Netezza DBA: June 2014

Friday, June 27, 2014

nzbackup vs nz_backup

There are 2 commands to perform backup of Netezza. One is nzbackup. The other one is nz_backup. For nzbackup, not much you can play with it except you need to specify the database to backup. However, nz_backup provides more command options eg, specific table to backup, include where clause etc. Here's the command options for nz_backup.

Usage: nz_backup -dir <dirname> -format <ascii|binary|gzip> [ optional args ]
-or-
nz_restore -dir <dirname> -format <ascii|binary|gzip> [ optional args ]

Purpose: To backup (or restore) one or more tables.

An nz_backup must be run locally (on the NPS host being backed up).

An nz_restore can be used to restore data into a remote NPS host. Just
include the "-host" switch, or set the NZ_HOST environment variable.

Note: When doing an "nz_restore -format binary -host <xxx>", the script
issues an "nzload -compress true". This nzload feature only exists
in nzload as of NPS 4.6. If you want to do this on an older version
of NPS (4.0 or 4.5) then:
o Install a copy of the 4.6 client toolkit somewhere on your box
(it can be used against the older server releases)
o Add it's bin directory to the start of your search PATH
o Then invoke nz_restore

These scripts can process a single table, multiple tables, or an entire database.

The data format that is used can be either
ascii -- which is very portable.
binary -- which is the database's compressed external table format. This
is much faster, and results in significantly smaller backup sets.
gzip -- ascii, which is gzip'ed on the NPS host.

The data can be written to (or read from) disk files or named pipes.
If you use named pipes, another application is used to consume
(or produce) the data.

These scripts just concern themselves with the DATA itself. When backing up
a table, the DDL is not included. When restoring a table, the script expects
the table to already exist. It will not create it. It will not truncate it
(so if the table currently has any data in it, that data will be left untouched
by this script).

To backup tables requires the following permissions:

GRANT SELECT ON _VT_HOSTTXMGR TO <user/group>;
GRANT SELECT ON _VT_HOSTTX_INVISIBLE TO <user/group>;
GRANT SELECT ON _VT_DBOS_CONNECTION TO <user/group>;
--To obtain information about transactions

GRANT LIST ON <DATABASE|dbname> TO <user/group>;
--The user must have access to the database that contains the tables

GRANT SELECT ON <TABLE|tablename> TO <user/group>;
--The user must have access to the tables themselves, and their data

GRANT CREATE EXTERNAL TABLE TO <user/group>;
--The user must be able to create external tables, into which the
--data will be unloaded

To restore/reload a table requires an additional permission:

GRANT INSERT ON <TABLE|tablename> TO <user/group>;
--The user must be able to insert (i.e., reload) data back into the tables

Options: REQUIRED ARGUMENTS
==================

-dir <dirname> [...]

The full path to the directory in which the data files will be written
to (or read from). This directory must already exist and permit write
access to it. The directory name can be as meaningful as you wish to
make it.

If you are running this script as some linux user other than 'nz', please
note that it will actually be one of the 'nz' processes that writes the
data into this directory. So linux user 'nz' must also have write access
to it. If you are using named pipes (rather than files) then this is the
directory where the named pipes will be created.

Examples:
-dir /backups/backup_set_17
-dir /snap_storage/bu/customer_db/2006_11_18
-dir /tmp

If desired, you may split the backup files up across multiple directories/
file systems. Each thread can be associated with a separate "-dir <dirname>"
by specifying them on the command line. If you use this feature, then the
number of directories specified must match the number of threads.

-format <ascii|binary|gzip>

Identifies the format to be used for the output files.

ascii Universal in nature, but typically results in larger files and
slower performance.

binary The database's compressed external table format.

gzip ascii, which is then compressed (using gzip). By definition,
compressing and decompressing data uses up a lot of CPU cycles (i.e.,
it takes a long time). When using the binary format (compressed/
external), the work is done in parallel across all of ths SPUs ... so
it is very quick. But this option uses the NPS host to gzip/gunzip the
data. You will (almost always) want to use multiple threads in order
to get more of the host's SMP processors involved in order to speed
things along. The sweet spot seems to be about 8 threads, though you
can certainly use a larger/smaller number if you want to break the
backup files up into more/fewer pieces.

best --to-- least
====== ====== ======
Speed: binary ascii gzip
Size: gzip binary ascii
Universality: ascii gzip binary

OPTIONAL ARGUMENTS
==================

-h or -? or --help display this help

-[rR]ev or -V print the software revision of this program

-host <name/IP> host name or IP address [NZ_HOST]
-u <user> user name [NZ_USER]
-pw <string> password [NZ_PASSWORD]
-db <database> database name [NZ_DATABASE]

-t <tablename> [...]

# Table(s) within the database to be processed. If none are specified,
# then all tables in the database will be processed.
#
# If you have a file that contains a list of tablenames to be backed up,
# (separated by spaces and/or newlines) it can be used via the following
# syntax:
#
# -t `cat /tmp/the_list_of_tables`

-threads <nnn>

# Breaking the backup into multiple threads (per table) can increase the
# overall thruput, especially for large tables. This will also have the
# effect of creating smaller backup files, since each one will now be
# broken up into multiple pieces.
#
# By default, only a single thread will be used. You can specify a number
# from 1..31. Whatever value you specify for the backup must also be used
# for the restore. In general, the sweet spot seems to be about 6 threads.

-script <scriptname>

# Instead of backing up (or restoring) the data from disk files, you can use
# named pipes -- allowing another application to consume (or produce) the data
# on the fly. To use named pipes, specify that application/script here. The
# script will be automatically invoked and passed (as arg 1) the fully rooted
# pathname of the named pipe that is it supposed to use.
#
# For example scripts, see the file(s)
# nz_backup.script_example
# nz_restore.script_example

-whereclause <clause>

# Normally, nz_backup is used to backup the entire contents of a table --
# all visible rows. This option allows "you" to tack on a WHERE clause
# to the data that gets selected ... allowing "you" to backup a subset
# of the table. All the power (and responsibility) is put into your
# hands. Do wrap the clause in double quotes so it will be passed into the
# script correctly. Examples:
#
# -whereclause "customer_key = 2"
# -whereclause "customer_key in (1,3,5)"
# -whereclause "region_name = 'AMERICA' or region_key = 0"
# -whereClause "order_date between '1998-01-01' and '1998-12-31'"
#
# Because this clause gets applied to all tables being backed up, you would
# probably only want to backup a single table at a time (when using this
# clause) ... since the clause will typically contain column names that are
# specific to that table.
#
# This clause only applies to backups (not restores). Its use will be logged
# in the output of this script (as well as in the pg.log file).

-ignoreTxID

# Tables are individually backed up, one at a time. Since a backup may span
# many hours, this script insures that the backup represents a consistent
# point-in-time by using the transaction IDs attached to each row.
#
# This switch will override that feature ... and backs up each table with
# whatever data it contains when the backup (of that particular table) is
# kicked off. This insures that your backup will include ALL of the data
# in a table that has been committed (so you're not susceptible to long
# running or stale transactions).
#
# This switch is primarily of use with NPS 4.0. In later releases, this
# script is able to do things differently.

-dropdata

# This is for testing purposes only. As the name implies, the backup will
# be written to /dev/null, resulting in no backup at all. This is useful
# for testing the performance of the NPS components that are involved
# (SPUs/S-Blades ==> Host), while excluding the speed/overhead of your host
# storage.

-sizedata

# This is for testing purposes only. Like "-dropdata", but rather than
# sending the backup data directly to /dev/null it will first be piped
# thru "wc -c" in order to count the number of bytes in the backup
# stream (e.g., to provide you with actual sizing information). So it
# has the performance characteristics of "-dropdata" ... but provides
# you additional information.
#
# You can use "-format <ascii|binary>" and 1 or multiple "-threads <nn>"
# when using this switch.
#
# Each table will include the following pieces of information
#
# Info: source table size 80,740,352
# Info: backup file size 76,574,691
#
# And summary lines for the entire backup set will display the
#
# TOTAL source table size : 48,267,526,144
# TOTAL backup file size : 46,088,648,210
#
# The table size is whatever the table size is -- the amount of storage
# space it using on disk (as reported by nz_db_size or nz_tables). This
# script doesn't know/care if the data is compressed on disk (e.g, CTA0/1/2).
# Nor does the script know if there are any logically deleted rows in the
# table (taking up space in the table, but which would not be part of a
# backup data set).
#
# The backup size is the amount of storage that would be required if the
# backup data set was actually written to disk. This would represent either
# the ascii or the binary (compressed external table format) version of the
# data ... whatever you chose.
#
# To get sizing information for a full nzbackup, e.g.
# nzbackup -db DBNAME -dir /tmp
# you would use a command line such as this
# nz_backup -db DBNAME -dir /tmp -format binary -sizedata

Outputs: Status/log/timing information will be sent to standard out ... and will
include information about any ERROR's that might be encountered.

Exit status: 0 = success, non-0 = ERROR's were encountered

Examples: $ nz_backup -format binary -dir /backupdir

$ nz_restore -format ascii -dir /tmp -db my_db -t table1 -script /tmp/my_script

Comparison: nzbackup/nzrestore nz_backup/nz_restore
================== ====================
NPS CLI Utility Add-on Script
Backup Type
Full X X
-differential X
-cumulative X

Granularity
Entire Database X X
Individual Table(s) nzrestore X

Formats Supported
Compressed Binary X X
Ascii X
Ascii gzip'ed X

Output Destination
Veritas (NPS 4.0) X
Tivoli (NPS 4.6) X
Disk X X
Named Pipes X

Multi-Stream Support nzbackup (6.0) X

DDL included as part X Use the nz_ddl* scripts
of the backup set

Friday, June 13, 2014

Netezza Versioned Tables

Versioned tables come about as a result of doing an
ALTER TABLE <tablename> [ADD|DROP] COLUMN ...

This results in multiple data stores for the table. When you go to query
the table, NPS must recombine the separate data stores back into a single
entity. This action will be performed automatically and on-the-fly. But
it does result in additional query overhead. Therefore, it is a best
practice to reconstitute the table as soon as practical by doing a

GROOM TABLE <tablename> VERSIONS;

Notes: The maximum number of table versions allowed is 4 (which means you
can perform at most three ALTER TABLE commands before doing a GROOM,
since this number includes the original table version itself)

SELECT the_table.database as "Database",
MAX
( CASE WHEN the_table.objclass = 4910 THEN SUBSTR(the_table.objname, 3)::nvarchar(128)
WHEN the_table.objclass in (4951,4959,4963) THEN NULL
ELSE the_table.objname::nvarchar(128)
END
) AS "Table Name",
TO_CHAR ( nvl(SUM ( used_bytes ),0), '999,999,999,999,999,999' ) AS "Size (Bytes)",
count(distinct(case when objclass in (4959,4963) then objid else null end)) as "# Of Versions"
FROM _V_OBJ_RELATION_XDB the_table
left outer join _V_SYS_OBJECT_DSLICE_INFO on
(
the_table.objid = _V_SYS_OBJECT_DSLICE_INFO.tblid and
_V_SYS_OBJECT_DSLICE_INFO.tblid > 200000
)
WHERE
the_table.objclass in (4905,4910,4940,4959,4951,4953,4961,4963)
and the_table.objid > 200000
GROUP BY
"Database",
the_table.visibleid
HAVING "# Of Versions" >= 2
ORDER BY 1, 2;

Wednesday, June 11, 2014

Check if statistics out of date

You can use query below to check if statistics are outdated. From nzsql, connect to a database in particular and run the query below:

nzsql -d database <<eof

SELECT
"Table Name",
the_label as "Statistics State",
count(*) as "# Of Columns",
TO_CHAR ( _v_table.reltuples, '999,999,999,999,999') AS "Table Rowcount"
FROM
(
SELECT
_v_relation_column.name as "Table Name",
CASE WHEN _v_statistic.recent IS NULL THEN 'not maintained '
ELSE
CASE WHEN _v_statistic.recent = 85 THEN 'Full '
WHEN _v_statistic.recent = 341 THEN 'Express '
WHEN _v_statistic.recent = 149 THEN 'Basic '
WHEN _v_statistic.recent = 1 THEN 'Full Min/Max OK '
WHEN _v_statistic.recent = 257 THEN 'Express Min/Max OK'
WHEN _v_statistic.recent = 129 THEN 'Basic Min/Max OK '
WHEN _v_statistic.recent = 0 THEN 'Full Outdated '
WHEN _v_statistic.recent = 256 THEN 'Express Outdated '
WHEN _v_statistic.recent = 128 THEN 'Basic Outdated '
WHEN _v_statistic.recent = 169 THEN 'Min/Max Only '
WHEN _v_statistic.recent = 170 THEN 'Unavailable '
WHEN _v_statistic.recent = 16554 THEN 'Unavailable '
ELSE _v_statistic.recent || ''
END
END as the_label,
_v_relation_column.objid
FROM
_v_relation_column
left outer join _v_statistic on
( _v_relation_column.name = _v_statistic.tablename AND
_v_relation_column.attnum = _v_statistic.attnum
)
WHERE
_v_relation_column.objid > 200000
AND upper(_v_relation_column.type) in ('TABLE','SECURE TABLE')
) the_subquery
LEFT OUTER JOIN _v_table ON the_subquery."Table Name" = _v_table.tablename
WHERE trim(the_label) NOT IN ('not maintained','Full','Express','Basic','Unavailable')

GROUP BY "Table Name", "Statistics State", "Table Rowcount"
ORDER BY "Table Name", "Statistics State";

eof

Tuesday, June 10, 2014

unload and load data using External Table

There a few ways you can transfer Netezza data across another Netezza database. One way is to use nz_migrate. Another way through Aginity Tools -> Import and the third way is through External Table in Aginity. Note: You can only use this method from Aginity

Below example, dumps the data from the query to your local PC temp directory while connecting to the source database

UNLOAD DATA

CREATE EXTERNAL TABLE ext_db 'c:\temp\ext_db.dat'

USING

(

remotesource 'odbc'

delimiter '|'

ignorezero false

ctrlchars true

escapechar '\'

logDir 'c:\temp'

boolStyle 'T_F'

encoding 'internal'

nullValue 'N'

maxErrors 1

)

SELECT database, owner, cast(createdate as date) as ts from _v_database;

While connecting your Aginity to your target NPS host database, this method can be used to load the data

to your say Development environment.

LOAD DATA (in different NPS host database)

create external table abc

(

DATABASE NATIONAL CHARACTER VARYING(255),

OWNER NATIONAL CHARACTER VARYING(255),

TS DATE

)

using

(

dataobject ('c:\temp\ext_db.dat')

Remotesource 'odbc'

Encoding 'INTERNAL'

EscapeChar '\'

CrInString true

CtrlChars true

Delimiter '|'

NullValue 'NULL'

)

;

Thursday, June 5, 2014

Netezza Groups

There two types of groups in Netezza. One type is the typical permission, privileges on specific objects in the NPS. Another group is called Resource Sharing Group (RSG) for workload management. You can use this group to define resource utilisation percentages from total NPS resources.

Object Permission Group

CREATE GROUP CELEBI_USR_PETL_ST_GRP;

Or, to create group and add user(s)

CREATE GROUP FIRST_GROUP WITH USER1,USER2;

or add user in group using alter

ALTER GROUP FIRST_GROUP ADD USER USER1;

or drop user from group

ALTER GROUP FIRST_GROUP DROP USER USER1;

RSG Group

CREATE GROUP EBI_ENR_USER WITH
SESSIONTIMEOUT 30
QUERYTIMEOUT 300
DEFPRIORITY NORMAL
MAXPRIORITY NORMAL
RESOURCE MINIMUM 10
RESOURCE MAXIMUM 70
MAXIMUM 30 ;

How to lock Netezza userid

How do you lock Netezza userid? You actually cannot find any way from NZ Admin tool or Aginity or even from the portal. Then, how do you lock a Netezza userid? Below command will show you:

SYSTEM(ADMIN)=> update _t_user set uselocked=true where usename='TEST1';
UPDATE 1

If you want to unlock, you can use NZ Admin tool.

Wednesday, June 4, 2014

find who owns what objects

How to you find out who owns what in the Netezza databases. You need to know this because sometimes you want to drop a particular object, NPS returns error like below:

DROP USER: user "BIADM" owns objects, cannot be removed

So how do you find out which objects belongs to BIADM?

Here's how:

SELECT OWNER, THE_OBJECT_TYPE, OBJNAME

FROM (

SELECT

usename AS OWNER,

(CASE WHEN objclass = 4902 then 'database'

WHEN objclass = 4903 then 'group'

WHEN objclass = 4904 then 'user' END ) AS the_object_type,

objname

FROM

_t_object,

_t_user

WHERE objowner = usesysid

AND objclass IN (4902, 4903, 4904)

AND objdb = 0

AND objowner <> (SELECT usesysid FROM _t_user WHERE UPPER(usename) = 'ADMIN')

UNION ALL

SELECT

_t_user.usename AS OWNER,

lower(the_class.objname) AS the_object_type,

the_object.objname || ' in database ' || the_database.objname AS objname

FROM

_t_object as the_object

LEFT OUTER JOIN _t_object the_class ON

(the_object.objclass = the_class.objid),

_t_object the_database,

_t_user

WHERE the_object.objowner = _t_user.usesysid

AND the_object.objclass NOT IN (4902, 4903, 4904)

AND the_object.objdb = the_database.objid

AND the_object.objowner <> (SELECT usesysid FROM _t_user WHERE UPPER(usename) = 'ADMIN')

) tab

WHERE OWNER='BIADM'

ORDER BY 1,2,3;

Tuesday, June 3, 2014

Netezza Admin / DBA role privileges

In Netezza, there is a default admin user which you can use. But how about creating a second admin user? You can assigned the permissions below.

create group DBA_GROUP;

-- Object permission
grant all on USER to DBA_GROUP;
grant all on MANAGEMENT VIEW to DBA_GROUP;
grant all on MANAGEMENT table to DBA_GROUP;
grant all on SYSTEM VIEW to DBA_GROUP;
grant all on SYSTEM TABLE to DBA_GROUP;
grant all on PROCEDURE to DBA_GROUP;
grant all on LIBRARY to DBA_GROUP;
grant all on AGGREGATE to DBA_GROUP;
grant all on FUNCTION to DBA_GROUP;
grant all on SYNONYM to DBA_GROUP;
grant all on SEQUENCE to DBA_GROUP;
grant all on EXTERNAL TABLE to DBA_GROUP;
grant all on DATABASE to DBA_GROUP;
grant all on GROUP to DBA_GROUP;
grant all on TABLE,VIEW to DBA_GROUP;

-- Admin permission
grant ALL ADMIN to DBA_GROUP;

create user NZADMIN with password 'abc123';

alter group DBA_GROUP add user NZADMIN;

create group DBA_GRA_GROUP with resource minimum 50 resource maximum 100;

alter user NZADMIN in resourcegroup DBA_GRA_GROUP;