Oracle to Postgres Migration - part 2

Workshop
Oracle to Postgres Migration
Part 2 - Running Postgres
2016-06-22 @IDM
Chris Mair
http://www.pgtraining.com

2016-06-22OracletoPostgresMigration-part2
The Workshop
very quick walk through for Postgres-DBAs to-be
• installation, getting support, the configuration
files, psql, understanding transactions, the
query-planner and locking, backups, system
tables, streaming replication, hot standbys,
connection pooling, load balancing and even
automatic failover all with life-demos and
condensed into just three hours - will we finish
on time?

Getting Support
• very good community support through mailing
lists: psql.it list / Italian and ofﬁcial list (English)
and many others
• commercial support - in Italy for example from us
at PGtraining (three free lancers) or 2ndQuadrant
(SRL), in Austria from Cypertec (GmbH) et al
• don't forget managed hosting offerings from
Amazon Web Services (PostgreSQL RDS),
Heroku and others

Installing Postgres
• from your distro (note that the second digit is the
major version 9.0 and 9.5 are ﬁve years apart
and some distros carry outdated versions)
• from the ofﬁcial repos at www.postgresql.org/
download/ - all major package formats supported
• from source (it is easier than you think:
everything can be compiled in a minute or two)

From Source, You Say?
• yeah, why not?
# Centos 7
yum -y install wget
yum -y install gcc make zlib zlib-devel libxml2 libxml2-devel
readline readline-devel openssl openssl-libs openssl-devel
useradd -m -s /bin/bash pg95
chmod 755 /home/pg95
su - pg95 -c 'wget https://ftp.postgresql.org/pub/source/v9.5.3/postgresql-9.5.3.tar.gz'
su - pg95 -c 'tar xf postgresql-9.5.3.tar.gz'
su - pg95 -c 'cd postgresql-9.5.3; ./configure --prefix=/home/pg95 --with-libxml
--with-openssl'
su - pg95 -c 'cd postgresql-9.5.3; make -j 2 && make install'

Sample Setup (v.1)

Conﬁguration
• use initdb to create the "cluster" (as in "instance
of postgres serving a set of databases", not as in
a set of machines)
• conﬁguration is in $PGDATA/postgresql.conf
(at the very least check out listen_addresses,
max_connections, shared_buffers and
work_mem)
• ACLs are in $PGDATA/pg_hba.conf
su - pg95 -c 'bin/initdb -D data'
# instance is fully contained in PGDATA=/home/pg95/data now

Starting and Connecting
• pg_ctl is your friend (put this line in /etc/rc.local
and make it executable):
• psql is the universal client:
su - pg95 -c 'bin/pg_ctl -D data -l log start'
[root@p0-primary ~]# su - pg95
Last login: Wed Jun 22 08:47:36 UTC 2016 on pts/0
[pg95@p0-primary ~]$ bin/psql postgres
psql (9.5.3)
Type "help" for help.
postgres=# q
[pg95@p0-primary ~]$

One Elephant at Work -
understanding transactions
• let's generate a ﬁle with single inserts:
• and load it into the database:
• experiments - what happens if:
• you add a begin/commit around the inserts?
• you create an unlogged table?
• you set synchronous_commit to off?
for (( i=0; i < 50000; i++ )) do
echo insert into big values ( $RANDOM ) ;
done
psql postgres -c "drop table big; create table big (x int);"
time psql postgres --quiet < inserts.sql
} outcome will
pretty much depend
on disk type...

understanding the planner
• let's generate a large table with an index:
• and look at the plans for queries such as:
• experiment - what happens if:
• you switch off auto-analyze (parameter autovacuum
= off in postgresql.conf), restart the server, drop and
recreate the table and repeat the experiment?
select random() as x into big from generate_series(1, 1000000);
create index ix on big(x);
explain select count(*) from big where x < 0.00001;

understanding MVCC and locking
• thanks to MVCC, "normal" operations such as update/delete/insert
do not need to lock a table, you can do a:
in one session while the table is fully usable on another session.
only if you try to update/delete THE SAME row, will the second
session be blocked.
• there are, however, operations that need locks on whole tables,
typically I've seen:
• truncate
• DDL statements such as ALTER TABLE
• I've seen situations were postgres instances were very "laggy", while
the system load was low due to lock contention
begin;
update person set name = 'Chris' where id = 1;
-- wait

Useful System Tables
• pg_stat_activity - list of sessions and what they're doing:
select pid, usename, state, query from pg_stat_activity;
• pg_locks (beware for example of AccessExclusiveLock locks on
user tables):
select locktype, database, relation, (select relname from pg_class where
oid = relation), pid, mode from pg_locks;
• pg_stat_all_tables - to check among other things auto-analyze is
good:
select relname, last_analyze, last_autoanalyze from pg_stat_user_tables;
• and many more

Backups
• cold backups - just shut the server down and archive the
$PGDATA directory
• online backups - pg_dump or pg_dumpall:
• pg_dump is per database (or table) with options, for example
binary output
• pg_dumpall is needed to backup the cluster-wide info such
as users
• psql and possibly pg_restore (to read the binary format) are
needed to restore the DBs
• demo as time permits

No More Elephants
• Have a look at Josh Berkus' 7 ways to crash Postgres:
• no updates
• out of disk space
• deleting stuff
• out of RAM
• bad hardware
• too many connections
• zombie locks

More Than One Elephant
• the other meaning of the word "cluster" is somewhat vague - here are some
Postgres features that I currently like to use:
• streaming replication: stream database operations to other nodes in real time
(optionally as 2-safe replication - i.e. at least one slave must have ack’ed a
transaction), this can be cascading too
• hot standby: issue queries on any secondary node (this includes doing online
backups on a secondary to save load from the primary)
• instant failover: promote a hot standby node to primary node instantly with a
single operation for high availability setups
• third party software allows much more, including master-master setups
• recent developments have much enhanced the streaming capabilities, for
example pglogical and BDR - eventually these will be merged into Postgres
(see for example my presentation on BDR)

We've Been Doing it the
Whole Time ;)

Setting up Streaming
Replication with a Hot Standby
• 5 minutes instruction by Cybertec
• our setup scripted for reference:
PRIMARY_IP=10.0.1.123
SECONDARY_IP=10.0.1.124
# primary setup
su - pg95 -c 'bin/initdb -D data'
sed -i "s/#listen_addresses = 'localhost'/listen_addresses = '*'/" /home/pg95/data/postgresql.conf
sed -i "s/#wal_level = minimal/wal_level = hot_standby/" /home/pg95/data/postgresql.conf
sed -i "s/#max_wal_senders = 0/max_wal_senders = 3/" /home/pg95/data/postgresql.conf
sed -i "s/#wal_keep_segments = 0/wal_keep_segments = 1024/" /home/pg95/data/postgresql.conf
sed -i "s/#hot_standby = off/hot_standby = on/" /home/pg95/data/postgresql.conf
echo "host replication all $SECONDARY_IP/32 trust" >> /home/pg95/data/pg_hba.conf
su - pg95 -c 'bin/pg_ctl -D data -l log start'
# note: use ssl and don't use trust auth in production, also have a look at the feature "replication slots"
# and if you're doing online backups on the standby see 25.5.2. Handling Query Conflicts in the manual
# secondary setup
su - pg95 -c 'mkdir data && chmod 700 data'
su - pg95 -c "bin/pg_basebackup -h $PRIMARY_IP -D /home/pg95/data --xlog-method=stream"
su - pg95 -c "echo 'standby_mode = on' > data/recovery.conf"
su - pg95 -c "echo "primary_conninfo = 'host=$PRIMARY_IP'" >> data/recovery.conf"
su - pg95 -c "echo "trigger_file = '/tmp/promoteme'" >> data/recovery.conf"

Streaming Experiments
• screenshot from another demo (with machines
africa and asia):

L'Appetito vien mangiando
• from the point of view of the application:
• hey, a connection pool would be handy!
• mmm.... in case of failover to the standby, how
am I notiﬁed that I need to change my JDBC
URL?
• come to think of it, it would be cool to off-load
read-only queries to the secondary server(s),
but I don't want to handle that logic by myself...

Enter pgpool-II
• pgpool-II is a middleware that does exactly this:
• it hides Postgres servers behind one port 5432
• it does connection pooling
• it does load balancing with the ability to pre-parse queries and send read-only
once to the standbys
• and much more:
• it can do replication by sending the same queries to multiple servers (this is
master-master replication even, but it is less efﬁcient and more fragile than doing
it with streaming replication)
• it has a built-in watchdog for high availability setups with two pgool-II servers and
virtual IPs
• etc.

pgpool-II
• here is a pgpool-II presentation from the author of the software - this is
what we want to do (from the linked presentation):

The pool is ready!

Experiments
• demo what we have on p2, enable query logging
on p0 and p1 to see the load balancing in action,
see what happens if p0 or p1 goes down!
• our setup for reference:
# note: make a db user nobdody for the monitoring and make a pg_hba.conf entry on p0 and 01 too...
useradd -m -s /bin/bash pgpool
su - pgpool -c 'wget -O pgpool-II-3.5.3.tar.gz http://www.pgpool.net/download.php?f=pgpool-II-3.5.3.tar.gz'
su - pgpool -c 'tar xf pgpool-II-3.5.3.tar.gz'
su - pgpool -c 'cd pgpool-II-3.5.3; ./configure --prefix=/home/pgpool --with-openssl --with-pgsql=/home/pg95'
su - pgpool -c 'cd pgpool-II-3.5.3; make -j 2 && make install'
su - pgpool -c 'cp etc/pgpool.conf.sample-stream etc/pgpool.conf'
su - pgpool -c 'cp etc/pool_hba.conf.sample etc/pool_hba.conf'
su - pgpool -c 'cp etc/pcp.conf.sample etc/pcp.conf'
sed -i "s/^backend_/#backend_/" /home/pgpool/etc/pgpool.conf
sed -i "s/^pid_file_name = '/var/run/pgpool/pgpool.pid'/pid_file_name = '/home/pgpool/pgpool.pid'/" /home/pgpool/etc/pgpool.conf
sed -i "s/^logdir = '/tmp'/logdir = '/home/pgpool'/" /home/pgpool/etc/pgpool.conf
sed -i "s/^health_check_period = 0/health_check_period = 1/" /home/pgpool/etc/pgpool.conf
echo "backend_hostname0 = '$PRIMARY_IP'" >> /home/pgpool/etc/pgpool.conf
echo "backend_port0 = 5432" >> /home/pgpool/etc/pgpool.conf
echo "backend_weight0 = 1" >> /home/pgpool/etc/pgpool.conf
echo "backend_hostname1 = '$SECONDARY_IP'" >> /home/pgpool/etc/pgpool.conf
echo "backend_port1 = 5432" >> /home/pgpool/etc/pgpool.conf
echo "backend_weight1 = 1" >> /home/pgpool/etc/pgpool.conf
echo "pgpool:d41d8cd98f00b204e9800998ecf8427e" >> /home/pgpool/etc/pcp.conf # empty password
su - pgpool -c 'nohup pgpool -n 2> log &'

Failover
• one of the cool features of pgpool-II is that
events from nodes attaching/detaching can be
scripted
• demo (if time permits) how to instruct pgpool-II to
connect to the standby over SSH and touch the
trigger ﬁle to trigger a promotion to primary
• however, always be aware that automatic failover
can be tricky (test well!)

A Simpler Pool
• if you don't need load balancing and automatic
failover, I recommend PgBouncer
• PgBouncer is "only" a connection pool, but it
does that job really well
• you can also combine pgpool-II and PgBouncer

'k thx bye ;)

Oracle to Postgres Migration - part 2

More Related Content

What's hot (20)

Viewers also liked (15)

Similar to Oracle to Postgres Migration - part 2 (20)

More from PgTraining (7)

Recently uploaded (20)

Oracle to Postgres Migration - part 2