonitor or a ingstop -f will cause the server to hang.
With the changes for b110332, a row locking session may go into a cpu loop.
E_DMA43F_LG_SHMEM_NOMORE or E_CL102D_LK_EXPAND_LIST_FAILED errors may appear even though the logging/locking resource limits have not yet been reached.
Massive 'IN' clauses can cause server to crash silently
IMPXML SEGV's trying to close XML input file.
QSF memory is gradually consumed until an "out of memory" failure occurs. This submission writes to DBMS log under control of tp qe128 and seems to magically reduce impact of QSF problem. (GENERIC) The execution of a row_producing procedure that returns enough rows to require more than one GCA_TUPLE (pre-fetch) buffer will incorrectly leave a positive lock count in the QSF object descriptor for the cached query plan. This causes the query plan to become "stuck" in QSF. Many such plans will eventually fill the QSF cache with objects that cannot be removed and will crash the server.
If a cache is saturated with modified pages from a single table and the table is closed, the server may hang waiting for free cache buffers.
In a VMS environment with a heavy load of IMA queries, the session(s) issuing the IMA queries may become unresponsive (while other DBMS sessions continue their work). In iimonitor, the hung session is shown in state CS_COMPUTABLE with CS_EDONE_MASK. Starting an IMA query from another session may cause the session to be in CS_CNDWAIT state.
When running lockstat -statistics the output contains no information for the "Locking System Statistics" section.
OPTIMIZEDB fails with E_SC0206. The error log shows a SIGSEGV in memcpy() being called from qes_skbuild(), qen_tsort() . . . Problem occurs during a TSORT.
A rule which is fired after a delete loops unexpectedly when a subsequent rule inserts rows into the table against which the original delete was issued. The delete will typically fail with: E_US1068 Integer overflow detected in query.
VCBF doesn't write ii.$.rcp.log.log_file_parts information in config.dat for being taken into account in the calculation of the total number of log_writer threads
The JDBC Server overwrites BLOB (Binary Large Object) data with the value from the last BLOB when inserting or updating 2 or more short BLOB values in one statement.
The JDBC client does not map the II_CHARSET definition ISO885915 to a valid Java encoding.
The JDBC driver throws "ExceptionInInitializerError" when running in the Sun 1.4 browser plug-in for Internet Explorer.
The schema names returned by the JDBC meta-data method getSchemas() are now in the proper form to be used as parameters to other meta-data methods.
JDBC driver protocol errors occur when closing a result-set which contains a BLOB that was not accessed with a getXXX() result-set method.
The JDBC meta-data method getPrimaryKeys() will now return keyed table structure information if no explicit primary key is declared for the table.
The Ingres JDBC driver result-set method getString() will now return a zero- length string for date columns which contain an Ingres 'empty' date value. A data truncation warning is also generated which can be accessed with the getWarnings().
The JDBC driver registers an exception when executing a database procedure with one input or BYREF parameter when it returns a value.
A multi-threaded JDBC application may receive a "NumberFormatException" when retrieving dates or timestamps.
The JDBC driver now supports UTF8 character set for query text, query parameters and query results.
A segmentation violation would occur in the JDBC server. The error message displayed was "Connection not established!, Message: ca.edbc.util.EdbcEx: Unable to establish connection due to communications error." A core dump was generated.
OpenAPI will now correctly retrieve char() and varchar() columns which are longer than 32,000 characters in length.
OpenAPI now requires initialization at IIAPI_VERSION_3 and connections at IIAPI_LEVEL_2 before Unicode data types are permitted to be used.
The OpenAPI null indicator for non-nullable columns contains random garbage.
4GL call frame/report commands can fail with error E_RW100A when passing parameters via the ut.exe system. This is caused by strings being incorrectly delimited on some platforms.
Reports display asterisks when value would still fit. With a mask of zz,zzz,zzn.nn- we are seeing ********** even with a value of 14,693,278.80- which should be fine. The value is displayed correctly in Ingres 6.4.
Backslashes in parameter values for a report result in E_RW1001.
ISQL is taking significantly longer to complete a query than it did on a 6.4 installation.
ABF applications with implicit integer to money conversions cause an E_AR0003 message to be printed.
ISQL will suffer an exception (SIGSEGV) if you complete a query, save results to a file, and then page down.
When compiling a C++ file with passing integer parameters to a procedure, can receive: "C2664: 'IIpLQprvProcValio': cannot convert parameter 7 from 'int *' to 'char *'. Type pointed to are unrelated; conversion requires reinterpret-cast, …"
A 4GL frame registered to receive a DB Event hangs when a DBevent is received. The DBevent is not processed until another DBevent is triggered. When another DBevent has been activated, the DBevent is processed and the frame becomes unusable.
The money template of +'$$$$$$$$$$$$n.nnCr' does not append two trailing spaces for positive values as documented in Section 18-20 of the Using Character Based Query and Reporting Tools guide. As a result, it is not possible to align the decimal point.
REPORT does not respect user customization to "param" in utexe.def in which '(%S)' is changed to (%S) to allow passing of parameters within double quotes.
When compiling an ABF frame which utilizes the CommandLineParameters() 4GL function as if CommandLineParameters() = 'NOPARAMS', the compilation fails with E_OS0010.
RBF: Report fails when attempting to use a DATE or C data type with the MIN, MAX or SUM functions in the report footer.
An entryfield of data type 'Integer' with InputMasking and a formatstring such as +nnnnnn or nnnnnnnnnn/n accepts only one number. Attempting to enter more numbers will cause OpenROAD to 'beep' and new numbers are not accepted anymore.
RBF does not show ANY and COUNT as aggregates for a C data type.
Scrolling up (using the scroll down key, usually mapped to PageUp) on Windows platforms behaves differently than on UNIX. On a form that is longer than two screen full, instead of scrolling a screen full minus one line, on Windows VIFRED scrolls up about a screen full and a half and places the cursor in the middle of the screen (on UNIX the cursor is placed at the top of the screen.
Concatenating VARCHAR variables and columns in excess of 2044 characters causes error E_RW0027_size_too_large.
Some values are too long to be displayed in the units column of the parameter table in the configure screen when using CBF.
ISO 8859/15 (Latin + Euro sign) has been corrected from IS885015 to IS885915.
Unload operations such as COPYDB and UNLOADDB, in character mode, will now produce copy.in and copy.out scripts which are generic to platforms.
Pressing the Cache function key in CBF, when the cache_sharing parameter has been made derived in dbms.crs, causes an error.
"administrator" is now created as a privileged user when IIDBDB is created.
The help text for the SQL SET command is incorrect in both SQL and ISQL. It does not match what is said in the SQL Reference Manual.
Non-numeric values can be specified for JDBC configuration parameters which do not have an initial value but are supposed to be numeric values.
INGSETENV does not set characters from non-ASCII ranges in the symbol.tbl file when the code page is set to SLAV852. The character 9F, which is a small letter c with a caron above in code page 852, is converted to small letter c only. This causes a problem to Slavik users as they are unable to set variables to a value which uses native characters.
XMLIMPORT gives "E_CO005 COPY: can't open file …" error when trying to import data from an XML file which contains only metadata information, such as one created by using "genxml -metadata_only".
Displaying E_US100D (and other long error messages) can cause the Terminal Monitor to abort with an exception (SIGSEGV) when referencing long table names.
OPTIMIZEDB fails with a memory exhausted error (E_OP091A) when optimizing tables which contain compressed NULL-able VARCHAR columns.
VisualDBA and Ingres/NET now support the new DB2UDB Enterprise Access Gateway. IVM and VCBF now support the new DB2UDB Enterprise Access Gateway.
VisualDBA can now be used to populate a database.
VisualDBA reports SYSTEM ERROR 8 when opening Nodes -> <name> -> Servers -> INGRES -> Users -> <userid> -> Open Windows. The correct answer would be <no open windows>.
A small number of Ingres CL messages weren't managed correctly by the IVM preferences when specifying notification levels. They couldn't be drag-and dropped in the dialog, and weren't affected if the category they belong to was set to a given state. Affected messages are those where the error number is not followed by the two character CL facility string. E_CL251D, E_CL251E, E_CL251F, and E_CL252F have been identified as having the problem.
"copydb dbname -add_drop [table_name ...]" generates a copy.in script which completely omits the CREATE statement(s) for the table(s).
The list of privileges is now displayed correctly in the VisualDBA Create Profile dialog under Windows 2000.
Gateway branches are now displayed correctly in ivm and vcbf.
The fix to problem INGDBA0150 inadvertently changed the behavior of COPYDB to always emit table related statements if the '-add_drop' parameter is specified. COPYDB should only implicitly emit table related statements, if no explicit scope is specified on the command line.
Gateway start/stop messages in the errlog.log don't appear at the right places in IVM (i.e., when the gateway branches are selected). They appear at various other places, under an incorrect component.
In the VDBA Performance Monitor, when trying to start a replication server (or selecting the Startup Settings tab) before the runrepl.opt file has been created, there is a misleading "Permissions Error [...]" message instead of the "file not found" error.
VDBA: SqlTest returns a "cannot open cursor" message instead of providing the SQL error when trying to execute "SELECT * FROM tab1 WHERE colx1=N1", where N1 is a numeric column.
VDBA: The width of the SQL/Test database combo-box has been increased, in order to allow displaying 24 (average) characters.
When creating / altering a User or Role definition with VisualDBA, there is now password checking between the password and "confirm password" controls.
A REPLICATOR server now can be started from VisualDBA if not logged under the Ingres user id.
There is no more problem in VisualDBA if, in [V]CBF, the Name server local_vnode parameter is different from the host name.
When using the Tables option in the VisualDBA "rollforward" or "checkpoint" dialogs, the execution of the command no longer fails if more than one table is selected.
When the DBMS is installed with the SQL92 compliant option, VisualDBA no longer fails when trying to launch a remote command.
In the VisualDBA properties pane for a user, and in the Alter User dialog box, the Ingres user now appears as having the RMCMD privileges.
Trying to execute SYSMOD from VDBA against an SQL/92 Ingres instance fails with an error: "This Command Can Only Be Launched by the Ingres Account".
Stop options are available in IVM for some (but not all) services, such as Remote Command and ICE servers, when they have already stopped.
Text is being truncated in various VDBA dialog boxes.
VDBA now provides an appropriate error message when trying to drag and drop a gateway table into an Ingres database, if a column of the gateway has an unsupported data type.
IIA now works correctly when specifying the "[...]identification on first kB of file" option with a DBF file.
IIA proposes duplicated solutions when importing a CSV file where the fields are quoted.
The Relocatedb, Infodb, Fastload, and Alter Database VDBA features are not disabled when connecting to an Ingres 2.0 server, and instead are failing with a "Command not Accepted by Server" error.
In the IVM, or VDBA About box, the pop-up which is seen when the cursor is over the CA logo will appear briefly every second or so even when the cursor is over the text portion of the dialog box.
IJA on UNIX writes out "^M" at the end of each line in the generated SQL file.
VDBA: Ctrl-V and Ctrl-C do not work correctly in the edit fields of table name and table column in the Create and Alter Table dialog boxes.
The With Grant Option feature does not work in the VDBA Grant Procedure dialog box accessed by selecting a procedure in a DOM window, and choosing Grant in the right click menu.
The VDBA Grant Procedure dialog box, available when selecting a user, expanding grants/granted Procedures, and choosing Create in the right click menu, is failing with an error Grant procedure failed. (This dialog is not the same as when selecting a procedure and choosing "Grant" in the right click menu.)
VDBA: In the monitor tree, if the selection is on the last branch and the searched text is in an upper sibling branch, then the Find operation will not find the tree item successfully.
In the VDBA Copydb dialog box, if all database objects are unchecked in the list, OK button should be disabled.
VDBA: No help is available through F1 key in the Force Refresh dialog.
Context help for the VDBA Create or Alter CDDS dialog box brings up an incorrect help topic.
Some checkboxes are truncated in IVM panes.
VDBA: Duplicate short-cuts, and minor format errors in the tool-tip text in various tools, menus and toolbars.
The DOM window for Replicator in VDBA displays the Replication version as 2.5 for Ingres 2.6 Replication.
Changing directories in the save environment dialog in VDBA does not work.
The VDBA 'About' box does not correctly display on Solaris.
Error message S_ST0533_reformat_log does not fit in the VCBF message box.
IIA does not show usage message when passed an invalid flag on the command line.
IIA will now import data on UNIX.
You can use the y-umlaut (0xFF) character for table names within VDBA.
A report called from ABF or VISION produces a core dump file when there is a syntax error in a SELECT clause.
The SQL Fortran pre-compiler incorrectly emits code for maintaining SQLSTATE if the program source contains a character variable within it's DECLARE section whose first six characters are 'SQLSTA' (Case insignificant).
Linking an embedded COBOL program which uses a BLOB data handler fails with errors "LNK2001: unresolved external symbol _IILQlpd_LoPutD, _IILQled_LoEndD, _IILQlgd_LoGetD".
Compiling ESQLC programs which include "EXEC SQL INQUIRE_SQL('attribute')" statements to return a numeric error number will produce errors or warnings similar to: "warning: passing arg 5 of 'IIpLQisInqSqlio' from incompatible pointer type".
Pre-compiling a COBOL program with a database procedure that produces no rows, always returns "E_EQ0317 There are 0 argument(s), statement requires more".
Micro Focus COBOL programs calling Ingres database procedures may terminate with an exception (SIGSEGV).
The ADA pre-compiler does not recognize short_short_integer as a declared type on Linux.
Pre-compiler output for the ADA language (from EQA and ESQLA) is not properly tailored to the needs of the GNAT ADA compiler available on Linux; the '-agnat' option needs to be supported.
Executing the command "ingstart -gateway_type" where gateway_type is oracle, rdb, sybase, mssql, or informix results in the message "No definition found for specified configuration name".
Executing the example database procedure from the Ingres/ICE Example Database Procedure page causes a 'partner abruptly released association' error message.
The Ingres/ICE C programming API returns no error when attempting to update or retrieve an association between a business unit or role.
The Ingres/ICE C programming API returns no error when attempting to update or retrieve an association between a document and a role.
When inserting a database definition into the Ingres/ICE repository using the Ingres/ICE C programming API, no error is returned when an invalid database user is associated with the definition.
The image used on the login screen of the ICE Administration Tool requires updating to reflect new branding.
Registered files password.html and user.html in the ICE Business Unit "icetutor" do not exist.
Creating an ICE web user with an invalid negative database user id can cause an exception in the ICE server process.
Using the ICE administration tool to view the "iceTools" Business Unit source file buMMenu.html fails. Login to the ICE Administration Tool as "ingres", expand the Business Unit Environment, iceTools, pages. Select buMMenu.html the error "unable to open file" is returned.
Using the ICE administration tool to view the "tutorial" Business Unit source file if5.html fails. Login to the ICE Administration Tool as "ingres" and select Business Unit Environment. Expand tutorial, pages and select if5.html the error "cannot open file" is returned.
Unable to open the page displayed in the Error Management branch of the ICE design tool. Login to the ICE design tool as a user with rights to execute the application, and then select Helpful, Error Management. A page image with no text is displayed.
Attempting to delete a profile role association using the ICE C API providing an invalid role identifier does not return an error.
A user is unable to view the romance.gif facet in the ICE Administration Tool.
A user with the correct permissions is unable to define an ICE server location using VDBA. Also, defining an HTTP ICE location using VDBA causes an exception in the ICE server.
Multiple simultaneous connections from an Ingres/ICE client running from a web server causes an exception to occur in the ICESVR process resulting in a 'Client abruptly released association' error.
Ingres/ICE server variables are not displayed correctly in the Web ICE Administration Tool servar.html page.
ICETranslate: Incorrect ICE 2.0 syntax is generated for equality testing. No support for testing that a variable is defined.
Attempting to delete a location associated with a business unit using an invalid location identifier with the Ingres/ICE C API results in an abruptly released association.
The Ingres/ICE server does not start if the installation is configured for SQL92 compatibility.
Running a report from ICE will fail if a single quote is used to pass parameters.
There was a shared library loading error during Apache server startup.
The configuration of an Apache web server for use with the Advantage Ingres Web Deployment Option can be simplified by providing the Advantage Ingres Web Deployment Option configuration file that can either be included in the Apache configuration file or appended to it.
The ICE server databases are not being upgraded when performing an upgrade of an Ingres instance.
Installing Advantage Ingres SP1 on the top of Advantage Ingres 2.6/0201 fails if the previous setting for II_SYSTEM contains a hyphen.
The 'OK' and 'Cancel' buttons overlap in an IVM dialog window.
When installing Advantage Ingres on a Solaris machine with the local language set to Italian, IIDSFREE fails. The symptom is displayed as insufficient space for the transaction log file. It is caused by "df -k" returning a column headed "Kbyte" but the SED command in the script is looking for "Kbytes".
The Getting Started Guide has a new font which will not make it look "fuzzy".
XML cannot be displayed from the dynamic SQL console. The message "Cannot view XML input style sheet. Please correct the error and then click the refresh button, or try again later" is returned.
The Installation guide has been replaced with the Getting Started Guide.
The messages.txt file will include all of the latest error messages.
Upgrade failure between Ingres 2.0 and Advantage Ingres 2.6/0207 caused by the mismatch of transaction log file sizes.
For double-byte versions of Advantage Ingres, if the transaction log is being installed in a location that has more than one directory and has localized characters embedded in any of the directories, RCPCONFIG will fail. This has been fixed.
Cannot find the help files for some Advantage Ingres GUI tools (i.e. WINSTART, IILINK) when pressing F1.
Successful installation of Advantage Ingres 2.6 is dependent upon the existence of a C compiler.
Performance over Ingres/Net on Windows is significantly slower (approximately 4x) in Ingres II using OpenROAD 4x (or other applications) against gateways or Ingres than it was using OpenROAD 3.5 and drivers. Also, response time in the application over Ingres/NET is much slower (2-10 times) than running locally. The response time differences are most notable in batch-like operations (e.g., an application that runs 1000s of simple updates or queries).
The GCC server may crash if receiving an incoming connection from a Windows client.
Attempting to connect to Ingres when II_ TIMEZONE_NAME is set to an undefined time zone causes "E_FORMAT: Couldn't look up message 9033d" to be reported.
The GCC server is not building the GCA_ERROR message correctly (for GCA_PROTOCOL_LEVEL higher than 2). For example, when a remote connection is lost, the local session gets a GCA_ERROR message for error 'E_GC2210 Session Failure. Local abort condition' with error code set to 0 instead of the appropriate non-zero generic error code.
The GCC server process memory will not be reclaimed when executing a database procedure via JDBC where a procedure parameter has been registered as an output parameter without providing an input value.
Data corruption when joining Ingres and EDBC Server tables has been fixed.
Installations using the UTF8 character set will only be able to connect via Ingres/NET to other UTF8 installations.
Programs which use the UNIX system call seteuid() to change the effective user ID at runtime may have connections aborted due to problems verifying the user's identity.
When autocommit is enabled, the ODBC driver does not always recover from query errors.
The ODBC driver returns SQL_NO_DATA for catalog function calls as SQLTables or SQLColumns if the table_name or column_name contains an underscore and the length of table_name or column_name is 32 characters long.
The ODBC driver sometimes displays the wrong precision for numeric columns.
The ODBC driver does not map some extended ASCII characters according to the rules of ANSI.
The ODBC driver routine SQLSetConnectAttr() or SQLSetConnectOption() returns the error "Invalid argument value (sqlstate HY024)" if SQL_TXN_READ_COMMITTED or SQL_TXN_REPEATABLE_READ was specified, and the target database is earlier than Ingres 2.0.
The ODBC driver doesn't allow the application to set the transaction isolation level in the middle of a session, and returns error "Operation invalid at this time".
The ODBC driver doesn't recognize the ODBC escape sequence when used in query "DECLARE ..... User". This will generate either a hang or a DBMS server error.
The ODBC driver routine SQLGetTypeInfo() returns an error for the data type arguments SQL_TIMESTAMP, SQL_DATE, and SQL_TIME, if the driver is initialized as version 2.
The ODBC driver fetches data incorrectly for a table containing more than one large object. The contents of the last large object column in the row are displayed for prior large object columns.
The ODBC driver does not support SQL_C_NUMERIC.
SQLGetData causes a memory overlay while retrieving a BLOB to Unicode data type. This can result in various errors such as: "System.NullReferenceException: Object reference not set to an instance of an object" or "System.TypeLoadException" in .NET environment.
Gateway server class db2udb is not supported in the ODBC driver.
Sorting arrays of equal elements is extremely slow.
Provided support of the Trace Window functionality on UNIX platforms the same way as it is offeredon Windows.
Provided support of the Trace Window functionality on UNIX platforms the same way as it is offered on Windows.
REPLICATOR: The replication thread in the DBMS server may generate an exception (SIGSEGV) and bring down the whole server if a table level deadlock is encountered when trying to distribute a transaction.
Updating key columns of data that contain no corresponding shadow table entries results in inserting a second row into the target database.
Running ARCCLEAN in a SQL92 compliant installation results in error 'arcclean: iiindexes persistent expected 1 row, but found 0' while modifying a shadow table. If shadow and archive tables are not present, error message is 'Could not modify dd_input_queue – no iitables row found. Could not modify dd_distrib_queue – no iitables row found.'
REPLICATOR: REPMGR and the replication process hang periodically. Both have to be shutdown and restarted in order to continue processing.
REPLICATOR Server: In some cases Replicator may silently fail to replicate all transaction rows to a target yet the distribution queue gets cleared normally.
If Replicator and Enhanced Security are installed and enabled (also requires C2 Security), and the Replicator distribution queue manager threads attempt to move rows from the input queue to the distribution queue, the DBMS server crashes with the error "E_SX000D_INVALID_SESSION A SXF request specified an invalid session identifier".
Replicator: '-ORT' flag does not open the connection to a closed target.
The replication thread in the DBMS server may generate an exception (SIGSEGV) and die if a table level deadlock is encountered while opening the needed resources.
REPLICATOR: When the Repserver connection to a target database is lost due to the IIGCC going down, the Repserver shuts down due to EMX being reached. On an insert repeated error: "1571 -- D_INSERT Error executing procedure xxxxxxxrmi -- 13172742 -- API function cannot be called in the current state", the server shuts down. This is easily reproduced by the simplest fill peer <-> fill peer replication scheme, and killing the target IIGCC.
REPLICATOR: Sometimes, when handling a collision in LastWriteWins mode, the REPLICATOR server would fail with error 1614 - RESOLVE_MISSING_UPDATE or error 1588 - RESOLVE_INSERT. Examining the underlying base and shadow rows in the source and target databases failed to explain such behavior. At times, restarting the server could clear the collision.
Deadlock message shows that the deadlocked session already holding a lock on the resource. This might be seen with E_DM9042, E_DM9045 and E_DM9043.
The archive process fails to restart and issues message E_DMA464 which indicates an invalid record exists in the transaction log. The situation may be caused when Ingres is killed (possibly during operating system failure) after force abort when the previous consistency point occurred during the writing of a log record that spans log pages. Issuing LOGDUMP after the described force abort may also result in E_DMA464.
Blanks may be lost in strings containing many blanks when the strings are copied out and then copied back into a database via COPYDB or UNLOADDB/RELOADDB. For some strings, an extra blank may appear.
The DBMS server crashes with E_DM9C01, E_DM9301. Buffer involved is btree root, btree index or fhdr page. Involves two sessions of which one is readlock nolock and the other is using table level locking.
When running a program which has been linked against the Ingres libraries and the program does a fork() and exec(), file descriptors are left open.
Outer join on two tables returns more rows than it should when the query contains an 'IS NULL' predicate on an attribute from the inner relation in the outer join, and this attribute does not appear in the select list. QEP shows a right join.
IIGENRES, IISETRES, and IIREMRES do not log changes to config.dat in the config.log.
A Global Temporary Table id can be seen to be in use by more than one session. The sessions involved hang in a deadlock during an attempt to escalate to table level locking.
Unprotected updates to internal structures may very rarely cause the DBMS server to call abort() from iiCLpoll() causing the server to die without any related error messages. Later on the recovery process (dmfrcp) will report E_DMA469.
COPY out of a NULL-able money column using a 'WITH NULL (X)' clause when X is an integer value, will have this NULL value incorrectly treated as a regular value during COPY in.
E_CLFE07, E_SC022F, E_SC0123, E_CLFE06, E_CLFE06, CSMTp_semaphore failure (in DBMS log only), plus an exception (SIGSEGV) in QUremove() from scs_terminate(). Seen when a session dies (or is killed) unexpectedly.
The DBMS Server fails with an exception (SIGSEGV) in GC_send_sm() when Ingres/NET sessions are still connected and "ingstop –force" is run. INGSTOP hangs displaying 'Waiting for the logging system to go down'.
Query using a select with outer joins performs poorly. The query optimizer generates and chooses a bad QEP in which a sort is executed when a secondary index could have been used.
IMA: ima_server_sessions contains incorrect values in session_id (junk byte at end of each session_id) and ima_server_sessions_extra contains no rows.
A simple select query fails with E_OP048C_SORTCOST consistency check error.
Deleting via IPM a session that is waiting on the 'LG LCTX mutex' results in E_CL250F_CS_REQUEST_ABORTED, E_DM942B_CRASH_LOG, E_DM00F1_MUTEX_LOCK, E_UL0306_EXCEPTION, E_DM904A_FATAL_EXCEPTION. The DBMS server is terminated.
If II_THREAD_TYPE is set to INTERNAL, the DBMS server may under unusual circumstances loop endlessly within the group page buffer manager, causing sessions connected to the server to hang. No error messages are produced, and IIMONITOR and IPM will also both hang.
If you create a Global Temporary Table with the same name but a different structure to an existing permanent table, attempts to use the table will cause E_OP0082 errors and an exception (SIGSEGV) in opc_keyqual in the error log.
DMFACP abnormally terminates. The following error sequences may appear in the errlog.log: E_DMA465, E_CL0F0A, E_DM9014, E_DM922F, E_DM9855 E_DMA459, E_DMA45E, E_DMA467, E_DMA464, E_CL0F0A, E_CL0F0A, E_DM9014, E_DM922F, E_DM9843, E_DM9815, E_DM9855. In a dual logging instance, the dual log is disabled due to errors E_DMA458, E_DMA45E, E_DMA467. The checksum noted in the log file buffer is 00000000.
TUXEDO: INGRES/TMS process can fail on attempted rollback if transaction failure is due to (TUXEDO) transaction timeout period being exceeded. Error seen is "E_LQ00D1 - error at the TM-RM's XA interface. transaction demarcation call: invalid XN THRD state Return value is XA_ER_RMERR". Ingres/TMS servers and application servers should be rebuilt when applying patches with this fix.
SELECT from a UNION VIEW with many ANDs and ORs in the WHERE clause returns E_OP0681.
Loading statistics on certain tables fails with E_OP093F for some of the fields. "E_OP093F Actual number of cells retrieved (nn) differs from the number of cells declared (mm)".
Query on tables with high number of indices (both user defined and from constraints) causes a hang in the query optimizer (OPF). Cntrl C will not interrupt the query. IIMONITOR must be used to remove the session.
Creating a distributed database fails when more than one DBMS server is running if the current configuration prevents multiple DBMS servers from accessing the same database (e.g. sole server). The following errors are reported: E_US0014, E_RQ0042, E_RQ0006, E_QE0506, E_SC0206, E_LQ0009, E_DU3511_NOCR_FE_CATS and the STAR server may die. The solution requires II_DBMS_SERVER to be set in the local environment and II_LDB_SERVER to be set to point to the same DBMS server in the Ingres environment.
SELECT from IMA tables ima_qsf_rqp and ima_qsf_dbp incorrectly returns no rows.
There are problems when executing queries which compare float8 to float4.
TRACE POINT DM420 sometimes displays negative values for FIX CALLS and other fields.
PERFORMANCE: Performance degradation since upgrading to Advantage Ingres on a single processor machine running with internal threads and IO slaves on an OS threaded release. Tests show best results obtained when using zero IO slaves.
On 64-bit operating systems, "optimizedb -zc dbname" returns: "Error displaying msg 264476", where otherwise the informational message I_OP091C would be displayed.
Ingres STAR: E_RQ004A, E_RQ0006, E_RQ0011, E_QE0511, followed by an exception (SIGSEGV) in rqu_put_qtxt(), followed by repeating E_QE0002 errors. The Star server becomes inaccessible and Ingres has to be shut down.
E_OP0783_ADI_FICOERCE consistency check on a query using function aggregates and outer joins.
OPTIMIZEDB runs very slowly after a patch containing the fix for problem INGSRV1911/INGSRV1389 is applied.
LOCKSTAT sometimes displays negative numbers in values listed under the section titled Locking System Statistics.
Certain queries which contain a 'not exists' predicate return a wrong result set. Queries are of the form: "SELECT column FROM table1 WHERE NOT EXISTS (SELECT * FROM table2 WHERE condition)".
The DBMS server reports E_DM9049 and an exception (SIGSEGV). Stack shows: dmr_get, dmf_call, qen_tsort, qea_sagg, qen_qp, qen_tsort, qea_fetch, qen_qp, qen_sejoin, qen_tsort, qea_replace, qeq_query, qef_all, scs_sequencer. The DBMS goes on to report E_DM004A and E_QE009C. The DBMS server dies after repeating the last two error messages.
When running with Masterpiece, E_DM93F5_ROW_LOCK_PROTOCOL is output to the errlog.log, when deleting a record.
A query with a WHERE clause containing more than one instance of restriction predicate collections in the form of (a = v1 AND b = v2 AND ...) OR (a = v3 AND b = v4 AND ...) or ... may result in an exception (SIGSEGV) in QEF.
Poor query plan chosen for a query following an outer join query in the same transaction.
The DBMS server returns fewer rows of data when doing a FSM join on a hash structure base table with another table.
COPYDB/UNLOADDB will generate a syntactically incorrect copy.in file when processing a table with a primary key constraint. The keyword KEY is missing from the statement '...ALTER TABLE table_name ADD PRIMARY KEY column_name ...'
In rare instances after a DBMS server crash, the recovery server will fail in its attempt to recover a transaction against a multiple location table. The following errors will be reported by the recovery server: E_DM9C5E_DM2T_ADD_RETRY_CNT, E_DM9C71_DM2T_ADD_TBIO_INFO, E_DM9679_DMVE_TABIO_BUILD, E_DM9662_DMVE_FIX_TABIO followed by various DMVE errors.
OP0082 is generated when a query is run and the tables are optimized, but runs OK with no statistics.
When the last attribute of a table is the key, if another column is dropped from the table (using ALTER TABLE...), then subsequent selects on the table using the key will fail with an E_QE0080 Error trying to position a table.
If the user executes a query in ISQL that includes an ORDER BY and generates a large result set and the user interrupts the query before it completes, the error message E_DM9715_PSORT_FAIL is inserted into the error log several times.
Left join query fails with "E_OP0082 consistency check - unexpected exception occurred". This is caused by an exception (SIGSEGV) in opb_gbfbm(). This is only seen after applying a patch containing the fix for problem INGSRV1273.
Query containing outer join(s) returns an incorrect result. The QEP incorrectly shows that the ordering from an inner relation in a right join can be used further up the QEP. In fact, the right join has returned NULLs for non-matching values which breaks the ordering, resulting in the incorrect result.
The DBMS server crashes with E_DM9397_CP_INCOMPLETE where buffer is for a secondary index. dmd_check information written to crash log file. Occurs in instances where the base table and its secondary index are of differing page sizes.
Persistent index on table with constraints has iiindexes.system_use = 'U'. When table is modified, iiindexes.system_use = 'G'. Consequence of this is that the index is not included in the copy.in script generated by COPYDB/UNLOADDB.
Select of -119988 months return 923 years and 8 months instead of -9999 years.
Deadlock occurs on iirelation when running a "modify to truncate" or an aggregate of a column in iitables.
E_DM002A returned to RAAT on an IIraat_table_open call if INTERNAL threads are enabled.
"SELECT SUM (CASE col1 WHEN 1 THEN col2 ELSE 0 END) FROM tablename" where col2 is a decimal type gives error "E_SC0206 An internal error prevents further processing of this query". The errlog.log shows E_PS0004.
Query containing LIKE predicates with no pattern matching characters OR'ed to equals predicates fails with "E_OP089B Consistency check: PJOIN implied".
In a multiple DBMS instance with INTERNAL thread type set and group buffers, running large update transactions, a DBMS server may appear to hang with 99% CPU utilization.
E_US07FD is generated when updating a table containing an update rule which updates another table's primary key.
Retrieving long data types (BLOBs) core dumps, using "EXEC SQL GET DATA " and data handlers when using "esqlc -prototypes file.sc" to preprocess the embedded source file.
Apache 1.3.22 configured with ICE fails to start apache server with error: "Cannot load .../ingres/ice/bin/apache/oiice.1.a into server".
The MKRAWLOG script does not correctly check raw partitions for availability to take a raw log file.
Error E_CL1F07 is not emitted if a user doesn't have READ access to the specified time zone file. Instead no warning is reported and the connection is made assuming GMT time zone.
Adding a referential constraint using 'base table structure' option fails with 'E_US16D5 Table specified no longer exists'.
"ingstop –force" and "ingstop –immediate" may fail with the DBMS server logging an exception (SIGSEGV) in psq_end_session().
On a B-tree structured table, after updating a row using a secondary index, updating a row using secondary index triggers a procedure, which leads to an infinite loop. The symptom for this problem is the same as problem INGSRV1123.
A query of the form 'SELECT ... FROM ... GROUP BY ...' results in "E_OP0082 consistency check" and the database server either crashes or becomes unresponsive.
The DBA is unable to execute SYSMOD against a database owned by another user.
Running "verifydb -mruninteractive -sdbname dbname -odbms_catalogs" against a database with constraints containing the WITH (INDEX = BASE TABLE STRUCTURE) clause incorrectly gives errors: S_DU1693_MISSING_INDEXTUP and S_DU0320_DROP_FRO M_IIINTEGRITY.
After loading a large ISAM table with pagesize greater than 2k, VERIFYDB falsely reports W_DM5047, W_DM500C or W_DM5050, W_DM500C.
OPF: inaccurate join estimates for a join between a table with statistics and a table without, leading to poor performance. The fix for this problem has been implemented via the new configuration parameter opf_stats_nostats_factor. When opf_stats_nostats_max is ON (to fix problem INGSRV890), this new parameter will scale the result of the opf_stats_nostats_max join estimates. To scale the opf_stats_nostats_max result by value, where value must be greater than 0.0 and less than or equal to 1.0, do the following:
iisetres 'ii.node_name.dbms.config_name.opf_stats_nostats_max' ON
iisetres 'ii.node_name.dbms.config_name.opf_stats_nostats_factor' value
opf_stats_nostats_factor defaults to 1.0 when not set or out-of-range.
"ckpdb +j dbname", followed by a table modify, followed by a series of table inserts (100,000 or so rows), followed by a ROLLFORWARDDB leads to W_DM134D, E_DM93E0, E_DM9650, E_DM964E, E_DM1306, E_DM1301.
A restriction applied on a column from a union view in which the column from one of the underlying selects is nullable and from another is not null can produce the wrong number of rows.
E_OP048B_COST consistency check returned when running a query.
ALTER TABLE DROP COLUMN WITH CASCADE can cause referential constraints to be dropped in which there is no dependency between the constraint and the dropped column.
CSMTp_semaphore failure reported in the DBMS log by the same thread that reports E_DM9238 in the errlog.log. This is a thread that is part of a parallel sort.
Division of two decimal type constants returns incorrect result. For example, "select sum(3.0)/sum(6.0)" returns 0, rather than the expected value of 0.5.
A query runs for an exceptionally long time if the DMF cache is increased above a certain threshold. Query which runs in less than 10 minutes requires in excess of thirty hours, and takes many millions of DIO and CPU units.
"optimizedb -zh –r" may show statistics count 0 for both cell0 and cell1, and gives a bad QEP.
The MIN and MAX functions will return the correct values when applied to VARCHAR fields in all instances.
A statement level RULE with a qualification that invokes a data base procedure that does not have any parameters defined will always fire, even if the qualification does not apply.
"alterdb -delete_oldest_ckp" will delete ALL checkpoint files if the checkpoint being deleted is a directory.
If running IIMONITOR to display formatted session information, the DBMS server hits an exception and if format_sessions() is included in the stack the DBMS server then fails to allow any further connections and must be restarted.
A complex view with numerous left joins and left outer joins produces error "E_AD1012 An attempt to place a null value in a non-nullable datatype".
Queries that use a collation table to make case insensitive queries may experience an "E_OP03A2 consistency check" error if OPTIMIZEDB has been run against the database.
When performing an online checkpoint, a catalog update to iirelation can possibly fail and put logging/journaling into a non-recoverable state.
When running AUDITDB on journals containing CREATE VIEW statements, the LOCATION field displayed can sometimes contain garbage.
Executing a select using a btree index is slower and reads more leaf pages than expected after a large number of keys have been deleted.
UNLOADDB and COPYDB have been fixed to generate views in the correct order, when transferring data from a 6.4 database to a 2.6 instance.
The order for generating permissions when unloading a 6.4 database from a 2.6 instance has been fixed.
Queries with NOT IN sub-selects in which both comparands are expressions can produce very slow query plans. For example, SELECT ... FROM ... WHERE a+b NOT IN (SELECT c+d ....
The "COPY TABLE table_name ( column_name=... ) FROM ..."
command does not assign the correct default value for columns not specified in
the column list. When a column not supporting NULL is omitted from the column
list, the implicit default value for the column type is used (E.g. 0 for INT),
rather than the explicit default defined for that column.
Please note that after applying the fix for this bug, columns defined with NOT
NULL NOT DEFAULT will cause "E_US0ADB No value specified for mandatory column 'column_name'"
to be reported if an attempt is made to copy in data and the column is not
specified in the column list.
In a database procedure with a local variable called result, the statement return result; causes a US0979 error. The same procedure compiled without incident on all earlier releases of Ingres.
E_DM002A followed by E_QE001D may occur if qef_sort_mem is exhausted and session resorts to a DMF sort. QEP will contain an FSM join.
The Ingres DBMS server hangs. Detailed analysis using a debugger shows a number of sessions waiting on an unnamed mutex in one of the following routines: TRdisplay(), TRflush(), TRset_file(), TRrequest(), write_line(). The owner thread is usually suspended on a LOG-IO event and will not contain any of these routines in its call stack. This problem will only be seen on platforms that support OS threads.
"E_US14E4 Invalid options have been specified for a non fast-load COPY" is generated. This error would happen after doing UNLOADDB, and then trying to reload back into a journaled database.
SELECT dbmsinfo('transaction_state') always returns 1. Please note that after applying the fix for this bug, Ingres will not open a transaction for queries which only reference dbmsinfo().
This is a refinement of a previous fix to this problem which would also suppress starting a transaction for singleton selects which only referenced constant values (E.g. "SELECT 1", or "SELECT 1 INTO :hostvar").
IMPXML (child process created by XMLIMPORT) requires excessive memory when processing a large import file and may fail producing a core file.
Optimizer may incorrectly apply restrictions to SELECTs in a UNION query in which the SELECTs are on the same nested, aggregated VIEWs, but with differing WHERE clauses.
In some case OPTIMIZEDB fails with "E_OP08A2 E_OP08A2_EXCEPTION Unexpected exception occurred within query compilation", and an examination of errlog.log shows an exception (SIGSEGV) within opc_ratt.
Binary operations involving decimal operands and operands of some other type (e.g. character) can yield an E_AD1127 error.
Subtracting a negative date interval from an absolute date may give a wrong result. This happens only when the result should be in a following year and the interval is given in months, for example: date('10-dec-2002') - '-1 month'
Row producing procedures (RPPs) can generate a variety of odd results - including error E_QE030C_BYREF_IN_ROWPROC.
A select query involving a not exists subselect returns incorrect results - example returns 992 rows when the result should be 0 rows. The QEP shows that the not exists subselect has been transformed into an outer join.
USERMOD always recreates non-persistent indices in the default location.
A valid outer join query may return 0 rows. This can occur when the query's ON clause has an equi-join predicate on 2 columns and it's WHERE clause has a value restriction on one of the columns (e.g. "a > 0") and an "IS NULL" restriction on the other.
UPGRADEDB: Calls to set directory permissions using "chmod" may fail due to a portion of the constructed location directories path used being truncated at 32 characters. Error messages are benign, since directories typically already have correct permissions.
Executing a script generated by COPYDB would trigger an "E_US14E4 Invalid options have been specified for a non fast-load COPY" error if a HASH structured table had been modified to an ALLOCATION value greater than eighteen (18) pages.
An aggregation query with at least one grouping column of date type produces more rows than expected (the same date grouping column values appear more than once in the output). The data was likely loaded with a version of Ingres previous to 2.6.
The value for server_pid in ima_server_sessions gives the PID of the current DBMS server thread not the server itself.
Queries involving views and executed through an embedded 'C' program may cause DBMS server process to die with an exception (SIGSEGV) and report of an E_PS0004_EXCEPTION error.
Utilities such as UNLOADDB, COPYDB, IMPXML, etc., formerly generated unique copy filenames when multiple users with similar names own tables with the same name in a database by appending a unique integer value to the base portion of the colliding filename(s). For example, if user1 and user2 both owned tables named "atable", filenames atable.use and atable0.use would be generated. The filename algorithm has been changed to now generate filenames of the form tablename.username and will no longer truncate the username suffix, or append digits to the tablename.
A session started with "SET SESSION WITH DESCRIPTION = '<user_sess_id>'" loses the description in IPM when processing a query having a long command string. The description shows as <None> while the query is processed, and returns to '<user_sess_id>' when processing is complete.
Complex queries that executed in Ingres 2.5 may fail with "E_OP0302 too many boolean factors defined in query - query too complex".
There is increased memory consumption by the W4GLDEV process due to memory leaks and frequent crashes using OpenROAD 4.1.
UNLOADDB and COPYDB fails for any quoted string which contains seventy (70) or more characters on a single output line. Failure symptoms may include a very large copy.in file and/or an exception (SIGSEGV).
A query containing a 'NOT IN' subselect by default is transformed into an outer join leading in some cases to poor performance compared to the same query executed with a subselect join. A new configuration parameter opf_old_sub has been provided to allow the override of this default behavior. When opf_old_subsel is ON 'NOT IN' and 'NOT EXIST' subselects will no longer be transformed into outer joins. To initialize this value issue the following command:
iisetres "ii.node_name.dbms.config_name.opf_old_subsel" "ON"
or
iiinitres opf_old_subsel
to initialize setting to OFF, then use CBF to change the value to ON.
Exception (SIGSEGV) may occur in function rdf_unfix() when inserting with a large number of default values.
The DBMS server hangs with sessions in CS_CNDWAIT state waiting for OPF sessions to terminate.
UPGRADEDB: Output contains warnings that some tables, check constraints and/or procedures were not redefined during the upgrade. This problem happens when upgrading a database with front end objects. The UPGRADEDB SQL output file contains the error E_UD0000C.
E_DM9C75_DM2D_CLOSE_TCBBUSY, E_DM0060_TRAN_IN_PROGRESS, E_SC0122_DB_CLOSE may occur if, during rollback, lock escalation occurs and fails for any reason (deadlock, timeout, etc).
Text for message E_DMA495_LG_BAD_SHR_TXN_STATE is missing causing DBMS server to report "E_CL0902_ER_NOT_FOUND No text found for message identifier". "ULE_FORMAT: Couldn't look up message 3a495 (reason: ER error 10902)" occurs in circumstances where this error should be displayed.
Queries against IMA tables cause a memory to be allocated which is not later reclaimed, leading to gradual growth of the DBMS server process.
Using Row Producing Procedures may cause the DBMS server process' virtual memory size to increase because of a memory leak. The amount of memory lost is proportional to the size and number of tuples processed, and may be significant.
A query which joins a union view to another entity may under some circumstances cause the server to crash with an exception (SIGSEGV) in opc_cxest().
A series of connects / disconnects causes a gradual increase in the virtual memory size of the DBMS server (memory leak).
E_DMF003_MUTEX_UNLOCK, E_DM904A_FATAL_EXCEPTION may occur when the buffer manager encounters an I/O error.
Sessions hang in an LKEVENT state awaiting the arrival of free cache buffers.
Pre-compiling an embedded SQL program with a singleton row producing procedure (RPP) call (no BEGIN/END loop) generates a duplicate label of the form IIrtEn. This causes the subsequent host language compile to fail.
The DBMS server may hang, IIMONITOR or IPM will show sessions in a CS_MUTEX state on either a RSB or LLB mutex. Stack dump shows some threads in the scavenge() function, others in LK_request_lock() and/or LK_do_unlock().
E_DMA455_LGREAD_HDR_READ error occurs during rollback of XA transaction.
Write Behind threads are in DI io_sem mutex wait state an inordinate amount of the time.
Error messages US0F1C and US0F80 have been modified to correct SET TRANSACTION syntax.
Exception (SIGSEGV) in aduucmp.c:MakeCE() while performing offline recovery or rollforward of Unicode data.
Suppress WARNINGS during ABF image compilation.
DEL is not recognized as a valid synonym for the IINAMU DELETE command.
An insert or update on a table with a unique secondary index fails with E_CL103B and E_DM9386. Text for error E_CL103B indicates lkb=0.
ADF resolves the function instances to Unicode functions for operands where neither the input nor the output parameters are Unicode types, resulting in incorrect results.
Inserting data values in long nvarchar columns does not work. Error "E_AD2005 ADF routine found DB_DATA_VALUE with an invalid length" is reported.
BTREE index corruption is possible when iitables.keys_per_page > 512. The corruption can occur during any REDO recovery of a BTREE index page, either during ROLLFORWARDDB or during RCP redo recovery processing. NOTE: Application of this fix invalidates previous checkpoints and a new checkpoint should be taken after application of this fix.
A hash join query with enough rows from both join sources to cause overflow processing may generate some result rows, and then terminate with an E_SC0206 error.
E_DMA406_LG_WRITE_BAD_STAT followed by "E_US002B Could not open the iidbdb database" after shutting down one of the servers in a multi-DBMS server installation.
Occasionally an INTEGRITY, RULE, or CONSTRAINT might not be applied, or an obsolete version applied, because of a problem in managing the RDF cache on platforms where the size of a pointer is greater than four bytes.
The IIMONITOR stacks function for Tru64 under INTERNAL threads has been enabled.
An Ingres instance configured for INTERNAL threads, DMCM (Distributed Multi-Cache Management), and multiple DBMS servers would have sessions hang in LKEVENT state.
Certain client jobs become stuck in CS_COMPUTABLE state. Stack traces show the session spinning within gfault_page.
There is poor Ingres performance and CPU utilization when running under Tru64 UNIX on Non-Uniform Memory Access (NUMA) architecture hardware (GS series).
USERMOD does not take advantage of parallel index creation when rebuilding non-persistent indices.
Bad SQL syntax is generated by COPYDB if both the '-parallel' and '-no_persist' flags are specified, and the last index processed for a table having at least one non-persistent index is persistent.
AUDITDB emits errors E_DM1201_ATP_WRITING_AUDIT or E_DM1212_ATP_DISPLAY if the current table compression type does not match the table compression type for a particular journal record for that table.
UNLOADDB, when called internally from UPGRADEDB, will incorrectly generate DROP PROCEDURE statements for system generated procedures.
ROLLFORWARDDB fails with E_DM133E_RFP_ROLLBACK_ERROR but does not mark the database inconsistent.
TUXEDO: INGRES/TMS process can fail on attempted rollback if transaction failure is due to (TUXEDO) transaction timeout period being exceeded. Error seen is "E_LQ00D1 - error at the TM-RM's XA interface....the error code is: transaction demarcation call: invalid XN THRD state return value is XA_ER_RMERR". NOTE: INGRES/TMS servers and application servers should be rebuilt when applying patches with this fix.
Select from an ISAM table registered in a star database causes the error E_OP0A01_USELESS_PLAN.
UPGRADEDB: May report "Error while converting the iidbdb database". errlog.log will contain errors E_DMA00D_TOO_MANY_LOG_LOCKS, E_DM9046_TABLE_NOLOCKS, E_DM004B_LOCK_QUOTA_EXCEEDED, E_DM0034_LOCK_QUOTA_EXCEEDED.
Registering a gateway table to a STAR database using REGISTER AS LINK, causes an E_PS090D error.
Inserting data into a table patched (converted) into HEAP structure through "verifydb -mrun -sdbname db_name -otable table_name" may silently fail. Using a patched table for INSERT is not supported, as the only approved function of this VERIFYDB option is scavenging otherwise unrecoverable data. A table should always either be modified or rebuilt after being patched with VERIFYDB.
A Write-Behind thread may loop forever while flushing groups in a multi-DBMS server shared cache installation.
A Write-Behind thread may receive E_CL2517_CS_SMPR_DEADLOCK when trying to acquire buf_mutex. The Write-Behind thread then terminates and eventually other threads will be blocked on the mutex.
Not all supported options for INGSTART are displayed in the usage message.
UNLOADDB and COPYDB of a table with a COMMENT ON COLUMN gains additional spaces in the quoted comment if the comment wraps around the end of line.
UPGRADEDB: If during upgrade there is a need to rebuild procedures, views, and/or permits, UPGRADEDB is displaying output from an internal invocation of UNLOADDB which includes text advising the user to run unload.ing prior to destroying the database. This message is misleading, as there is no need to run this script, or to destroy and rebuild the database.
XMLIMPORT does not work when an XML file being imported has data from multiple tables in it. The error "E_US0845 Table utl40_table does not exist or is not owned by you" is given out. This is happening because we now open the data handle for the table data files while we are reading the metadata, but as we read successive files, we are overwriting the file handle of the previous one.
When II_DECIMAL is set to ',' (decimal comma) the statement: 'CREATE PROCEDURE procname AS DECLARE avar=DECIMAL(10,3); BEGIN avar=2,34; END' will produce error E_US09C4 and cannot be loaded.
Errors 'E_US0845' and 'E_US0834' are seen when unloading a distributed database.
A printing of decimals(31,10) is handled incorrectly in reports, and as a result asterisks are printed out instead of a decimal.
Transaction abort while SET NOLOGGING is in effect fails to mark database inconsistent.
The user is unable to remove the value of a parameter which has a null default value in CBF.
E_DMA434_LGK_VERSION_MISMATCH may occur during recovery process (RCP) startup if a concurrent non-server process (e.g., LOCKSTAT ) is simultaneously connecting to and creating the logging/locking shared memory segment.
Execution of a COPY <table>() FROM '<file>' when <table> has one or more NULL-able long data type columns (BLOBS) may fail with errors E_AD7004_BAD_BLOB, E_SC031A_LO_PARAM_ERROR and E_US0005. The problem is sporadic and only occurs on 64-bit platforms requiring pointer alignment on eight byte address boundaries.
'ALTER TABLE tablename ADD columnname columntype WITH DEFAULT NOT NULL UNIQUE' causes 'E_SC0206 internal error' and an exception (SIGSEGV) in cui_idunorm() on a table where a column had previously been dropped.
A hash join of many rows all with the same join column values can result in a QE0018 error. It can also produce the wrong answer when it is an outer join (right or full).
Recovery process (RCP) may suffer an exception (SIGSEGV or SIGBUS) during startup while handling a "disable dual logging" event.
Under certain circumstances E_DMA464_LGREAD_BADFORMAT or other LGREAD errors can occur after creating very large log records (as when CKPDB performs an online checkpoint, produces BI dump records). For this to occur, a large log record spanning multiple log pages must have been forced while its transaction was suspended waiting for sufficient contiguous log pages to write the record.
Sporadic INGSTART failures in which IIGCN, or the archiver (DMFACP) is reported to have failed to start. On further examination, process is seen to have started successfully, and failure report is incorrect. This timing issue has only been seen with Tru64, and the frequency of occurrence is dependent on unknown environmental factors.
On Linux IPCCLEAN does not catch both versions of 'resource deleted' from IPCRM.
SQL commands contained with environment variables ING_SET, ING_SYSTEM_SET, or ING_SET_DBNAME will be executed using applications privileges when that application is run. Behavior has been changed to ignore DROP, CREATE, DELETE, UPDATE, INSERT, MODIFY, or SELECT commands and to issue a warning.
Hash joins are well known to be an alternative to sort merge join, and have been implemented in 2.6, but since Service Pack 1 (SP1) have been disabled by default. A hash join algorithm should perform at least as well as an equivalent sort merge or partial sort merge join, and usually much better, and the change associated with this SIR enables Hash Joins by default. However, in certain cases, the performance of certain queries has been seen to suffer. If this is the case, the global default for this feature may be set to off by adding "ii.machinename.dbms.*.opf_hash_join: off" to $II_SYSTEM/ingres/files/config.dat. Control of this at the session level may be managed with TRACE POINT OP162. Setting this trace point sets the activation state for this feature opposite to the value set in config.dat, clearing the trace point restores the feature activation state to the global default.
The DBMS server crashes when inserting 46 characters into a LONG BYTE column when the page_size is 8K and the stack_size is the default (65536).
The transform of "not exists" to outer joins is too restrictive, even when a simple restriction predicate is ANDed to a correlation predicate. The transform is not applied in other cases when it should. Specifically, if the where clause contains other restriction predicates, the flattening potential is sometimes missed.
Poor memory management in the parameter manager facility may cause a gradual increase in the virtual memory size of the DBMS server process.
The Advantage Ingres Service does not register for the shutdown event. It also does not shutdown Ingres when there are active sessions spawned from another service.
New releases of Ingres that introduced additional reserved words inevitably result in syntax errors in applications written long before that use one of the reserved words as an entity name (table, column, etc.). The request is to make Ingres parsers treat reserved words in a context sensitive manner so they only cause errors when used in an ambiguous context.
Right outer hash joins may produce extraneous result rows.
If a hash join is performed in which the join columns are VARCHAR or TEXT type and one or more of the columns has trailing blanks though a matching value does not (or the number of trailing blanks is different), pairs of matching rows may not be found. Change to trim only for VARCHAR (and NVARCHAR). Trailing blanks are significant for TEXT.
A hash outer join generating enough rows in the outer source of the join to overflow onto disk can omit some of the non-joined rows from the outer join.
The SET [NO]OPTIMIZEONLY statement currently requires the executing user to have tracepoint authorization. This is not consistent with other OPF set statements (e.g. SET [NO]QEP) that do NOT require tracepoint authorization.
A view was created that selects a case expression from the result of an outer join. When a query is issued on the view with a where clause restriction on the case expression, the wrong number of rows may be returned. The same problem may occur without views, when a case expression is used in the where clause of an outer join query, and a column referenced in the case expression comes from the inner side of the outer join.
Repeat queries run in a tight loop in a highly concurrent environment with other instances of the query may very occasionally return incorrect results.
A hash join between 2 sources, one of which has a row size in excess of 8K can cause an exception (SIGSEGV).
Some complex queries may get OP0889 errors. The queries observed so far with this problem are big enough to be compiled with the new enumeration heuristic and contain 2 or more subselects OR'ed together in the where clause.
A join between a TEXT column and a VARCHAR column returns fewer rows than expected. The problem could occur joining a TEXT column to other data types, too (including another TEXT column), and could also occur in an aggregation query in which one of the grouping columns is TEXT.
Incorrect results may be produced by complex queries with numerous nested joins in which a FSM join is placed above a hash join.
A hash join or aggregation using columns of C data type may produce incorrect results. Hash join/aggregates on Unicode columns or CHAR/VARCHAR columns with alternate collation may also produce incorrect results.
An outer join query in which the inner table of the outer join is a view that itself contains an outer join will generate an E_OP0395 error (unexpected outer join ID found).
An aggregate query in which a grouping column also appears in an equijoin predicate in which the other column is not exactly the same type (the columns were i2 and i4 in the problem script) may produce apparent garbage for the grouping column.
A select in a for-loop in a procedure in which an order by clause contains the same column more than once, and that column does NOT appear in the select-list, will cause an exception (SIGSEGV) in QEF. A SEGV may also happen when an "order by" in a database procedure select statement references a column not in the select list (or redefined by "AS" clause).
XA global transaction may be rolled back if first updating statement of any participating transaction aborts, possibly resulting in lost updates.
Error message E_US1594 incorrectly reports table name and composite key length when one of the columns is greater than the maximum btree key length.
The DBMS error handling for the ENABLE SECURITY_AUDIT statement has been fixed.
There are various problems with parallel sort synchronization and sort work file usage, leading to SRopen() errors, un-initialized semaphore errors, exceptions (SIGSEGV), and/or stuck parent/child threads.
The E_DMA464_LGREAD_BADFORMAT error is reported by the Archiver.
When running Ingres with Tuxedo, the following errors may be seen in the tuxedo log: "ERROR: Service xa_start returned XAER_RMERR, ERROR: Service xa_start returned XAER_PROTO". These errors will no longer occur. NOTE that the number of transaction entries allocated per application server in the Ingres-Tuxedo shared memory segment used to be II_TUX_XN_MAX/II_XA_SESSION_CACHE_LIMIT but it is now II_TUX_XN_MAX/II_TUX_AS_MAX which allows tuxedo users to lower II_XA_SESSION_CACHE_LIMIT and increase the number of transaction entries allocated per application server.
An exception could occur in LG_abort_transaction() trying to abort an XA transaction after a log full condition.
AUDITDB reports E_DM1206_ATP_TOO_MANY_TX and E_DM1205_ATP_NO_BT_FOUND errors when the -b and/or -e options (timestamp constraints) are used.
Tuxedo server may crash or get the following error during shutdown: "E_CX0003 Unable to free memory: cl_status = 70155 for control block = cx_cb_p". The application and TMS servers need to be re-linked after applying this patch.
Updating a VARCHAR column with a greater than 2K string in a database created with a custom collation sequence causes the DBMS server to crash silently with no error reported in the errlog.log.
Using Tuxedo for distributed transaction management, the following errors are reported in errlog.log: E_DM904A_FATAL_EXCEPTION, preceded by other errors indicating failure to PREPARE/SECURE or failure to ROLLBACK a distributed transaction. The errors can occur if a Tuxedo TMS server sends a rollback for a transaction while another Tuxedo TMS server is preparing the same transaction. Other errors might include: E_DMA406_LG_WRITE_BAD_STAT, E_DM9370_DM0L_SECURE, E_DM9505_DMXE_WILLING_COMMIT, E_DM9503_DMXE_ABORT.
E_DM9049_UNKNOWN_EXCEPTION is generated in the DBMS server (and then the RCP server) In the errlog.log you will also see the following stack dump: aduucmp, adt_kkcmp, dmv_compare_key, dmve_bid_check, dmve_btput The errors happen during REDO or UNDO (rollback) of an update to a btree table with NCHAR/NVARCHAR data, where the NCHAR/NVARCHAR column is part of the btree key.
The 'Move Log to Journals' button in IJA UNIX fails with "E_US1A22 Database name starting with ' ' contains invalid characters".
Upgrading a 32-bit database to Ingres 2.6 64-bit using UPGRADEDB succeeds, but then an insert/update of a table upgraded fails with E_RD0132_QUERY_TREE, E_PS0D15_RDF_GETINFO. The problem happens if the table has a check constraint, e.g. "CREATE TABLE T (col1 INT NOT NULL, CONSTRAINT ccc CHECK (col1 > 0))". During UPGRADEDB the check constraint which has a stored query tree should have been dropped and recreated, but it wasn't.
IIACCT ACCOUNTING: On a client(Windows)/server(HP-UX) instance, the accounting reports produced by iiacct + convacct + acctcom -h show all the activity belonging to the user "ingres" even when this activity belongs to other users.
The archiver ACPEXIT script in $II_SYSTEM/ingres/files is not called if you kill the DMFACP process or if you execute "ingstop –dmfacp".
When performing an "insert as select" against a STAR database, where the source table is in a remote database being accessed through a VNODE, the GCC server will hang if the OS IPC buffer subsystem becomes full.
The INGSTART utility now prints out a GCC's port information when starting up a GCC. It also waits for a GCC to accept connections before continuing on.
"copydb -with_proc -add_drop ..." may add more than one drop procedure statement for each procedure.
USERMOD fails when executed with II_TEMPORARY set to an embedded spaced location with "E_US1A22 Database name starting with 'files\' contains invalid characters".
MKRAWLOG fails with message "Symbol II_LOG_FILE_NAME must be defined" on hosts with fully qualified hostnames (i.e., machinename.domain).
A database could become inconsistent through a COPY IN of variable length data (i.e., varchar) which fails (i.e., due to a bad data file).
VDBA fails to properly create HTTP type ICE locations when using 64-bit Ingres as the server.
SYSCHECK may fail on a release of an operating system if it is newer than the one Ingres was compiled on.
The Replicator service internally gets an exception (SIGSEGV) with a stack overflow. This causes the Replicator server to not shut down properly unless a reboot is performed. This happens when a two-phase commit is in progress and it fails.
Queries containing decimal to integer comparisons fail withE_OP0791_ADE_INSTRGEN and E_AD200B, e.g.:
create table foo (deccol decimal(9,0), intcol integer);
select * from foo where deccol = 1;
select * from foo where deccol = intcol
The CACHEUTIL banner text is now displayed properly.
The LIST output from CACHEUTIL now does not wrap around by one character.
Performance enhancements have been made to OS threading in Ingres on HP-UX.
CS_su4_dump_stack() does not print the PC addresses at which functions are called if the function name is found.
The cache_lock help description in CBF under DBMS is incorrect. It should read: "The cache_lock status determines whether or not the cache pages will be locked into physical memory. This will help to avoid swap file paging. Attempts to lock more cache into memory than a system-specific limit will fail. Because of the impact on system resources, the use of cache locking is restricted to the super-user."
SYSCHECK incorrectly reports the number and size of shared memory segments required when dmf_separate is on.
Optimization of complex query takes a very long time when the system is running under load (Linux OS threading).
Seemingly simple query ignores obvious secondary index, choosing to do much more expensive table scan.
The user environment variable II_CKTMPL_FILE can be use to set the location of the checkpoint template file.
The 64-bit versions of Advantage Ingres did not allow allocation of more that 2GB of memory for the DMF buffer manager.
Sporadic occurrences of exceptions (SIGSEGV or SIGBUS) are seen within the OPF facility. This is seen on busy systems with queries that perform joins on columns with histograms.
While running OPTIMIZEDB, the client received an error "iiCLfdreg FATAL: fd 8535 surpasses max fd 8192" and the DBMS server then hung.
The USERMOD utility does not display the usage message when the input string is just a '-' character.
Errors in STARVIEW are generated when a failure occurred during a select of the SQL and TESTNODE.
In addition to the problems documented above, this service pack also contains fixes to problems that were resolved in Service Pack 1. For more information on these problems which were resolved in Service Pack 1, visit SP1 Generic readme.
For technical assistance with this product, contact Computer Associates Technical Support at http://ca.com/ for a complete list of locations and telephone numbers. Technical support is available 24 hours a day, 7 days a week.