Database Maintenance - PowerPoint PPT Presentation

About This Presentation
Title:

Database Maintenance

Description:

JSOC Review 17 March 2005. Overview. Q: ... Plan on weekly base backup ... Updating planner statistics. Preventing transaction ID wraparound failures. Obstacle ... – PowerPoint PPT presentation

Number of Views:603
Avg rating:3.0/5.0
Slides: 8
Provided by: franki150
Learn more at: http://hmi.stanford.edu
Category:

less

Transcript and Presenter's Notes

Title: Database Maintenance


1
Database Maintenance
  • Karen Tian
  • ktian_at_Stanford.EDU

2
Overview
  • Q How long can we keep running?
  • Our goal is to minimize downtime
  • Do maintenance while running
  • Challenges
  • Very large database a few TB
  • Transaction intensive applications
  • Currently we have SUMS and DRMS tables in the
    same DB
  • SUMS tables insert, update, and delete
  • Must vacuum to reclaim disk space
  • DRMS tables tables insert only, no update or
    delete, except for transient records
  • Static tables, need to vacuum to prevent XID
    wraparound

3
Routine database maintenance tasks
  • Base backup and WAL file backup
  • Vacuum
  • Reindex
  • PostgreSQL upgrade

4
Base backup and WAL file backup
  • Base backup
  • pg_start_backup()
  • file system snapshot
  • pg_stop_backup()
  • snapshot dump to tape
  • Plan on weekly base backup
  • The longer the interval between base backups, the
    longer it takes to process xlog to catch up
  • These base backup along with appropriate WAL
    files can make a consistent DB snapshot, perhaps
    useful for other purpose.
  • Archive WAL files
  • Clean up WAL files after a base backup

5
Vacuum
  • Purpose
  • Recovering disk space
  • Updating planner statistics
  • Preventing transaction ID wraparound failures
  • Obstacle
  • Long running transaction prevents vacuum from
    deleting dead rows
  • Currently SUMS and DRMS tables reside in the
    same DB. DRMS module may start long running
    transaction that prevents vacuum from removed
    dead rows in SUMS tables.
  • Plans
  • Switch to warm stand-by and perform vacuum on the
    primary
  • Separate SUMS from DRMS, i.e., run them off
    different PostgreSQL server instances.

6
Reindex
  • Why?
  • Index might become corrupted
  • Index might become bloated
  • A freshly-constructed index is faster because
    logically adjacent pages are usually also
    physically adjacent in a newly built index
  • Impact
  • Locks out writes but not reads of the index's
    parent table
  • Takes an exclusive lock on the specific index
    being processed, which will block reads that
    attempt to use that index

7
PostgreSQL upgrade
  • The internal data storage format is subject to
    change between major releases
  • Require backing up data and restore it on the new
    server
  • Must take the DB server down and switch to
    stand-by
Write a Comment
User Comments (0)
About PowerShow.com