Database Infrastructure SOP

From FedoraProject

Revision as of 14:13, 24 May 2008 by ImportUser (Talk)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Contents

Database - SOP

Contact Information

Owner: Fedora Infrastructure Team

Contact: #fedora-admin, sysadmin-main, sysadmin-dba group

Location: Phoenix

Servers: db1, db2

Purpose: Provides database connection to many of our apps.

Description

db1 and db2 are our primary database servers. db1 contains the MySQL instance, db2 contains postgresql. Each database server replicates to itself and the other through a dump style backup. In a normal situation, db1 runs only MySQL, not postgresql. While db2 runs only postgresql not MySQL. Which is running on which is defined in the puppet configs, specifically the node manifest for each server (nodes/db1.fedora.phx.redhat.com.pp and nodes/db2.fedora.phx.redhat.com.pp)

Troubleshooting and Resolution

Connection issues

There are no known outstanding issues with the database itself. Remember that every time either database is restarted, services will have to be restarted (see below).

Some useful queries

What queries are running

This can help you find out what queries are cuurently running on the server::

select datname, procpid, query_start, backend_start, current_query from
pg_stat_activity where current_query != '<IDLE>' order by query_start;

This can help you find how many connections to the db server are for each individual database::

select datname, count(datname) from pg_stat_activity group by datname
order by count desc;

Seeing how "dirty" a table is

We've added a function from postgres's contrib directory to tell how dirty a table is. By dirty we mean, how many tuples are active, how many have been marked as having old data (and therefore "dead") and how much free space is allocated to the table but not used.

\c fas2
\x
select * from pgstattuple('visit_identity');
table_len          | 425984
tuple_count        | 580
tuple_len          | 46977
tuple_percent      | 11.03
dead_tuple_count   | 68
dead_tuple_len     | 5508
dead_tuple_percent | 1.29
free_space         | 352420
free_percent       | 82.73
\x

Vacuum should clear out dead_tuples. Only a vacuum full, which will lock the table and therefore should be avoided, will clear out free space.

XID Wraparound

Find out how close we are to having to perform a vacuum of a database (as opposed to individual tables of the db). We should schedule a vacuum when about 50% of the transaction ids have been used (approximately 530,000,000 xids):

select datname, age(datfrozenxid), pow(2, 31) - age(datfrozenxid) as xids_remaining
from pg_database order by xids_remaining;

Information on wraparound

Restart Procedure

If the database server needs to be restarted it should come back on it's own. Otherwise each service on it can be restarted:

service mysqld restart
service postgresql restart

Koji

Any time postgreql is restarted, koji needs to be restarted. Please also see Restarting Koji

Mirror Manager

Anytime postgresql is restarted Mirror Manager will need to be restarted, no SOP currently exists for this

Bodhi

Anytime postgresql is restarted Bodhi will need to be restarted no sop currently exists for this.

Smolt

Anytime MySQL is restarted, Smolt will need to be restarted. no SOP currently exists for this

Note about TurboGears and MySQL

{{ Template:message/note | There's a known bug in !TurboGears that causes MySQL clients not to automatically reconnect when lost. Typically a restart of the !TurboGears application will correct this issue.