DBCC Secrets
Shhh
Mark Kusma
Manager, Tactical Support
Sybase CS&S
August 9, 2007
DBCC Secrets
What this session is
We will look at several DBCC commands that are primarily
undocumented
We avoid commands that will intentionally crash systems, or cause
major problems
We look at some commands that must be used with caution
What this session is not
This is not an exhaustive list of DBCC commands
There are close to 300 DBCC commands
DBCC Secrets
Well cover the syntax, version info, usage, and
potential issues of several DBCC commands. Well
also look at when and why to use them.
These commands should be used with
dbcc traceon(3604)
Not shown on subsequent slides
dbcc cacheremove
Syntax
dbcc cacheremove(dbid | dbname, objid | objname)
Version
Added prior to ASE 11.9
Usage
Used to remove a descriptor (DES) from cache (open objects)
If object is a table
the IDES (open indexes) and PDES (open partitions) is also removed
Writes dirty buffers to disk
Does NOT remove pages from data cache
In-memory statistics flushed to systabstats and sysstatistics
Places the DES back on the FREE chain
Cannot be used with system tables
Must have sa_role
Cannot be granted
dbcc cacheremove
Potential Issues
Object may be in use and cannot be removed from cache
When and Why
Use when system table info has been manually updated for this object
Example, changing uid in sysobjects to change owner
Use when scavenging is taking place and you cant increase the number of <item>
config param due to memory constraints
Either remove rarely used objects, or remove many and let ASE recache the active objects
Use to remove single statements from statement cache
See new option on dbcc purgesqlcache 15.0.2 Whats New
dbcc cacheremove
sp_monitorconfig 'open objects
go
Usage information at date and time: Jul 11 2007 10:37AM.
Name
Num_free
Num_active Pct_act Max_Used
Reuse_cnt
------------------------- ----------- ----------- ------- --------------------number of open objects
447
53 10.60
53
0
dbcc cacheremove(test,p_abc)
go
sp_monitorconfig 'open objects
go
Usage information at date and time: Jul 11 2007 10:38AM.
Name
Num_free
Num_active Pct_act Max_Used
Reuse_cnt
------------------------- ----------- ----------- ------- --------------------number of open objects
448
52 10.40
53
0
dbcc checkindex
Syntax
dbcc checkindex(tbl_name | tabid, indid [,bottom_up | NULL [,ptn_name |
ptnid] ] )
Version
Added in ASE 12.5.3 under CR 338495
Usage
Similar to checktable, but only executes for one index
Option bottom_up only for DOL tables
Partition option applies to 15.0, and higher
Must have sa_role or be the table owner
Can be granted, however, granted at database level (or all databases)
dbcc checkindex
Potential Issues
Index corruption may be found, requiring action
When and Why
Use when checktable will take too long to process all indexes
Use bottom_up to verify each row of a DOL table has a corresponding index leaf row
Catch 644-type errors
If error found, rebuild index and check others
dbcc checkindex
dbcc checkindex(abc_dol,2,bottom_up)
go
Checking table 'abc_dol' (object ID 784002793): Logical page size is 2048 bytes.
Table has 99 data rows.
Index has 99 leaf rids.
The total number of data pages in this table is 99.
dbcc dbcacheremove
Syntax
dbcc dbcacheremove(dbid | dbname)
Version
Added in ASE 12.5.3 and 12.0.0.8 ESD2 under CR 334132
Fix in 12.5.4 ESD3 and 15.0.2 for infected with 11 CR 438354
Usage
Used to remove a database descriptor (DBTABLE) from cache (open databases)
Places the DBTABLE back on the FREE chain
Cannot be used on system databases or user tempdbs
Most commonly addresses 3101 errors trying to load database
Has been on request list for long time
Preference was to fix the cause of wrong dbt_keep value
Must have sa_role
Cannot be granted
dbcc dbcacheremove
Potential Issues
Database may really be in use and cannot be removed from cache
When and Why
Typically for Msg 3101 (database in use) during load db
dbt_keep may be off due to some previous error or failed task cleanup
Previously had to cycle ASE to clear the keep count
Several lines of output indicate usage info and success/failure
Usage and result is also logged in errorlog
dbcc dbcacheremove
dbcc dbcacheremove(test)
go
Attempt to uncache the database 'test' with dbid 4.
The
The
The
The
The
descriptors keep count is 1 and the system tasks keep count is 0.
descriptors hot count is 0 and the detached keep count is 0.
per process keep count for spid 12 is 1.
database is set to current by spid 12.
database 'test' could not be uncached.
dbcc dbcacheremove(pubs2)
go
Attempt to uncache the database 'pubs2' with dbid 5.
The descriptors keep count is 0 and the system tasks keep count is 0.
The descriptors hot count is 0 and the detached keep count is 0.
The database 'pubs2' has been uncached.
dbcc delete_dol_datapage
Syntax
dbcc delete_dol_datapage(dbid, objid, page, [noblock])
Version
Added in ASE 12.5.3 ESD3 and 15.0 ESD2 under CR 383241
Usage
Used to deallocate a corrupt datapage from a DOL table
Requires an EX_TAB lock on the table, unless noblock is used
Must have sa_role
Cannot be granted
Cannot be used with Archive Databases at this time
Use only table scans after this, such as bcp to recover good data
dbcc delete_dol_datapage
Potential Issues
Indexes pointing to this page may cause errors and should be dropped and recreated
Data WILL be lost
When and Why
6xx errors on a page and trying to bcp data out fails
As a last resort, you can deallocate the page so that bcp can run
Example Setup
create table abc_dol (a int, b char(1500)) lock datarows
go
insert abc_dol values(1,'ABC')
go 100
create index x_abc_dol on abc_dol (a)
go
dbcc delete_dol_datapage
declare @vers int, @page int, @dbid int, @obj int
select @dbid = db_id(), @obj = object_id('abc_dol'), @vers = @@version_as_integer
if @vers < 15000
select @page = first+1 from sysindexes where id = @obj and indid = 0
else
select @page = firstpage+1 from syspartitions where id = @obj and indid = 0
dbcc delete_dol_datapage(@dbid,@obj,@page)
Forced deallocation of page on dbid = 4, objid = 784002793, pageid = 890, object
will be consistent for table scans only.
Some info from the page being deallocated in dbid 4: pgno = 890, objid =
784002793, ptnid = 784002793, nextpg = 0, prevpg = 0, timestamp = 0000 00001cad.
select count(*) as "Index Count" from abc_dol
select count(*) as "Table Count" from abc_dol (index 0)
Index Count
----------100
Table Count
----------99
dbcc lock_release
Syntax
dbcc lock_release(spid, level, type, dbid, objid, [page [,row] ] [,force])
Version
Added in ASE 12.5.3 ESD2 and 15.0 ESD2 under CR 374681
Usage
spid must have been badly terminated (P3_PSS_SICK or failed termination)
level is the lock level: table | page | row
type is the lock type and depends on the lock level
ex_tab, sh_tab, ex_int, sh_int
ex_page, up_page, sh_page - page required
ex_row, up_row, sh_row - page, row required
force must be used if the target spid has an active transaction Caution!!
Must have sa_role
Cannot be granted
dbcc lock_release
Potential Issues
Forcing the release of a transactional lock may cause inconsistency
of data
An open transaction may still be a problem for truncating the log
(See dbcc dbreboot at end of presentation)
When and Why
Use to clean up phantom locks not released by a badly terminated
spid
spid may be holding a lock that prevents accessing or dropping
object
dbcc lock_release
dbcc lock_release(12,'table','ex_int',4,880003135)
go
Spid 12 was not badly terminated. Cannot release the given lock.
dbcc lock_release(12,'table','ex_int',4,880003135)
go
The transactional 'exclusive intent' lock for spid 12, dbid 4, objid 880003135,
pageno 0, rowno 0 cannot be released. Use the 'force' option.
dbcc lock_release(12,'table','ex_int',4,880003135,force)
go
The 'exclusive intent' lock at 0x0000000200e3ddb0 for spid 12, dbid 4, objid
880003135, pageno 0, rowno 0 was forcibly released.
dbcc logprint/printolog/traceprint
Syntax
dbcc logprint(multiple lines,can be printed)
dbcc printolog(only one string allowed)
dbcc traceprint(destination depends,on traceflags 3604,3605,3623)
Version
Added prior to 11.9
Usage
logprint requires sybase_ts_role
printolog requires sa_role or ha_role
traceprint does not require specific role
Does not include the errorlog prefix info:
00:00000:00021:2007/06/24 19:37:40.83 server
dbcc logprint/printolog/traceprint
Potential Issues
May affect errorlog monitoring applications
When and Why
Use in a script to output progress information or
debugging info
Can be used to trigger errorlog monitors to page
someone, or take action
Use to record when certain procs are run, such as
threshold procs, etc.
dbcc logprint/printolog/traceprint
dbcc logprint('This is a test', 'for the TechWave', 'Presentation')
go
Errorlog output:
00:00000:00011:2007/07/16 22:31:26.72 server
00:00000:00011:2007/07/16 22:31:26.72 server
00:00000:00011:2007/07/16 22:31:26.72 server
This is a test
for the TechWave
Presentation
dbcc printolog('This is for the TechWave Presentation')
go
Errorlog output:
00:00000:00011:2007/07/16 22:34:43.51 server
This is for the TechWave Presentation
dbcc logprint/printolog/traceprint
dbcc traceoff(3605)
dbcc traceprint('This is a test', 'for the TechWave', 'Presentation')
go
ASE Console output (does not go in errorlog):
This is a test
for the TechWave
Presentation
dbcc traceon(3605)
dbcc traceprint('This is a test', 'for the TechWave', 'Presentation')
go
Errorlog output:
This is a test
for the TechWave
Presentation
dbcc logprint/printolog/traceprint
dbcc traceoff(3605) -- must be off, otherwise 3623 ignored
dbcc traceon(3623)
dbcc traceprint('This is a test', 'for the TechWave', 'Presentation')
go
Errorlog output:
00:00000:00011:2007/07/16 22:42:14.06 server
00:00000:00011:2007/07/16 22:42:14.06 server
00:00000:00011:2007/07/16 22:42:14.06 server
This is a test
for the TechWave
Presentation
dbcc traceon(3604,3623) -- also sends to isql session, but no prefix info
dbcc traceprint('This is a test', 'for the TechWave', 'Presentation')
go
Errorlog output:
00:00000:00011:2007/07/16 22:42:32.20 server
00:00000:00011:2007/07/16 22:42:32.20 server
00:00000:00011:2007/07/16 22:42:32.20 server
This is a test
for the TechWave
Presentation
dbcc memdump
Syntax
dbcc memdump([optional string])
Version
Added in ASE 12.5.1 under CR 323061
Usage
Used to capture a shared memory dump
Required setup
sp_shmdumpconfig add,dbcc,null,10,<dir>
sp_configure dump on conditions,1
see also sp_configure number of dump threads
<some activity>
Execution
dbcc memdump(Test Point 1)
Must have sa_role to execute
Cannot be granted
dbcc memdump
Potential Issues
Engines are halted by default while the memdump is captured this is best
When and Why
Can be used when no other way to trigger memdump (no error or message)
Tech Support may request memdumps when diagnosing issues related to
performance, proc cache fragmentation, or other
Memdump analysis requires access to source code
The optional text is recorded in the errorlog and can be used to identify the
test case, or test progress
dbcc memdump
dbcc memdump(Start test)
exec myproc1
exec myproc2
dbcc memdump(After proc2)
exec myproc3
exec myproc4
dbcc memdump(End of test)
go
dbcc proc_cache
Syntax
dbcc proc_cache(free_unused | flush_elc)
Version
Added in 12.5.4 ESD5 and 15.0.2 under CR 446389
Usage
free_unused frees unused procedures from cache
flush_elc flushes pages from Engine Local Cache to global pool
Should be used with care and when system is not heavily used
Must have sa_role or sybase_ts_role to execute
Cannot be granted
dbcc proc_cache
Potential Issues
Procs flushed from cache may need to be read from disk and recompiled
Removing pages from ELC may increase spinlock contention on proc cache
global pool
When and Why
Use to clean and defragment proc cache without recycling ASE
Try this if you receive a 701 error
Pages in ELC may not be used due to multi-page allocs
free_unused can be used when changing optimization goals/settings
dbcc proc_cache
dbcc proc_cache(free_unused)
go
Procedures in cache before free unused:
Procedures in cache after free unused:
dbcc proc_cache(flush_elc)
go
erlocal_count before flush: eng0:
erlocal_count after flush: eng0:
1571
0
15
0
dbcc proc_cacherm
Syntax
dbcc proc_cacherm(type, dbname, objname)
Version
Added prior to 11.9
Usage
type is V,P,T,R,D,C,F, or S (must be uppercase)
corresponds to View, Proc, Trigger, Rule, Default, Cursor, SQLJ Function,
SQL function
Must have sybase_ts_role to execute
Cannot be granted
dbcc proc_cacherm
Potential Issues
Objects removed from cache may need to be read from
disk and recompiled
When and Why
Remove objects from procedure cache
May help with some 3702 errors
3702: Cannot drop the <objtype> <objname> because it is currently in use
Use monCachedProcedures to get a list of objects
currently in proc cache
dbcc proc_cacherm
select
case ObjectType
when 'stored procedure'
then 'dbcc proc_cacherm(P,' + DBName + ',' + ObjectName + ')'
when 'trigger procedure'
then 'dbcc proc_cacherm(T,' + DBName + ',' + ObjectName + ')'
when 'view'
then 'dbcc proc_cacherm(V,' + DBName + ',' + ObjectName + ')'
when 'default value spec'
then 'dbcc proc_cacherm(D,' + DBName + ',' + ObjectName + ')'
when 'rule'
then 'dbcc proc_cacherm(R,' + DBName + ',' + ObjectName + ')'
end
from master..monCachedProcedures
go
dbcc showrecovery
Syntax
dbcc showrecovery[(dbid | dbname)]
Version
Added in 12.5.3 under CR 353375, enhanced in 12.5.4 ESD7 under CR 370410
Usage
If dbid or dbname is specified, the database must be in recovery
boot time, LOAD DATABASE, LOAD TRAN, ONLINE DATABASE, or MOUNT DATABASE
If omitted, information is shown for all databases in recovery
Must have sa_role to execute
Cannot be granted
dbcc showrecovery
Potential Issues
None
When and Why
Database recovery occasionally takes a long time
One customer scenario was over 30 hours
Use this command to display information about the recovery process
and help determine why it is running long
dbcc showrecovery
dbcc showrecovery
go
No databases are being recovered.
dbcc showrecovery
go
RECOVERY STATISTICS FOR DATABASE 6
General:
Spid: 12
Command: LOAD DATABASE
Type: LOAD DATABASE
Pass: Initialize
(lots of output)
dbcc serverlimits
Syntax
dbcc serverlimits[(all)]
Version
Added in 12.5 under CR 214917, enhanced in 15.0 under CR 360050 for long idents
Usage
Some people familiar with the command, but not the all option
If all specified, additional output is shown with more internal information
Anyone can execute, however, output goes to stdout where ASE was started
Use traceflag 3604 to display on screen requires sa_role
dbcc serverlimits
Potential Issues
None
When and Why
Identify various server limits, such as
max columns in a table
max indexes on a table
max length of table, column, proc names, etc in 15.x
Limits based on 2K, 4K, 8K, 16K pagesize, such as
max size of row
max rows per page
dbcc serverlimits
Use isql w120 option
dbcc serverlimits(all)
go
Max number of variable length columns in a system table
(MAX_VAR_SYSTABS)
: 81
Max number of times a statement in stored procedure can be recompiled
(MAX_NUMBER_OF_RECOMPILATIONS): 10
Max length of a Sybase version string
(SYBASE_MAX_VERSION_LEN)
: 100
dbcc tune(des_bind)
Syntax
dbcc tune(des_bind, dbid, objid | objname)
Version
Added in 11.9.1 under CR 139739
Usage
Must specify dbid, not dbname
Marks the DES as hot, meaning it wont be scavenged and is guaranteed to
be in memory
Does not survive ASE recycle
Must have sa_role to execute
Can be granted access to all tune options
dbcc tune(des_bind)
Potential Issues
A DES is used and cannot be scavenged
Object cannot be dropped until unbound
When and Why
Relieve pressure on the DES spinlock
Check sp_sysmon Object Manager Spinlock Contention
Code can be skipped since we are guaranteed to have an active DES for this
object
object can be table, proc, default, rule, view, trigger
dbcc tune(des_bind)
dbcc tune(des_bind,4,abc_dol)
go
dbcc cacheremove(test,abc_dol)
go
Some one else is using this object. DES not Removed
drop table abc_dol
go
Msg 3702, Level 16, State 1:
Server 'acid1502', Line 1:
Cannot drop the table 'abc_dol' because it is currently in use.
dbcc tune(hotdes)
Syntax
dbcc tune(hotdes, dbid)
Version
Added in 11.9.1 under CR 139739
Usage
Must specify dbid, not dbname
Must have sa_role to execute
Can be granted access to all tune options
dbcc tune(hotdes)
Potential Issues
None
When and Why
Displays DES information for objects currently
marked as hot within the specified database
dbcc tune(hotdes)
dbcc tune(hotdes,4)
go
DES chain (bind state) for all objects in dbid 4:
Current database (4) test:
DES at 0x20158a8e0
ddbid=4
OBJECT structure at: 0x20158a90c
objname=abc_dol dol_rowno=14 dol_vattcnt=7 dol_status: 0x04:
(0x0004(DOL_ROW_POSSUNCMT))
dobjstat.objid=976003477 dobjstat.objuserstat=0x0
dobjstat.objsysstat=0x43 3 (O_USER) (0x0040 (O_LOGGED))
dobjstat.objsysstat2=0x18200 (0x00010000 (O2_ELEVEN90_OBJECT),
0x00008000
(O2_LOCK_DATAROWS), 0x00000200 (O2_NOVARCOL))
dbcc tune(des_unbind)
Syntax
dbcc tune(des_unbind, dbid, objid | objname)
Version
Added in 11.9.1 under CR 139739
Usage
Must specify dbid, not dbname
Removes the hot status from the DES and allows the descriptor to be freed
(see cacheremove) / reused or the object to be dropped
Must have sa_role to execute
Can be granted access to all tune options
dbcc tune(des_unbind)
Potential Issues
DES for this object can be reused
When and Why
Remove hot status from an object previously marked
hot with des_bind
Must be done before dropping object
dbcc tune(des_unbind)
dbcc tune(des_unbind,4,abc_dol)
go
dbcc cacheremove(test,abc_dol)
go
drop table abc_dol
go
dbcc istraceon
Syntax
dbcc istraceon(<traceflag>)
Version
Added prior to 11.9
Usage
Used to determine if a traceflag is on
Can be done in batch code
New built-in, switchprop, added in 12.5.4 ESD4 and 15.0.2 (0 is off)
Anyone can use
dbcc istraceon
Potential Issues
@@error variable is set (0 = on, -1 = off)
When and Why
Use in a script to see if a particular traceflag is set
dbcc istraceon
dbcc traceoff(3604)
go
dbcc istraceon(3604)
select case @@error
when 0 then 'traceflag is on
when -1 then 'traceflag is off
end
go
---------------traceflag is off
dbcc dbreboot
Syntax
dbcc dbreboot(<cmd>,<db1> [,dbn])
Version
Added in 12.5.4 ESD1 and 15.0.2 under CR 368529
Usage
<cmd> can be report, reboot, reboot_norecovery, shutdown,
shutdown_load, restart, restart_norecovery
Can specify up to 16 databases
dbcc dbreboot
Potential Issues
Running tasks will be terminated
When and Why
Database may be incorrectly marked as in use
spid received 3475 error, or 12583
3475: There is no space available in SYSLOGS to log a record for which
space has been reserved.
12583: Unable to log a record due to internal error %d.
Both errors will retry at 1 minute intervals
If need to add more space, use reboot_norecovery option
Task failed during termination processing phantom spids
dbcc dbreboot
Options
report Shows a report on the specified database(s).
reboot Database is shutdown and restarted with recovery.
reboot_norecovery Database is shutdown and restarted and left in
a not recovered state. Can be used to add more space.
shutdown Database is shutdown and left in an unusable state.
shutdown_load Database is shutdown and left in a state that allows
LOAD DATABASE to be done.
restart Restart and recover a database that has been shutdown.
restart_norecovery Restart database and left in a not recovered
state. Good when there are problems with recovery.
dbcc dbreboot
drop database tb2
go
Msg 3702, Level 16, State 4:
Server 'acid1502', Line 1:
Cannot drop the database 'tb2' because it is currently in use.
dbcc dbreboot(report,tb2)
go
---------- Active Processes and Transactions in Database 'tb2' ---------Spid Program
Transaction
Status
CPU
I/O
Blk Error
---------- Operation on Database 'tb2' Completed Successfully ----------
dbcc dbreboot
dbcc dbreboot(reboot,tb2)
go
---------- Shutting Down Database 'tb2' ------------------- Re-starting Database 'tb2' With Recovery ---------Recovering database 'tb2'.
Started estimating recovery log boundaries for database 'tb2'.
Database 'tb2', checkpoint=(1051, 9), first=(1051, 8), last=(1051, 9).
Completed estimating recovery log boundaries for database 'tb2'.
Started ANALYSIS pass for database 'tb2'.
Completed ANALYSIS pass for database 'tb2'.
Started REDO pass for database 'tb2'. The total number of log records to process is 2.
Completed REDO pass for database 'tb2'.
Recovery of database 'tb2' will undo incomplete nested top actions.
Started UNDO pass for database 'tb2'. The total number of log records to process is 2.
Undo pass of recovery has processed 1 incomplete transactions.
Completed UNDO pass for database 'tb2'.
Started filling free space info for database 'tb2'.
Completed filling free space info for database 'tb2'.
Started cleaning up the default data cache for database 'tb2'.
Completed cleaning up the default data cache for database 'tb2'.
Recovery complete.
Database 'tb2' is now online.
---------- Operation on Database 'tb2' Completed Successfully ---------drop database tb2
go
DBCC Secrets
QUESTIONS?
