PG_REORG Utility for VACUUM FULL online

pg_reorg is a utility made by NTT for reorganizing the table structure.

Concept is simple, if you have all the require pointers and data in same page, then accessing those is much more faster. This is what pg_reorg provides to a user.

Following are some options, which pg_reorg provides.
-o [ —order-by] columns:
This option makes pg_reorg to oraganise the table data as per the mentioned column. At the backend pg_reorg will creates a new table using CTAS and SELECT Query include ORDER BY clause with columns mentioned with -o.
-n [—no-order] tablename:
When this option is being used, then pg_reorg, does the VACUUM FULL ONLINE. Now, question is how it must be doing. Simple Concept, create a new table using CTAS and create a trigger on current table to track the DML. As the New table got created play those tracked DML on new table. It works well. This option is only for table which has primary key.

pg_reorg by default does the CLUSTER of tables and it follows same concept, i.e without locking table do the CLUSTER.

After performing all the options, pg_reorg does the ANALYZE on the table.

Following are some elog information, which it performs at backend:
elog(DEBUG2, "---- reorg_one_table ----");
elog(DEBUG2, "target_name    : %s", table->target_name);
elog(DEBUG2, "target_oid     : %u", table->target_oid);
elog(DEBUG2, "target_toast   : %u", table->target_toast);
elog(DEBUG2, "target_tidx    : %u", table->target_tidx);
elog(DEBUG2, "pkid           : %u", table->pkid);
elog(DEBUG2, "ckid           : %u", table->ckid);
elog(DEBUG2, "create_pktype  : %s", table->create_pktype);
elog(DEBUG2, "create_log     : %s", table->create_log);
elog(DEBUG2, "create_trigger : %s", table->create_trigger);
elog(DEBUG2, "create_table   : %s", table->create_table);
elog(DEBUG2, "delete_log     : %s", table->delete_log);
elog(DEBUG2, "lock_table     : %s", table->lock_table);
elog(DEBUG2, "sql_peek       : %s", table->sql_peek);
elog(DEBUG2, "sql_insert     : %s", table->sql_insert);
elog(DEBUG2, "sql_delete     : %s", table->sql_delete);
elog(DEBUG2, "sql_update     : %s", table->sql_update);
elog(DEBUG2, "sql_pop        : %s", table->sql_pop);
Interesting Hunh.

Some Comparison between clusterdb and pg_reorg is given in below link:
http://reorg.projects.postgresql.org/index.html

With everything, DBA has to take care of few things:
While pg_reorg is going on one table, the DBA should not let anyone to execute DDL Changes and Index on the currently reorganizing table.

Comments

Post a Comment

Popular posts from this blog

xDB Replication from Oracle to PPAS

EnterpriseDB Replication tool has been now modified and add new functionality in it. Previous Replication Console, which had been made, was using the DBA Management server and Publication & Subscriptions were dependent. Now, xDB Replication has been divided into two components: 1. Publication Server : (Master Server) 2. Subscription Server: (Slave Server) Some Background of Publication Server. Publication Server has following components: 1. JAVA Based Daemon. 2. MetaDatabase of Publication Server. Publication Server keeps its information (Information about the Primary Database, Tables Details etc.) in the MetaDatabase under schema _edb_replicator_pub Publication server is independent and has no dependency on Subscriptions Server. By default publication server uses the 9011 port. However, if user wants, [S]he can use different ports too. For running Publication Server on different port, use the following command: $EDBHOME/jre/bin/java -Djava.awt.headless=true -jar $...
Read more

Does UPDATE Change ROWID in Oracle?

Yesterday, there was a discussion which was going on OTN (Oracle Forum). I was also part of that discussion. Discussion was about the "Does UPDATE statement change rowid in Oracle?" As per my comment of "UPDATE does not change rowid", one of the user has pointed me that UPDATE statement sometime changes the rowid in Oracle and given me following example. SQL> ed Wrote file afiedt.buf 1 create table moving_rowid ( 2 col1 number primary key, 3 col2 number, 4 col3 number, 5 col4 varchar2(10) 6 ) 7* organization index SQL> / Table created. SQL> insert into moving_rowid values( 1, 2, 3, 'foo' ); 1 row created. SQL> insert into moving_rowid values( 2, 3, 4, 'bar' ); 1 row created. SQL> select rowid, col1, col2, col3, col4 2 from moving_rowid; ROWID COL1 COL2 COL3 COL4 ---------------- ---------- ---------- ---------- ---------- *BAEADxsCwQL+ 1 2 ...
Read more

PostgreSQL Database Link to Oracle Database on Linux

I have seen question, like "How to make Database Link from PostgreSQL to Oracle?", always floats in PostgreSQL Community Forum. So, I thought to do some research on it and write a Blog. Cybertec (One of The PostgreSQL Database Company) has released a PostgreSQL Module ODBC Link, using which user can make Database link to any other database, including Oracle, MS SQL Server, PostgreSQL etc, which have ODBC compliant Data source. Lets see how you can make ODBC Connection from PostgreSQL to any ODBC compliant data source and fetch data. Installation of this module is very simple. Following are the steps which user can follow: 1. Install the unixODBC driver on your linux machine. user can use following useful link for Downloading and Installing the unixODBC Driver: http://www.unixodbc.org/ 2. Download ODBC-Link from following location: http://www.cybertec.at/download/odbc_link/ODBC-Link-1.0.4.tar.gz 3. Untar the downloaded file as given below: tar -zxvf ODBC-Link-1.0.4....
Read more