Sybase数据库技术，数据库恢复专家

not be used in a SELECT which is not part of the declaration of a cursor or which is not inside

a stored procedure"

on the same column can choose deferred update mode, when direct update mode is possible.

As a result performance of the UPDATE query can be negatively impacted.

An example query is:

update t1 set c2 = 3 from t1, t2 where t1.c1 in (1,2,3,4,5,6,7,8,9) and t1.c2 = t2.d2

where table t1 is being updated and the query joins tables t1 and t2 and there is a IN-list

on column c1 of table t1.

modified pages to disk at the end of recovering a database that is only modified on

the surviving node.

"quiesce hold" command may result in error 2244 "Unable to open database dbname".

Error 2244 may also be raised when issuing a "quiesce" command on a database that is already in a 'quiesce' state.

5861 but hits 10886.

This might cause some issues later when the process resumes on the new node.

modules 'cm__check_page_range' or 'cm_choose_buffer_pool'.

2. In "Specify Master Device page" of Create Cluster Wizard, there will be a dialog if specified devices do not have I/O fencing capability.

3. In "Specify the System Procedures Device" of Create Cluster Wizard, there will be a dialog if specified devices do not have I/O fencing capability.

4. In "Specify the System Database Device" of Create Cluster Wizard, there will be a dialog if specified devices do not have I/O fencing capability.

5. In "Specify Server Instances" of Create Cluster Wizard and Add Instance Wizard, there will be a dialog if more than one instance will be created on one host.

6. In "Specify Name and Path" page of Database Device Wizard, there will be a dialog to show if the specified device is I/O fencible.

7. In "General page" of Database Device Properties, there will be a label to show if the device is I/O fencible.

8. In "Device Information" of Add database wizard, there will be a dialog to show if the specified device is I/O fencible.

trailing blanks when the column is involved in an equjoin with some character

data type expression. Also hash/merge join was not available between character data type

column equijoin under nocase sort orders.

may be reported in the error log together with a stack trace which includes

the modules 'dol_putdrow', 'undo_dol_delete', 's_renormalize' and 's_recrtproc'

while concurrent sessions write a new query tree in sysprocedures whereas

the current session needs to rollback its current work from it.

distinct_hashing or group_hashing operator may claim

too many buffers (typically the value of "max buffers per

lava op") as seen through the "set statistics plancost"

output.

in the error log together with a stack trace which includes the modules

'clfr_recovernode' and 'clfr_recovery' during node failover recovery.

This change reduces the number of inserts by skipping faults for which

an exclusion is defined with sp_dbcc_exclusions and skipping inserts in dbcc_counters.

The pre-fix behavior can be enabled using exec sp_dbcc_updateconfig null,

"enable excluded faults inserts", '1' and exec sp_dbcc_updateconfig null,

"enable dbcc_counter inserts", '1'

Currently it can return 'Critical internal application error' along with 'The specified User does not exist'

Sybase Adaptive Server Enterprise DDL Generator Utility/15.0.2/1/S/1.4.2/ase15x/Thu Dec 13 19:23:59 PST 2007

error log together with a stack trace which includes the modules

'ptn__setup_field_access', 'wrongpage' and 'sort_read' while ASE was in

the process of reporting error 696 "An attempt was made to fetch logical page

... Page belongs to <value> and not to <value> ..." due to a sort operation

having failed.

for 'data-only' row with minimum row length of <value>." may be reported in the

error log followed by a stack trace in module 'collocate' when the DBCC PAGE

command is run with the print option value of 4 on a datarows-only-locked table

that has variable length columns with a size value greater than 256. The modules

'prvarlen_cols' and 'prDOL_ROWS' will also be reported in the stack trace.

named cache to be deleted at some instance with out releasing physical memory.

ALTER TABLE ADD PARTITION command is run while concurrent activity on the target table

takes place.

circumstances when executing a query using CASE

expressions involving inside comparisons to

NULL literal values and ANSINULL session option

is not active.

Customer would like to be able to have some databases be local, and some proxy.

a product server down by mistake.

in module cipcmsg_alloc() due to lack of free messages to be sent over network.

in the presence of global configuration with size less than 255 pages can lead to successful

operation instead of failing.

value on the same option failed due to the code error in sp_configure

stored procedure.

Backup Server: 4.145.2.22: [107] Error for device '<archive_device>'. Attempted to transfer xxxxxx bytes, but yyyyyy bytes were transferred by operating system call write/read."

creation of a temporary table in the local temporary database of an

instance.

http://msdn.microsoft.com/en-us/library/ms187054.aspx

And looking at how it works, it is nothing but what we already had in

mind for the incremental dumps. The snapshot would be an archive

database that would use the original database and the modified pages

section instead of a dump + modified pages section.

Creating a snapshot will also be very similar to creating an archive database,

but building the original diskmap would be much easier. Then,

a couple of new fields in the dbtable would tell ase that it has to

forward a write defined by (xactid, page) to a set of snapshots, while a daemon

would be writting the original pages as the transactions commits,

if the pages had not been copied already.

I see the project pretty easy to implement, and we would kill two birds

with one shot:

- Database snapshots

- Incremental dumps. These could require having internally two types of snapshots.

A real one, with access to the original page, and a light one, with access

only to the list of modified pages but not the pages themselves.

page count might be set incorrectly to a wrong value and the processes suspended on LCT might continue.

associated with a particular procedure from the procedure cache.

dbcc deleteplan(db_id/db_name, obj_id/obj_name, plan_number)

Delete the procedure plan specified by plan_number

dbcc deleteplan(db_id/db_name, obj_id/obj_name, 'all')

Delete all plans associated with this procedure object.

was aborted because it used too much stack space."

together with a stack trace which includes the modules 'memalloc' or 'mempagecount'

may be reported in the error log when running a complex query.

in the management nodes section. This applies to ASE and qrmutil.

it is possible that during parallel recovery ASE tries to online an archive

database before the scratch database this archive database is bound had a

chance to come online, causing the recovery of the archive database to fail.

algorithm, traceflag 15385

set minmax off

corrupted during upgrade during online database after the database is loaded across platform where

byte-order is different, if the table has more than 255 columns.

in a MASS and BCM thread that is trying to downgrade the physical locks on the MASS.

The user can not input any starting port larger than this maximum value.

enhanced to include the database name and the object ID of the system table whose table descriptor

could not be installed. Similarly error message 952 "Database '%.*s' cannot be opened because a

system index descriptor cannot be installed." is enhanced to include the object ID of the system

table whose index descriptor could not be installed.

ASE may raise a message "usinitpktsz: Unable to allocate overflow buffer of size xxxxx from pool".

would often have their result rows stored in a worktable when the cursor OPEN command is executed.

Subsequent cursor FETCH commands would read rows from the worktable, thus insulating the cursor from

changes to the tables upon which the cursor was defined, in between FETCH commands. In ASE 15, the

optimizer attempts to avoid creating worktables for better performance. In this CR, command line

trace flag 457 has been introduced to force ASE 15 to internally convert many of the cursors that

would have used a worktable in ASE 12.5, but would not use on in ASE 15 to insensitive cursors to

force materialization of the cursor rows when the cursor OPEN command is executed.

then the message "current process infected with 11" in module 'cm_bind_cache' may be reported

in the errorlog.

error 14 (Found invalid magic numbers and / or table of contents entries on the quorum device. This

device may be corrupted.). This is most likely to occur in simulated cluster environments with an NFS

mounted quorum device and a low configuration for "quorum heartbeat interval".

"An invalid PARAMFMT stream was received." when a query is sent to ASE.

contains a NOT EXISTS subquery that references the target table (tab1) may insert extra rows

into tab1 in ASE 15 compared to pre-ASE 15 versions. The ASE 15 behavior is the expected

behavior, but to maintain backwards compatability, command line trace flag 9580 has been

introduced to force pre-ASE 15 behavior.

Workaround is to:

Login with username/password to kill DBMS processes on remote instances.

OR

Login to the instance of the cluster where the DBMS process is local.

OR

Define external login mappings (using sp_addexternlogin) for Kerberos logins for each instance.

the correct syntax is set proxy...restrict role

http://infocenter.sybase.com/help/index.jsp?topic=/com.sybase.help.ase_15.0.sag1/html/sag1/sag1814.htm

The message header indicates the name of a stored procedure called by the trigger and not

the trigger's name.

when SP_DROPUSER is run although the user to drop is not actually dropped.

infected with 11" in the module 'cm_delete_cache' may be reported in the error log.

'process infected with signal 11" with stack trace in the module of jcnstGetObject()

and LeEmitSndOp::_LeEoRouteResults().

execution plan if the groupby itself also has its own correlated subquery.

The better plan would be evaluating the groupby only once and storing

the intermediate result in a worktable.

error 1289. Stack traces reported might contain the word "des__revalidate".

logical page <pageno> was read while accessing...", or a 891 error, "rying to unhash a buffer

which is in Dirty state.." may be reported in the errorlog after database binding operation fails.

like "Spinlock Pdes Chain Spinlocks at address owned".

supports the correction of '100039' faults reported by DBCC CHECKSTORAGE for user tables; this

fault describes an incorrect extent OAM page hint.

where the tempdb is bound to a named cache.

is created and dropped from different instances in the cluster.

could happen during execution.

"MERGE JOIN" operator, then a SEGV may be generated while copying

data to the output tuple. The stack will have the following

calls on it:

LeMJOp::getNextInner()->LeMrgUnionOp::_LeOpNext()->LeMrgUniContext::getNextTup()

LeUnionContext::copyRsVal()->LeVtuple::copyCols()->LeColumn::LeColCopy()->memmove()

The stack may appear different with different FBO compilers.

waiting for appropriate time.

failover-recovery of instances, each time holding up one or more internal PCM-tasks. This can be noticed in sp_who output, with many PCM-tasks sleeping for no reason (i.e without any user issued command).

mixed log and data databases, a segmentation violation may be

reported in the ASE errorlog showing modules 'th_failover' and

'th_load_thresholds_into_memory'

export LIBPATH $SYBASE_JRE'/bin:'$SYBASE_JRE'/bin/classic'

To

LIBPATH=$SYBASE_JRE'/bin:'$SYBASE_JRE'/bin/classic'

export LIBPATH

failover recovery completed, a non-syslogs page could be used as a log page in a MLD database.

By default, the first reference to the table being deleted in the FROM list, whether it has a

correlation name or not, is used as the source of rows to be deleted.

Under command line trace flag 120, if there is a reference to the table being deleted without a correlation name, it will be used as the source of rows to be deleted, regardless of where it appears in the FROM list.

If no table without a correlation name is found, the first reference to the deleted table will be used as the source of rows to delete.

ETL hangs when saving a job with 20 projects.

Resolution:

The internal data structure used when working with job data was adjusted to grow with larger data sets.

in a mixed log and data database, might be blocked from completing the rollback.

'dbname' (dbid), 'objname' (objid), ..." may be reported in the error log mistakenly when

a session is killed or otherwise interrupted.

1. trace flag 7725 is on

2. abstract plan load is on

3. enable literal autoparam is on

if failed to associate APs with the parameterized text, text will be parameterized again

using 15.0.1-style autoparam name, and retry AP apply with the new text.

DBCC CHECKALLOC. In rare cases, a 3475 error may be reported

instead

not quiescent when the dump was performed.' could be returned when loading a 64 bit database dump to a platform where the byte order (endian type) is different; the dump must have been taken when there is heavy update activity on the source database.

in the module 'ind__ides_getnext' may be reported in the error log

with a stack trace which includes the modules 's_compile', 'substitute'

and 'ind_rowcopy' when a SQL command is in the process of being compiled.

This will result in ASE shutting down itself.

correctly rows from the INSERTED/DELETED tables.

may hang when Adaptive Server is configured with several

engines and some other queries (which ran in parallel too)

were aborted (for example sending CTRL+C).

propagated back to the ISQL client.

on the physical lock upgrade taking place, and this can lead to an indefinte

hang, stalling the progress.

2.Exclude "Engine Group" from cluster creation menu.

3.Add "Local Cache", "Engine Group" and "Local Temporary Database" submenuItems for instance creation menu.

interpolation assertion (ratio >= 0.0 && ratio <= 1.0) on diagservers.

stack traces in the errorlog.

with more than 2 table join inner to the outerjoin.

on the companion and a connection that is failing over have the same user login and

the connection that is failing over gets locks on syslogins catalog before the HA

takeover process does. The failover or failback does not complete and HA state on

the companion remains unchanged.

cluster environment.

been previously run against an ASE configured to use a non-English language.

setup may encounter SIGSEGV in libsylapij.so and generates a java stack trace, and then core dumps or hangs up.

You can work around this issue by moving the OEM license file from $SYBASE/SYSAM-2_0/license directory

before installation and moving it back to the directory after installation.

unexpected values. Sometimes stack traces may happen with reference

to add__computedcol module when using the table with such computed columns.

ETL hangs when using DB Bulk Load Sybase IQ when an invalid pipe load script is executed.

Resolution:

ETL was modified to properly handle this error condition.

encloded in brackets. For example "tcp:[2001:ec8:4008:1::123]:80"

Current thread: AWT-EventQueue-0

java.lang.NullPointerException

at com.sybase.aseplugin.bo.ASEBasePage2.getJDialog(ASEBasePage2.java:73)

at com.sybase.aseplugin.bo.prop.RoleProperties$RoleColumnPermissionPage.initTableComboBox(RoleProperties.java:2432)

at com.sybase.aseplugin.bo.prop.RoleProperties$RoleColumnPermissionPage.initComboBox(RoleProperties.java:2403)

at com.sybase.aseplugin.bo.prop.RoleProperties$RoleColumnPermissionPage.<init>(RoleProperties.java:2336)

while the statement cache is active.

updated on sysmessages resulting in sub-optimal plans.

remote server is configured with 'server login'.

are based on some magic numbers when statistics are not explicitly updated on the tables.

An option is provided to implicitly generate statistics for private temp tables. The statistics

are kept in memory until the temp table is dropped. Traceflag 2716 is used to turn on this option.

Note that external tools such as optdiag and sp_modifystats will not affect the in-memory

statistics. The feature does not support regular, partitioned and proxy tables, as well as

optdiag output, compatibility mode and DDLs in stored procedures when deferred compilation is

turned off.

dbid if the ltm marker is not turned off prior to the dump and ONLINE is attempted prior to running

dbcc dbrepair(dbname, ltmignore).

( please note this is different from the existing "cis connect timeout" parameter which applies only to the time taken to make the connection).

log and data mixed, is extended on to another device using the ALTER DATABASE command

without using the LOG ON clause, the newly added disk piece gets marked as

"data only" instead of "data and log". That is, no log space is added by the command.

Due to this, the database may run out of log space sooner than expected.

signal 11 stacktrace might happen in CgpUnion::_cgpInit() during compilation.

The stacktrace is more likely to happen if "append_union_all" is also turned off.

it could cause:

CT-LIBRARY error:

ct_results(): protocol specific layer: external error: There is a tds

protocol error. Premature end of the datastream was encountered.

e.g.

select 'Mary' for browse

* Release of the basic QPTune functionality

* Add missing stats support

* Save queries to user specified groups before any tuning exercise

* Change setting default optgoal during fix to avoid any conflict

* Other bug fixes and cosmetic changes to improve the existing utility overall

the "enable pci" configuration parameter to 0 and the "enable java"

parameter to 1. The message "ASE - Dynamic Pluggable Component

Interface is disabled" will be reported in the error log. This

can be done only on Adaptive Server 15.0.3 ESD#1 or later.

in the deferred mode.

"Invalid security token" when a query is sent to ASE confiured with multiple engines.

'bufdlink' may be displayed in errorlog while doing 'online database'.

compatibility mode. The default value is 0 (disabled).

modifying a partition, Adaptive Server Enterprise verifies that the user process has exclusive

access to the partition descriptor. If it does not the command fails and a 14127 error is raised.

text_tabcheck when running dbcc checktable after altering table change lock, which also follows

runtime undoing a dml transaction on the same table in a remote node.

system tables, with other tasks executing DDLs and utility command on the same database.

and one of the DISTINCT view SELECT column has an equisarg on it.

references:

The name server uses the selected load distribution policy and generates

an IOR with multiple profiles to balance the request between available servers and ports. If a dynamic load policy is selected, the NLL is used to determine

and balance the load of the individual servers.

There is no longer a way to select the 'load policy' for a cluster.

could be incorrect.

if you try a higher value than what is set with sp_configure "max resource granularity".

the installer should determine the installed components and update them to the latest available versions.

But the installer may additionally install the following connectivity components even if those are not installed components.

o Embedded SQL/C

o Monitor Client Library

o XA Interface Library for ASE Distributed Transaction Manager

o Connectivity Language Modules for chinese, japanese, spanish, french, korean, polish, thai, german and portuguese.

1> union

2) subquery on one side of the union,

3) derived table expression under subquery,

4) there are local variables in derived table expression,

5) there are builtin functions in derived table expression,

may cause stack trace to happen in LeEvRun().

a SARG against PT on a non-partitioning column may return too

few rows. The query must also contain a SARG against PT's

partitioning columns such that some partitions can be eliminated

during optimization. Take the following query as an example:

select * from S inner join PT on S.c1 = PT.npc and PT.pc >= 1

The predicate "PT.pc >= 1" can be used to eliminate some

partitions during code generation (pc is PT's partitioning column).

The equijoin "S.c1 = PT.npc" is a SARG against one of PT's

non-partitioning columns. In this case, once S.c1 has a value of

NULL, no more rows will be returned from this equijoin even after

S.c1 is refreshed to non-NULL value.

Note that this problem could also occur across stored procedure

or cached statement invocations.

s_getTreeOrPlan,open_user_table,open__byname,parse_name.

This may happen if procedure creates a table with computed column.

with the 'for load' clause on a node N1 does not return the

expected 930 error message.

in the statement cache.

view or derived table will not mark the joining predicate

as a sargable. This could result in a cartesian product scan

rather than an index lookup.

the same engine. This results in OMNI connections not distributed across the engines that can

potentially cause performance degradation.

the dataserver will continue to start with no valid listeners.

and col1 and col2 columns on table A are character fixed-width, non-nullable datatypes.

select B.FKcol1 + ',' + B.FKcol2 from A, B

where B.FKcol1 + ',' + B.FKcol2 = 'TEST5,T'

and A.col1 = B.FKcol1 and A.col2 = B.FKcol2

column groupby query (i.e. SELECT more columns than the Group By list and aggregate)

when one of the joining view/derived table has DISTINCT.

the same engine. This results in OMNI connections not distributed across the engines that can

potentially cause performance degradation.

that bind/unbind cache operation may fail with out error message.

'ssql_update_recompile_info' together with a stack trace which

includes the modules 's_getTreeOrPlan' and 'tdsrecv__dynexec'

may be reported in the error log and could bring the server

down if the configuration options "STATEMENT CACHE SIZE",

"ENABLE MONITORING" and "ENABLE STMT CACHE MONITORING" are

all enabled and a client application executed a dynamic sql

statement with the option "DYNAMIC PREPARE" active and the

internal lightweight procedure for the statement had to

be recompiled (for example due to a schema change).

are killed externally,the command "start cluster" fails to start the cluster.

example assiging the value 18446744073709551615 to a BigInteger field in Java will result in a

value of -1 being stored in the Java field.

Clearly there is no current way of maintaining statistics on tables which are only populated potentially during a transaction's lifetime. There needs to be some sort of statistical learning mode to handle this.

for a nested loop join even though there is no sarg on the store index child scan and

there already exists an index whose keys match the equijoin columns. For most time, based

on actual data distributtion, the existing index scan plan would be better.

dbid <value>, index <value>, oam page <value>, would cause a negative page count ...

Contact your System Admin." in the error log mistakenly when multiple sessions import

data in the same table concurrently through the bulk copy utility.

configuration parameter has no effect.

It always initializes to the default value

of 45.

when configured with Java in Database feature if JVM requests an extremely large

amount of memory due to Solaris bug 6671882.

message "infected with 11" with the modules "jvm_GetClassName" and "DBClassFileGetClassName", being

printed in the ASE errorlog.

detect spurious errors because it could skip a dirty cached page and

use the page on disk instead, if the owner instance has got a request to downgrade

the physical lock to a null lock.

parallel scan, the value of the UsedCount column of monOpenObjectActivity for IndexID = 1

is not incremented.

clause is executed, the datatype of the grouping columns

is VARCHAR and the data for those columns contains NULL values

and empty strings having only white spaces like ' '

Normally sort nodes would provide benefits such

as orderings for merge join/ORDER BY/GROUP BY,

or buffer manager optimization (i.e. better cache hits on an ordered join attribute).

This error could occur when there is a combination of multiple covered and non-covered

index scans within a query plan. Normally, it occurs in cases when the sorts are

very small, so the error is not noticible, since it only involves the estimate

of the CPU component. If the sort volume was large, then it would have overwhelmed

the incorrect CPU estimate, and the sort would have been eliminated.

However, if the input volume estimate for the

sort is incorrect, due to missing statistics, and this volume is actually very large,

then the addition of the sort operator could result in a noticible performance regression.

the database. This also makes AP capture more versatile.

It is documented that the connection level AP dump/load is disabled

when a USE <database> statement is issued, to avoid the tentative usage

of an absent user defined AP group.

However, when the standard in/out default AP groups are used, there is

no reason to disable the connection level dump/load - since these groups

are present in all database.

added a sort to break up the join order, so that there is more buffer manager cache available

for each side of the sort, that in turn would reduce PIO requirements of the query overall.

A new syntax is being used to add an ipv6 address in sp_addserver. The syntax

will encapsulate the ipv6 address with in brackets '[,]'. An example of

an sp_addserver command for an ipv6 address:

sp_addserver

'rmtsrv1',ASEnterprise,'[fd77:55d:59d9:167:203:baff:fecd:2b0d]:6268'

with a top DISTINCT operator might be ignored by optimizer.

if reformatting strategy is chosen.

stack trace in the module of 'LeEvals::LeEvRun()' and 'LeOrScanContext::setOrVtuple'

if the query plan chosen has OrScan in it.

may be reported in the error log together with a stack trace which includes

the modules 'vu_getviewdefn', 's_reinstallProcTree' and 'proc__linktopss'

when two concurrent sessions access a VIEW whose query tree needs to be

redefined by ASE.

any non-unique index or table scan. In 15.0 the unique index is only used rule based

if a FOR UPDATE clause was specified. The 12.5 unique index rules are

now supported for 15.0

sp_helprotect [name [, username [, "grant"

[,"none"|"granted"|"enabled"|role_name [,permission_name]]]]]

decide to do one Index Scan for each non leading column (like is today) or do only one Index scan for reading all non-leading columns.

latching a buffer that is destroyed.

provide incorrect results.

890 error reading "The mass containing the buffer 'BUF=<Buffer Address>, MASS=<MASS Address>, Buf#=<Buf count>, page=<Page Address>, dbid=<DBID>, Mass vdevno=<VDEVNO>, vpage=<Virtual Page #>,

Buf lpage=<Logical Page #>,Mass stat=<Mass Status>, Buf stat=<Buffer status>, size=<Page size>, cid=<Cache id> ('<Cache Name>'), Pinned xdes=<XDES Adress>, spid=<Spid #>' should not be written

in its current state."

Solaris, it lists all the Sun Solaris Patches required on Solaris 8. However,The Patch 110648-13 and 109210-12 are required only when the customer is using Sun Cluster and ASE HA option. This is not clearly stated in our release bulletion. we need to revise this and give more precise statement regarding this, which recommend customer to examine the patches carefully and decide whether to apply such patches or not.

housekeeper 'pr_gcq' | 'pr_tsq' | 'pr_tsq', [<engid>]\n [...]

to

housekeeper ( 'pr_gcq' | 'pr_tsq' | 'pr_asq', [<engid>] ) \n [...]

so that correct the duplicate flag for pr_tsq, and will allow users to print account stats queue correctly.

http://infocenter.sybase.com/help/topic/com.sybase.help.ase_15.0.sag1/html/sag1/sag1392.htm

change:

To make an option or options take effect for every new database, change the option in the model database.

to:

To make an option or options take effect for every new user defined database, change the option in the model database.

Note: changes to the dboptions of model do not affect tempdb or user defined multiple tempdbs upon restart of ASE. Only newly created user defined databases and multiple tempdbs will reflect the dboptions of model, with the "create database" command. A restart of ASE clears the objects and data contained in the temporary databases, but does not reset the dboptions.

to the current procedure execution but for the whole session.

shutdown of cluster/instance may result in SIG 11 reported in errorlog. Stacktraces could

contain the words "log_license_info", "open__byname".

statement in a stored procedure.

executed. This is only seen with the dataserver binary, not the diagserver binary.

the port ranges will need to be manually edited

in the Sybcluster XML file.

So, enhance the function of creating cluster to allow user enter starting port.

sp_downgrade will force the server to be in single-user mode before downgrade is performed.

by file reader interfaces.

'cm_cid_from_name' while creating named cache.

to either the source or destination ASE it always prints the same error:

ERROR|The URL 'bruno-xp:5000' has an incorrect format. The correct format is: <hostname>:<port number>.

This is very misleading.

set repthreshold <value>.

terminate_process(). This is a very useful diagnostic. The

query text printing has been reinstated.

and a clustered index on this table.

kernel PCI(M1): ERROR: JNI_CreateJavaVM; failed retrieving the JVMTI handle in GetEnv() with -4

kernel PCI(M1): ERROR: JNI_CreateJavaVM; PCA/VM FAILED; not enough memory

kernel PCI(M1): ERROR: jsigcache_destroy; Signature cache destroy request without a JavaVM handle to destroy

kernel PCI(M1): -> PCA/JVM Signature cache is detached

kernel PCI(M1): Failed to destroy the VM signature cache.

with certain errors (like, for example, a 226 error "... command

not allowed within multi-statement transaction.") could unexpectedly

close the tables used by the session who previously executed the same

stored procedure.

in Adaptive Server. Adaptive Server has been modified to avoid terminating the task, and will print

an informational message instead. Recovery errors will be avoided, but ASE engine will hang until the operating system responds.

When installing on a machine that has ETL installed or if previously installed and uninstalled, the initial automatic login to the repository fails.

Workaround:

User can successfully log on by manually providing the password (transformer).

may be reported in the error log together with a stack trace which includes

the modules 'pg_allocate' and 'bufdirty' when a stored procedure is executed

requiring a work table creation and cache binding for user temporary databases

was changed between executions. This will result in ASE shutting down itself.

PCI(T0): ERROR: pci_wakeup; Unable to wakeup slot: 00, sema: 25DCAA74->(0), value: 0

and usually ends with:

PCI(T0): ASE_PCI: PCI Launcher Boss thread died. This is a FATAL ERROR (errno = -1)

and then usually a stack trace. After this, no java operations work although other ASE operations may continue normally.

Request enhancement of -T364 to let 15.x native optimizer support this trace flag on server level and provide a new criteria for session level.

update tab1 set col1 = 5 where exists (select 1 from tab2 where tab2.col2 = tab1.col2)

may sometimes be executed in deferred update mode when it is possible to use direct update

mode, resulting in a performance regression.

TEXT/IMAGE field which involves internal worktable

(eg. orderby, groupby) processing is not supported.

Rewrite your query to avoid conversion from non

TEXT/IMAGE datatype to TEXT/IMAGE datatype." maybe be

reported when executing a query involving an explicit

NULL value being converted to TEXT/IMAGE datatype.

threshold procedure that requires sso_role and when a password has been attached to sso_role.

The threshold procedure is successfully invoked.

Cross-platform conversion for database bigTest: 30401 pages completed (10%).

Application Tracing doesn't write a carriage return ("\n") with japanese messages.

That's hard to read the query plan in the trace file.

This is a feature request to put in place a prompt to save the SQL data.

'pg__updateoam_thru_allocpg' and 'pg__recalc_oamentry' after loading a database dump

which was taken on a very active ASE environment.

and stack trace may happen from ex__doprint():

1. Asian languages (Chinese, Japanese and Korean) are used as default language of a connection

2. a create table/view command with longer than 255 bytes multi-byte object name

3. the object name starts with one single-byte character and followed with multi-byte characters

SDESes may be claimed (One for each table beyond 16) for the duration of execution.

in sdes for object '<value>' - there is no room in sdes aux keep pool." in the error log

together with a stack trace which includes the modules 'bufkeep', 'bt__insert_into_leaf'

and 'bulk__fast_idxupdate' when inserting rows through bulk copy in a Data Only Locked

table whose non-clustered index tree is being updated by ASE.

processed incorrectly and lead to the wrong results.

a offline database such as dump tran might hang because space usage

in some segment may cross the segment threshold.

SQLServer message 632, state 2, severity 20:

Attempt to move memory with an incorrect length of 16416. Maximum allowed leng

th is 16384.

this query needs execution of the DEFAULT or RULE or references materialized column

using global variables.

and the alter table does not involve dropping a column, ASE produces a

SIGSEGV in cri_ncrebuild().

may hang when the query had to abort due to a 1105 error,

"Can't allocate space for object <table_name> in database

<db_name> because <segment_name> segment is full/has no

free extents. If you ran out of space in syslogs, dump

the transaction log. Otherwise, use ALTER DATABASE to

increase the size of the segment.", caused by insufficient

space in tempdb for the worktables that the query internally

needed.

adding a column with a name that duplicates an existing column name.

that has just hit a break condition to not execute successfully - it only occurs if the sql is contained within a single batch.

Pop up window:

Could not start the Sybase SQLServer _MY125 service on Local Computer.

Error 1053: The service did not respond to the start or control request in a timely fashion.

(click ok)

No error in the ASE errorlog.

than 32 groupby columns and with more than 6 joining tables might

hit signal 11 and stacktrace during the call of s_ap_compat_gen().

It could also happen for more than 32 groupby column query under

compatibility mode.

called from terminate_process().

in cases where there are only a few long-running transactions at the time

the server was shut down or DUMP DATABASE was executed.

cancel execution of the query by typing Ctrl-C in the isql, some of the locks acquired by

that query could remain unreleased and server could deadlock later.

12.5.x.

BCP IN and when run time recompilation are needed.

aux keep pool." or an 803 error "Unable to place buffer <value> from cache '<value>' holding

logical page ... either there is no room in sdes or buffer already in requested slot." when

reported will be followed by diagnostic information printed in the error log.

failback operation results in a hang if Java was used prior to failback on the secondary server.

Secondary ASE errorlog reports "scheduler infected with signal 11" message with a stack trace that

includes keipanic().

Section that reads:

Sybase is considering an enhancement in future release that will revert behavior (group by implicitly orders result).

In ASE 15.0 ESD #2, traceflag 450 makes group by use the classic (non-hashed) sort method, thus making

the result set order predictable again -- but at the possible cost of slower performance. In a later release, a new

"optimization criteria" language command "set group_inserting {0|1}" is being considered which will let you control

this on a session level -- and especially via login triggers -- without requiring trace flags.

Should read:

Sybase created an enhancement in 15.0 ESD #2 that reverts behavior (group by implicitly orders result). Global traceflag 450 makes group by use the classic (non-hashed) sort method, thus making the result set order predictable again -- but at the possible cost of slower performance.

In release ASE 15.0.1, with CR# 424188, a new "optimization criteria" language command "set group_inserting {0|1}" is available which will let you suggest this to the optimizer on a session level -- and especially via login triggers -- without requiring trace flags.

changed by the crashed instance may not be correctly recovered by the instance that does the failover recovery.

(or higher) causes ASE fail to boot with "Segmentation Fault error (core dumped)."

cells. So there are a relatively large number of rows that qualify for the associated

value.

A weight of 0 will lead to a row estimate of 1 and the optimizer will tend

to scan the associated table too early in the plan leading to poor performance.

Two problems have been indentified here. First, the boundary condition check for

building a frequency cell was off by one. Second, the calculations used to predict available steps for creating frequency cells was incorrect. The fix for these problems

is enabled through trace commandline flag 2715.

timestamp value, throwing the following

exception:

Cannot build locator from invalid hex string

<00000100f1d933c200000d6....0>

because: For input string: "f1d933c2".

of the deferred compilation feature could be switched on or of by the system procedure

sp_configure.

when executing a query involving complex expressions with

several nested built-in function calls (like for example

IS_NULL(), STR_REPLACE(), CONVERT(), etc.) or long CASE

expressions.

for a table created with the "ignore_dup_key" clause and having TEXT

columns may fail with a fatal 7114 error reading "Page <Page Number>

is not a valid text page." together with a stack trace showing modules

"sendtext" and "get_txtpage".

be reported by DBCC CHECKALLOC some time after instance failure as a

result of recovery not completely cleaning up pending deallocations

resulting from a DROP command that happened just prior to instance failover.

stored procedure 'proc_name'. Must re-create procedure."

may be reported in the errorlog and also sometimes the

message "current process infected with 11" in the module

'qremap__mapptr' together with a stack trace which includes

the modules 's_getTreeOrPlan' and 'qremap_prochdr' when

executing a very big procedure which requires more than

24480 pages of memory.

operator on APL tables.

signall 11 and stacktrace could happen in GtcChildOptBlockN::GtHasCorrelatedChildOb()

for an aggregate over a union view/derived table under correlated subquery.

some transactions may be unnecessarily blocked until the end of the recovery

redo pass before being able to proceed.

'enable literal autoparam' and 'statement cache size' are set.

sort operator's estimated and actual input set sizes.

This is needed to validate the sort operator's cost

model.

The " Average Row width is %f " message has been removed from

the sort op's showplan output. The message was misleading as it

was actually the "estimated average row width" and of very little

use without the estimated cardinality and the actual row counts and

average input set row size.

# iofenceutil /dev/rdsk/c5t1d0

Workaround:

Create the pass-through device special file for a specific device before running iofenceutil. In the above example, the pass-through device file /dev/raw/raw# for the device "/dev/rdsk/c5t1d0" should be created before running the utility "iofenceutil".

Primary attributes that get migrated have no type defined in generated XSM element.

Check Model returns an 'Undefined Type' error.

permission mask is set to 0777 equivalent. This requires mask to be set to rwx, for example:

sp_jreconfig "add", "work_dir", "/some/path(ugo+rwx)"

finds that the index and data row are inconsistent or an invalid row id is encountered

in the table.

reported in the ASE -errorlog which contains the modules 'dispatchpca', 'pca_get_config_value' and

'pci_command'.

replication may cause the DSI connexion to go down with error 1028, if the

statement references tables from a different database than the source of replication.

a join involving a union (from a view/derived table) where part

of the union's SELECT columns is in the DISTINCT clause

and another part is involved in the join predicate only.

equi-join in the correlation clause between a column in the subquery and

a scalar expression which contains the outer reference to a main query

column.

could cause wrong results.

if the trusted directory has been defined improperly (meaning that it does not conform

to the documented syntax) then the server may enter a permanent 100% busy CPU

condition for which there is no solution other than server restart.

Stack trace logged in the errorlog will show modules vu_colnames,colname_driver,

col_open_range

1. an equijoin join predicate between table1's col1 and table2's col2

(table1.col1=table2.col2), and

2. another equijoin predicate between one of the above joining columns

and an expression involving only column(s) which are also equijoined to

the other table's column(s), such as table1.col1=isnull(table1.col1,table1.col1), and

3. a store_index (reformatting) plan is chosen for the nljoin inner side

between those two tables.

on the inner block's tables. When the store_index is created,

duplicates were not being removed. This can be a significant performance

issue for correlated subqueries that are theta-joined AND

there are additional predicates that are not used to position on

the reformatted table. The following query is an example:

delete outer

FROM outer

WHERE EXISTS (SELECT * FROM inner

WHERE DATEDIFF(year, innter.ValDate, outer.ValDate) = 0

AND DATEDIFF(month, inner.ValDate, outer.ValDate) = 0

AND outer.ValDate < inner.ValDate

)

Note that the outer and inner tables are theta-joined

(outer.ValDate < inner.ValDate). And that the DATEDIFF builtins

must be evaluated after rows are qualified by the theta-join.

raised together with a stacktrace in the ASE errorlog containing the modules 'eudf__convertSQL2Java',

'execudf' and 'LeRun'.

But in fact, dbcc checkalloc performs check of OAM page entries and does not check of Text column chains.

is not configured large enough, sometimes it could cause the stack corrupted and ASE goes down.

database that has mixed log and data on an instance that has a single engine,

might hang as a result of a livelock.

table as an intermediary, when the target table contains both an identity column and a text

column.

may hit SEGSEGV together with stack trace in the module of LeDeleteOp::_LeOpNext().

has been changed to 18.

from the install components list.

followed by a stack trace in the errorlog.

An extended column aggregate is a non-ANSI sql feature that Sybase introduced, which allow a column to be placed into the select list of a GROUP BY aggregate, which is not included in an aggregate expression nor is it in the GROUP BY list.

The processing of the view silently fails in this case and returns no results.

CREATE TABLE PROB_TABLE

(

A int NOT NULL,

B int NOT NULL

)

go

insert into PROB_TABLE values (1,2)

go

CREATE VIEW PROB_VIEW1

AS

SELECT PT.A as PT_A,

PT.B as PT_B

FROM PROB_TABLE PT

go

-- note the extended column aggregate in PROB_VIEW2

CREATE VIEW PROB_VIEW2

AS

SELECT PT_A

FROM PROB_VIEW1

GROUP BY PT_B

go

-- the following returns no results and silently fails

select * from PROB_VIEW2

go

and fails with the following error:

kernel PCI(M0): PCI Bridge run-time check failed - Bridge is not enabled (expected: AIX 5.3 or later found: AIX 5.2 or earlier)

To reduce CPU usage and spinlock contention, please change one parameter via

sp_configure 'lock address spinlock ratio', 1

query whose inner index scan has the same key column sequences as the ORDER BY

clause but with different direction on the leading column (i.e. ascending

vs. descending).

recompiled during subsequent executions.

causes an assertion failure.

Customer would like to get his database in "read only" mode, but having his maintenance user (Customer is using replication) still be allowed to apply commands.

Customer would also be able to dump his database (as maintenance user could fill up syslogs)

To summarize: Customer request for a new database status, where all user may have a read only access to data, and where only user with "replication_role" and/or "oper_role" (needed for dump/load command) or DBO user will have normal access permission.

15.0.2 finishes in 386400 ms. This is plan related, because we ran the 15.0.1 AP in the 15.0.2 server, and it finished in 1100 ms.

that have zero used/unused page counts. When the oam deleted entry is the

last oam entry on the current oam page, ASE may mistakenly terminate this

command execution without completing the oam cleanup.

non-leaf page, so it can't check out the corrupt index if someone of first keys is bad.

functional index. The correct error should be 11029.

when executing an outer join query with aggregate functions in the

SELECT list after the query plan is cached in procedure cache.

This problem could occur if statement cache is enabled or the query

in question is in a stored procedure.

a shared disk cluster so that the cluster is able to coordinate transactions

on a remote Adaptive Server.

than in previous major releases. REORG COMPACT and REORG RECLAIM_SPACE utilities will

now clear out OAM entries with zero used/unused counts.

"NT operating system error 87 in module 'nt386\ksource\blkmem\osmntmmap.c' at line 479 The parameter is incorrect"

followed by "kernel os_create_region: Failed to get virtual memory information".

attempts to execute the procedure raise errors 208 and 913. Using EXECUTE 'procedurename'

WITH UPGRADE corrects the problem.

11 together with a stacktrace being printed in the ASE errorlog. The stacktrace contains the

modules 'Cast_UnitextData_to_String' and 'jvmfini'.

may be hit (1105 error) whereas 'sp_helpdb' shows that free data space

is available. Alternatively, for mixed log and data databases, 3475

errors may be raised under the same space usage conditions.

will still recognize underscores, such as 'master_dev'.

include the modules 'srvdes_alloc' and 'omni_connect' may be reported in the errorlog

when accessing a proxy table or remote server when ASE has no more free memory. As a

workaround, increase the "max memory" configuration parameter.

may be reported in errorlog while running the DBCC CACHEDATAREMOVE command.

- Display Backup server name in usage of stored procedure sp_volchanged.

- Restrict instance name in Cluster not to exceed 27 chars.

count(column) in the select list of an outer join query could be

wrong.

re-create procedure." may be reported in the errorlog, or a 703 error, "You cannot run this procedure,

trigger, or SQL batch because it requires more than 24480 pages of memory. Break it up into shorter

queries, if possible." may be returned, and also sometimes the message "current process infected with

11" in the module 'qremap__mapptr' together with a stack trace which includes the modules

's_getTreeOrPlan' and 'qremap_prochdr' when executing a very big procedure which requires more than

24480 pages of memory.

the like clause and other expressions within derived table expression and there are at least one

local variable referenced in the derived table expression.

to grab a group of identity values reserved for that process.

incorrect huge number result for the server pagesize. When you run them together, it is

correctly reported as 2048 bytes. Problem is reproduced on windows only, dbisql or isql.

See repro section.

may hit SIGSEGV in make_negative_obj_id() and may cause one engine to fail.

user resource limits and audit settings

and a group_hashing strategy is used to calculate the count,

then an incorrect count of 0 will be returned. This could occur if a

group-by count aggregate occurs in a derived table or a view

and the derived table or view is used in turn as the inner member

of an outer join.

thread when the configuration 'cis idle connection timeout' is set to a non-zero value.

when one of the instances is brought down in the cluster.

when doing sp_configure and the config file is truncated.

with compatibility mode and raise warning messages accordingly.

hung in latch sleeping and unkillable.

the columns in the select list) is larger than the max row size for the server, then

the optimizer will avoid sort nodes at all costs to satisfy an ORDER BY clause.

This means that a very expensive sort node may be placed much earlier in the query

plan, in which the max row width is smaller. This could cause huge swings in

performance, if the number of rows at the top of the query plan is much smaller

than earlier in the join order. The allow_wide_top_sort optcriteria will avoid

checking for the max wide, for optimization purposes only, and allow a sort node

to be placed on top of the query plan for purposes of satisfying an ORDER BY.

This is consistent with 12.5 behavior for placement of sort nodes for ORDER BY.

The assumption is that rows formed from a select list of a query which has an ORDER BY

will fit into the sorter.

and returns the error "ssl tcp is not a valid protocol type on this platform"

kernel Detected failure of the Primary interconnect network link with instance <n>

SHOWPLAN ON always contained plans from a number of other statements which

the user did not directly execute. This has been fixed to restore the

9.0.2 behavior which is as follows:

If DBISQL is configured to display results as text, or read-only grids, the

SHOWPLAN output is just for whatever statement(s) you explicitly execute.

If DBISQL is configured to show results as editable grids, it requires more

metadata. That metadata is provided by the JDBC driver (JConnect) which

makes a number of database requests on our behalf. These requests

themselves cause SHOWPLAN output which users might not expect, but there is

nothing further which can be done about it. DBISQL also includes a plan

viewer window which is a good alternative to the textual SHOWPLAN.

An extended column aggregate is a non-ANSI feature, in which a reference to a column is made which is not in the GROUP BY list, and not contained in an aggregate, and is outside the WHERE clause.

SELECT

count(fifo_realized )

FROM

( SELECT * FROM OLOpenPos

UNION ALL

SELECT * FROM OLMaturedPos

) pos

ORDER BY

external_lentity_str

go

In the case above, the ORDER BY column is not in the WHERE clause, and there is a scalar aggregate in the select list. This results in 2 evaluations of the union all derived table, which could have wrong answer or signal 11 issues.

when a complex predicate that has strong filtering can be applied to the index. Instead

a non-covered more expensive index may be chosen instead, or a different join order may

be chosen. This can be seen when the "set statistics plancost on" output shows an

estimated row count for a covered index scan which is higher than the row count for the

rid join over a non-covered index scan. The behavior change is associated with the

optcriteria cr559034, which can be enabled with "set cr559034 on"

by an cancel attention generated from a killed connection.

view if there is a sargable correlated predicate on the column from the DISTINCT

view.

the new joint instance will have different run value on the new instance.

inserting values into the table with identity column.

may generate Msg 10334, "Permission related internal error was encountered". This

error may occur when the procedure execution follows the drop and re-creation of a table referenced in the derived table expression.

ASE may fail to boot with messages:

"SSL Plus security module EZ_SECURITY_BUILDER_DSA_MODULE load failed (2030).

SSL Plus security module EZ_SECURITY_BUILDER_RSA_MODULE load failed (2031).

...

SSL Plus v5.0.4 security modules loaded, 14 errors. Continuing startup without listed modules "

also releated to the count field in TDS_DONE this is a INT.

datatype (TEXT, IMAGE, UNITEXT) as an input parameter may generate an access

violation/segmentation fault with the function, LeAlignedDataRow::readRow, in the calling

stack. This failure can occur if the query plan contains a blocking operator (i.e. HASH

JOIN, MERGE JOIN, STORE or SORT OPERATOR in showplan output) above the SCAN OPERATOR that

reads in the LOB column value.

not work correctly if the query objects are not prefixed by the owner name.

- at least one SCAN OPERATOR is executed in parallel and

- that SCAN OPERATOR does 'Dynamic Partition Elimination' (as seen in showplan) and

- that SCAN OPERATOR evaluates a predicate on the partitioning column of a range partitioned table containing the "<" or "<=" operator (i.e. where ptncol <= @v),

too few rows may be returned.

configuration option on encrypted columns declared with a decrypt_default

value under certain circumstances.

correlation name, it fails to raise 7348 error.

For example, The following queries should raise an error:

select a.col1 'name', b.col1 'name' from a, b order by name

select a.col1 'name', b.col1 'name' from a, b

could be returned when executing a statement having in its

WHERE clause java user defined function calls for objects

referenced in the statement with local variables.

without any sarg and the 302 index selection diagnostics would incorrectly

list the index to be covered. It could happen for a non-correlated subquery

which itself has a correlated subquery.

has hit an exception in its previous executions.

The functions LeScanOp::_LeOpClose()or LeInsertOp::LeOpen()may appear

in the stacktrace.

maybe raised when doing delete operation on a DOL table that is bound as hot object using

DBCC TUNE 'des_bind' command.

is defined with a DEFAULT clause. In the same situation, CREATE PROXY_TABLE fails silently

and does not return any error.

unique indexes are present.

under the subquery could lead to stacktrace and access violation in

CgpMergeJoin::_cgpInit().

When we use the set command, we only see 1 beginxact and 1 endxact for a select

into proxy_table, whereas sp_configure "cis bulk insert batch size", 200 will

give us 1 beginxact and 1 endxact for every 200 rows inserted.

used to establish a CIS connection using SSL, is terminated.

contained in the jar does not represent a valid Java class file." can be raised.

of vtabremoteaccess.cpp might be printed or ASE Cluster Edition Error 12035 "Failed to return a row

for the monOpenObjectActivity monitoring table." may be returned when customer tries to create a

temporary table with a a table name >10 characters. The same error might occur for smaller temporary

tables in case the user created tempdb has a large enough name

when there is mixed datatype equijoin between the RI column and

another column. The similar mixed datatype equijoin under a view could also

lead to wrong result when the view is joined to itself and has to be

materialized into a worktable first.

table contains a unichar column of size larger than 127 characters.

And it can finish the work with only one connection.

registered in the dbcc_exclusions table like sp_dbcc_faultreport does.

during implicit conversion of INT value '<nn>' to a NUMERIC field"

Replication and owner" is solved and supported by Replication Server 15.2 ESD#1 onward.

where col1 =-(1)

union, it could lead to SIG11 in ReslopOp()->EvalConstExpr().

improved and now contains more information. The new message suggests that the sybpcidb database

may be missing, or it may need to be initialized with the installpcidb script.

query joining two or more tables having in its WHERE

clause a comparison between two expressions which

both result in a java object and the plan selected by

the optimizer decided to use a reformatting mechanism.

for a consolidated license file

when running a query joining a local table with a proxy table

after update statistics has been run on the proxy table.

errorlog for the module cmcc__getphysicallock.

expression from a view or derived table is SELECTed from an outer

join view inner side and the constant expression is further participating

in a simple WHERE clause (comparison involving other constant only)

when the outerjoin view is queried.

"unnamed-col" for predicate printing instead of the real column name.

a literal value. The workaround is to use a literal value and then use ALTER TABLE REPLACE to

replace the literal value default with a function default.

cluster failover, LOAD DATABASE and LOAD TRAN, because asynchronous prefetch

is incorrectly disabled.

them rather than fail with an error, and to reveal their existence through a new lock

type entry in syslocks.

downgrade. It should provide more infomation like which objid and indid the row has.

may not complete leading to stack trace.

If you are running 12.5.4 ESD1 or later on Linux 2.6 kernel, you can enable direct i/o for all devices by booting with traceflag 1648. This should be used in conjunction with 1649 to enable native Linux kernel async i/o.

should have at least 512k or 25 buffers whichever is greater. However in ASE Cluster Edition

this rule is broken when the cluster has more than single instance configured.

information about backups in the MDA tables to track :-

- backup start/end time

- datasize - raw and compressed (if compression used)

- backup type

As while building installmaster in NT platform, it won't generate the statement that could insert

version info into sysattributes.

JavaVM Threads execute, both operating for PCI Bridge managed memory, they could potentially

lockup because of a PCI Memory management flaw not yielding other ASE Tasks or Threads.

This problem could finally lead to an ASE server crash under shutdown or full hang when

in shutdown.

An option is checked and cluster setup log is saved.

In order to browse path of setup log in the following steps, when the "Browse..." button is clicked, Cluster Administrator will fail often.A Cluster Administratoe process will be extinguished.

and the spid terminated when selecting from a view containing a union all

and the select includes > 1 variable assigment from the same column.

corruption, such as first key of one non-leaf page is bad. The new 'force' option of

sp_post_xpload will force rebuild all the index except APL clustered index.

object and waiting to acquire lock on another remote object. This is due to limitations in the

deadlock detection algorithm. Optimization has now been added so that the algorithm will continue

with deadlock detection if tasks holding shared lock are waiting on less than 9 distinct remote

objects.

that are replicated using SQL Statement Replication may cause Rep Agent to

shutdown with error 9202, State 1 when attempting to process log records

related to SQL Statement Replication.

we might hit the 1127 error which means that the OAM pages of the object got full.

CgLavaUtils::CgTargetListExprs() might happen for query with an equi-predicate on

two columns of the same index.

when the text data is longer than 4KB and starts with an XML declaration element and

contains multi-byte characters.

particular sql statement has increased to be in excess of the previous limit of 1024

chars in sp_monitor_deadlock. (The columns are replaced with dummies in 12.5.4)

Cannot use DROP with 'test_c1_427145536' because 'test_c1_427145536' is a

constraint. Use ALTER TABLE.

But the command you do is drop default test_c1_427145536 and the message is because

you try to drop a created by declaration default.

.

all the binding information is stored in the sysattributes table in masster

database related to Multiple temporary databases ... those bindings are missing

for all their logins in the upgraded server.

So now this customer has to recreate all the bindings done with sp_tempdb in the

past again now from scratch.

The customer wants the feature of using dump and load to upgrade their database

versions and have all the binding information related to Multiple temporary

databases for all the logins to be present in the upgraded server.

maybe the feature would be

1) to relocate the binding info to some table other then sysattributes

2) create a stored procedure or instructions to migrate the Multiple

temporary database binding info from the older server to the

newer server.

is done on one of the instances, error 968(upgrade in progress) may be reported in the

errorlog while using database on another instance in the cluster.

To resolve this error, ensure that all users are out of the database, or reboot ASE and run:

dbcc tune(des_unbind)

customer changed locking scheme from APL to DOL and now the optimizer fails to choose the most ideal query plan. the customer had to execute ... 'update index statistics' ...

before the optimizer would succeed again in selecting the most ideal query plan.

WARNING! Changing the locking scheme on a table can cause the optimizer to alter its costing estimates for accesses to the table. The impact of changing the locking scheme should always be tested in a non-production environment before deployment to production.

Given that the clustered index on an APL table is a very direct

mechanism to access the data AND that there are Search Arguments

on all three columns of the clustered index, ASE most likely

bases its decided upon join order around a clustered index scan

of matter_participant.

Given that clustered indexes on DOL tables are not as strong as

they are on APL tables ASE ends up selecting a non-optimal index.

Update Index Stats ... 'update index statistics' ... after this

is executed ... ASE will then have histograms on additional

columns and then ASE is better equiped to select the correct

join order and query plan.

Conclusion:

the conclusion here is that changing locking schemes is not

directly tied to the need to run 'update index statistics',

rather, the additional histograms created by 'update index

statistics' will, in some circumstances, be required in order

for the ASE 12.5.x optimizer to select the most ideal query

plan.

pose server hangs when multiple ASE engines are in effect. This problem

is filed at IBM under PMR 51261,756,000.

A workaround for those server hangs PCI Bridge Virtual access approaches

are disabled under IBM AIX 5.3 ML7, making it the default server behavior.

This despite an understood performance loss.

IBM AIX servers can be booted with traceflag -T9403 allowing them for full

PCI Bridge access mode. Doing this, such poses strong risk to server hangs,

but won't cause performance loss on operations.

in instancename.

The stack trace could contain strings like "hk_chores_proc", and "bufwritedes_log_all".

and clients modify data in that database from the same node or from remote nodes,

Rep Agent may replicate some data with a delay.

table "systabstats" have the wrong data types.

ColName expected dtype actual dtype

pioclmdata real int

pioclmindex real int

piocsmdata real int

piocsmindex real int

illegal token was received" may result when accessing data

containing null bytes on an ASE server when character set

conversion using Unilib is employed.

destroyRemoteAccess and VTABRemoteAccess::`scalar deleting destructor'

Workaround:

Select the desired object in the tool palette and add it to sequence. The object is added at the end of the hierarchy

run and this query inserts into a proxy table that is created on a remote view and the

remote view is created on a table that contains an identity column and the datatype of the

identity column is not NUMERIC (i.e. it could be integer or smallint) and the remote view

has an instead-of trigger for insert and the instead-of trigger contains a query that

selects the identity column from the inserted pseudo table.

3452 regarding failure of upgrade steps 21 and 1501.

This is why ASE reserves space for the new security features in the network buffer memory.

You still need to increase additional network memory to extend the network buffer memory if you increase max network packet size.

Msg 632: Attempt to move memory with an incorrect length of 25600. Maximum allowed length is 16384.

a VIEW attached with an INSTEAD OF UPDATE TRIGGER on a text or image column may result

in Msg 156 "Incorrect syntax near the keyword 'WHERE'" error and/or a stacktrace in

'omni_tname'.

transaction. Update statistics changes will become part of the

transaction, and and all the work done by update statistics

will be removed by any rollback. If a multi-statement

transaction does not exist, then update statistics will continue to

use internal transactions to commit multiple minimal atomic parts

of the update statistics command which will remain even if the

command is aborted. Trace flag 2720 can be used to

continue with the pre-fix behaviour, in which case an attempt to run

this command in a multi-statement transaction is disallowed with an

error message.

the server don't update actual page number read so that this check failed with table corrupt.

add partition results in local indexes on the new partition(s) if a local index is present on the

existing partition(s).

If it doesn't exists, the page will be read from disk.

With the Cluster Edition, DBCC PAGE will first check if the buffer of the specified page is in the cache of the local node.

If it doesn't exists, it will check the caches of all nodes before reading the page from disk.

may hit stack trace in the module of metrics_write() together with curs_open()

on hpia64 platform.

target table column non-nullable while the early behavior

of ASE would make the target table column nullable even

when the ISNULL result can't be null for the query.

instance in the multinode cluster crashes.

direct mode may hit SIGSEGV together with stack trace in the module of LeDeleteOp::_LeOpNext()

if the query plan has merge joins.

compatibility mode for GroupBy query with a lot of aggregates.

error 414 saying the key size of a worktable exceeds the maximum allowable limit.

non-binary sortorder servers (which are typically case-insensitive). This will only

occur if reformatting is chosen for the join strategy. This could only

occur if an equi-join predicate contains (var)char columns and the distinct

select-list contains one or more columns that are not in any of the equi-join

predicates.

may sometimes result in a message "current process infected with 11" and the stack trace reported in the errlrog file if the procedure performs SELECT query on this table.

This is likely to happen if a procedure query tree is renormalized.

Stack trace may reference modules pre__replace_cpc and s_getTreeOrPlan.

C:\WINDOWS\system32>isql -Ulogin_ing

Password:

1> xp_cmdshell "dir"

2> go

. . .

02/28/2003 09:38 PM 113 zonedon.reg

2032 File(s) 379,119,114 bytes

48 Dir(s) 28,418,981,888 bytes free

(2087 rows affected)

(return status = 0)

C:\Sybase\ASE-15_0\scripts>isql -Usa -SASE15 -i instmstr -o saida

Password:

C:\Sybase\ASE-15_0\scripts>isql -Ulogin_ing

Password:

1> xp_cmdshell "dir"

2> go

Msg 10330, Level 14, State 1:

Server 'ASE15', Procedure 'xp_cmdshell', Line 2:

EXECUTE permission denied on object xp_cmdshell, database sybsystemprocs, owner dbo

(return status = -4)

To solve the situation :

C:\Sybase\ASE-15_0\scripts>isql -Usa

Password:

1> use sybsystemprocs

2> go

1> grant execute on xp_cmdshell to login_ing

2> go

1> exit

C:\Sybase\ASE-15_0\scripts>isql -Ulogin_ing

Password:

1> xp_cmdshell "dir"

2> go

. . .

2032 File(s) 379,119,114 bytes

48 Dir(s) 28,418,904,064 bytes free

(2087 rows affected)

(return status = 0)

1>

the last log page.

with sampling are concurrently executed on the same Data Only Locked table that has

at least one non-clustered index present.

In the event of a hang, the system 'ps' command will show an extra server process running, i.e. in a two engine system 'ps' would show three server processes rather than the expected two.

If the error results in a dead engine on the secondary companion, the error log will contain messages indicating the engine hosting the JVM has exited with a signal 11.

"Command has been aborted" error.

sp_engine 'online', together with Java statement are combined to one single SQL-batch when

the target engine used in the sp_engine command is the engine hosting the Java VM.

For example, select NULL + NULL + 'abc'.

are not treated with accent insensitivity.

's_free' with a stack trace which includes the modules

'kill_proc', 'metrics_dump' and 's_recompile' may be reported

in the error log after killing a session that was executing a

stored procedure which had to be recompiled and it was hang

because transaction log was almost full and the configuration

option "enable metrics capture" was set to 1.

this is not allowed as DOC CR 493203 says.

cluster. Any subsequent attempt to boot the server with the new configuration file thus generated will fail.

(<rid_pageid>, <rid_rnum>) for transaction (<sessionid_pageid>, <sessionid_rnum>)

that started at '<date>' in database '<dbname>'. Please contact Sybase Technical Support."

is added to replace few occurences of error 3301.

for dumping the log records of a particular session.

together with a stack trace that includes modules 'rec_undo_session' and 'abort__all'

when a transaction that allocated space in a data-only locked table is rolled back.

the usual message for error 701 " There is not enough procedure cache to run this

procedure, trigger, or SQL batch. Retry later, or ask your SA to reconfigure ASE with more procedure cache." is accompanied by a stack trace in the errorlog file.

The stack trace will indicate modules LePlanRelease and LeParentOp::LeOpProcessIOStats.

The issue will happen if options to display statistics were activated.

modules like 'bwfinish', 'exec_eop', etc or the message

"timeslice <negative number>, current process infected"

may be reported in the errorlog after a query with an ORDER BY

clause was executed which internally used a worktable that due

to its large size was created with the "Data Only Pages"

lock schema and Adaptive Server was configured for running

queries in parallel.

Right now an internal action by ASE conducted as 'sa' will appear to be the same as an action done by one using an external login as 'sa' and this causes auditors to believe there has been a breach of the security policy/rules regarding the use of the 'sa' account.

stacktraces being reported in the ASE-errorlog if there are users connected to ASE executing

Java in SQL statements.

found under that name. Make sure that name is entered properly." may be raised for a select into

query where dbo is specified as the table owner and the table contains a user defined datatype.

the local 'default data cache' for an instance when we reboot

the instance, the local 'default data cache' may sometimes

get re-instantiated after the reboot.

returning these rows. The problem is specific to non-sargable predicates (expressions.

in-lists, and or-clauses) where the predicate is placed in a RESTRICT operator below

an XCHG operator. For example:

select * from r where r.id = NULL or r.id = 8

This query wouldn't return rows with an id column value of NULL if a parallel scan

is done on "r". Note, the problem is unique to sessions where ansinull semantics are

off.

'memcpy' together with a stack trace which includes the

modules 'LeRun' and 'LeJoinContext::computeVirtualCols'

may be reported in the error log if a SELECT UNION ALL

query is executed involving an outer join, a TEXT column

from the inner table is selected and some NULL values

are returned.

requires it to use a cursor but ASE has disabled the support of updateable cursors against IQ.

on usage of the device immediately after creation of the device using 'disk init'.

However the unique index is not always suitable for the query.

often, Adaptive Server may report a false 863 error,

"Buffer resources in cache <cachename>, id <cacheid> are unavailable.

Please re-run this query or ask system administrator to re-configure buffer cache memory"

even when there are enough buffer resources.

15.0.3 and 15.5 ESDs. Previously, optimizer fixes which effect plan selection

were only only enabled by configuring "optimizer level" as "ase_current". The

ase_default optimizer level was ase1503esd1. The optimizer level for ase_default

is now ase156ir.

DISTINCT view and not all of the view's SELECT columns are participating in

the subquery's correlated equijoin predicates.

The work around is not to run SP commands right away after node shutdown or node failure

are skipped if plans are evaluated before the exhaustive search engine is entered.

The symptom may be seen by forcing a query plan which has a better

estimated cost than the query plan that is eventually chosen.

<cache name>..." may be reported in errorlog while running sp_poolconfig.

more logical IOs than expected.

from a segregated log and data database.

column, special OR strategy might not be used. ASE now provides a optcriteria

'use_mixed_dt_sarg_under_specialor' which when enabled, will allow special OR strategy to be

considered in case of mixed datatype SARGs in the IN/OR list. Additionally, traceflag 434 has been

provided to turn OFF this feature. The traceflag setting overrides the optcriteria setting.

sarg (col=<cnst>) on it. Similar wrong result could also happen for an equijoin

between a GROUP BY view/derived table and other table when there is a constant sarg

in the view for the joining column.

'select into/bulkcopy' option for the database is unnecessarily changed.

FOR UPDATE can not be used in a SELECT which is not part of the declaration of a

cursor or which is not inside a stored procedure.

parameters as BIGINT datatype.

trigger may encounter a SIGSEGV (infected with 11). "xact_sync_trigger" will usually be referenced in the stacktrace.

Bad pointer 0x4 encountered while remapping stored procedure.

This may be seen in the error log on execution of sql statement containing constants in

both the predicate and in a group by clause.

Should only happen with 'enable query metrics' enabled at point of proc creation.

If the algorithm is chosen then index_union appear as a hash union of several index scans followed by rid join in the showplan output.

The abstract plan would contain a hash_union_scan if this algorithm was chosen with dirty reads on DOL tables.

command.

allocation rules broken. Current page: %d; next page: %d; piece end: %d;

end-of-scan: %d.", may be reported while creating a in-memory database

using a template database which does not have full durability.

the sp_cacheconfig interface, fails due to insufficient 'max memory',

the suggested 'max memory' value printed as part of the following

error message:

"The current 'max memory' value '%d', is not sufficient to change

the size of cache '%s' to '%dM' (%d KB). 'max memory' should be

greater than 'total logical memory' '%d' required for the configuration."

is not sufficient if the server is rebooted after the change.

when the remote table also triggers a change in one of the column values being inserted.

Stack trace will report upsleepgeneric, mda_take_mutex, fdpsrv_init_reader.

in cases where many DMLs are simultaneously done on the same page.

IBM AIX 5.3. This problem comes down to a suspected IBM AIX 5.3 O/S issue that is filed

with IBM under PMR 51261,756,000. A design workaround was applied to ASE for IBM,

avoiding this problem and is default active but will cause performance problems to

those specific ASE Tasks as they all only execute from the ASE engine which is

physically bound to the JavaVM.

To overcome the functional and performance issue with ASE under AIX. the ASE should be

booted with traceflag -T9403 and have prior to the boot AIXTHREAD_SCOPE=S environment

variable setup under the shell responsible in booting the ASE.

in the error log you only see this.

00:00000:00016:2009/06/17 12:00:25.65 server Error: 2238, Severity: 20, State:9

00:00000:00016:2009/06/17 12:00:25.65 server QUIESCE DATABASE: Could not acquire begin lock due to LOCK INTERRUPTED.

But this spid are run begin tran in master database (this only happens on master database) and a other spid have issued a quiesce database master_tag hold master plus that a spid run the begin tran in master database receive a attention.

DISTINCT view and not all of the view's SELECT columns are participating in

the subquery's correlated equijoin predicates. Trace commandline flag 15315

can be used to workaround this problem.

the OAM page when running heavily concurrent multi-user applications

simultaneously inserting into the same table.

for query involving distinct operator and exists sub-query

may result in stack trace. This happens if a procedure is frequently renormalized. The stack trace shows the module "lastleft".

in the error log together with a stack trace which includes the

modules 'Optimize' and 's_compile' when executing an UPDATE

statement involving a join between two or more tables and the

table being updated has defined a functional index involving

columns which are being updated with values from columns

from the other table(s) participating in the join.

considered if stack traces are seen in 15.0.3 ESD#4 involving des_wkinstall()

For 578797-1, we added some codes under trace flag 7751 to clean up worktables at the end

of execution if worktables are not dropped.

DO NOT run TF 7752

which performs the UPDATE or DELETE with the selfjoin is executed by the non owner of a procedure

and the execution is performed using the deferred compilation mechanism.

queries using expression involving LIKE and variables or parameters ASE may fail to

generate parallel plan and report stack traces in the errorlog file. A trace will

show modules copy__eval and copy__eoptlike.

table with text/image columns.

segmentation violations may occur on tasks doing concurrent updates on the

same database, in ASE procedure lt_notify_flush().

has objid %ld, indid %ld, and used bit on, but reference bit off.", may

be reported by DBCC CHECKALLOC() after failover recovery in case of a

segregated log and data database.

to unpredictable results when their methods are executed within the JavaVM.

server due to the LDAP_SPIN spinlock being held may be seen involving the housekeeper

whilst checking ldap configuration.

reported even if user does not have valid permission when load transaction is executed for a temporary database.

If a parameterizable statement has a literal value whose precision exceeds the limit of

decimal\numeric (38) then a 2756 error may be received on recompile followed by a 706 error

and stack trace in the error log involving the function rm_prochdr(). The spid will be

terminated.

not explicitly specified in the command statement, resulting in an Oracle ORA-01442

error. This issue is specific only for Oracle and a traceflag 11237 is used to enable

this fix.

timestamp from log...", may be reported during ONLINE DATABASE after

loading a transaction log from a 12.5 server, into a 15.0 or later server

during which the message, "The pre-15.0 log for database '<db>' includes

a CREATE INDEX incompatible with this server version and operating

environment..." was reported.

Current syslogs free space does not match what dbcc counted.".

the remote server is started up later and the connection is established.

it could return wrong result if later this table is changed to a partitioned table when executing this procedure.

may be reported while running dbcc checktable() after running load database on a hppa64 platform from a dump which was taken

on a linuxamd64 platform.

This is a request to add a status bit indicating read-only and therefore not dumpable.

is reported for the database which does not need to be recovered.

may be reported during LOAD DATABASE of a database dump that was taken of a database with durability

NO_RECOVERY or AT_SHUTDOWN and a deallocation occurred while DUMP DATABASE was active.

As a workaround, dump the database while there is no transactional activity.

is enabled, then explicitly updateable cursors (declare statement contains and

update-clause) and dirty reads on APL tables could lead to a 311 error:

open mycursor

Msg 311, Level 16, State 2:

Line 2:

The optimizer could not find a unique index which it could use to scan table

'mytab' for cursor 'mycursor'.

This was observed after QPMetrics lazy dumping was introduced. The change

fixes the problem.

appear in the select list) and if the select list includes an encrypted

column, the execution of the query will hit a stack trace reporting a

null encryption key.

An exemplary query is:

table t2 (a2 int, b2 int encrypt with key1)

select t2.* from t1, t2

where t1.a1 = t2.a2

group by t1.a1

tables and the schema counts of the view upon which the instead-of trigger is created and

the log (syslogs) do not match, a 11060 error may be raised, resulting from too many (10)

attempts to recompile the query plan without succeeding.

and/or theta-only joins.