blob: e4f87e14a232e2aac35be96cbcb3f3a48f5086f9 (
plain) (
tree)
|
|
#!/bin/bash -ex
# {{ ansible_managed }}
#
# WARNING: Please read both the documentation and this script if you want to run it
# Authors:
# - Théophile Helleboid
# - Paul Bonaud
# Description:
# This helper tries to strictly follow the official documentation of PostgreSQL
# from the pg_upgrade page: https://www.postgresql.org/docs/9.5/static/pgupgrade.html
# Usage:
# WARNING: Please read both the documentation and this script if you want to run it
# 1. edit this file and remove the "DRY_RUN=1" line
# 2. execute this script as the postgres admin user
{% if postgres_pgbouncer_uri is defined and postgres_pgbouncer_uri %}
PGBOUNCER_PG_URI={{ postgres_pgbouncer_uri }}
{% endif %}
DATABASE_NAME={{ postgres_dbname }}
DATABASE_PORT={{ postgres_new_cluster_port }}
OLD_CLUSTER_VERSION={{ postgres_old_cluster_version }}
OLD_CLUSTER_NAME={{ postgres_old_cluster_name }}
OLD_CLUSTER_DATADIR=/var/lib/postgresql/$OLD_CLUSTER_VERSION/$OLD_CLUSTER_NAME
NEW_CLUSTER_VERSION={{ postgres_new_cluster_version }}
NEW_CLUSTER_NAME={{ postgres_new_cluster_name }}
NEW_CLUSTER_DATADIR=/var/lib/postgresql/$NEW_CLUSTER_VERSION/$NEW_CLUSTER_NAME
STANDBY_SERVER={{ postgres_standby_server }}
STANDBY_OLD_CLUSTER_NAME={{ postgres_standby_old_cluster_name }}
STANDBY_OLD_CLUSTER_DATADIR=/var/lib/postgresql/$OLD_CLUSTER_VERSION/$STANDBY_OLD_CLUSTER_NAME
STANDBY_NEW_CLUSTER_NAME={{ postgres_standby_new_cluster_name }}
STANDBY_NEW_CLUSTER_DATADIR=/var/lib/postgresql/$NEW_CLUSTER_VERSION/$STANDBY_NEW_CLUSTER_NAME
DRY_RUN=1
if [ "$(whoami)" != "postgres" ]; then
echo "ERROR: you must run this script as the 'postgres' admin local user"
exit 1
fi
{% if postgres_pgbouncer_uri is defined and postgres_pgbouncer_uri %}
# Test the configuration is correcly deployed on pgbouncer
psql $PGBOUNCER_PG_URI --command 'show databases;' | grep "^ $DATABASE_NAME "
# Test connection to pgbouncer is OK
psql $PGBOUNCER_PG_URI --command 'show pools;'
{% endif %}
if [ ! -f /usr/lib/postgresql/$NEW_CLUSTER_VERSION/bin/pg_upgrade ]; then
echo "ERROR: /usr/lib/postgresql/$NEW_CLUSTER_VERSION/bin/pg_upgrade is not available on this machine."
exit 1
fi
# We need ssh access to the standby server
ssh $STANDBY_SERVER "ls -A $STANDBY_OLD_CLUSTER_DATADIR 2>&1>/dev/null"
ssh $STANDBY_SERVER "ls -A $STANDBY_NEW_CLUSTER_DATADIR 2>&1>/dev/null"
if [ ! -d $OLD_CLUSTER_DATADIR ] || [ ! -d $NEW_CLUSTER_DATADIR ]; then
echo "ERROR: Make sure both old cluster and new cluster data dirs exists"
echo "ERROR: on both primary and standby servers ($STANDBY_SERVER)."
echo "ERROR: old_cluster datadir: $OLD_CLUSTER_DATADIR"
echo "ERROR: new_cluster datadir: $NEW_CLUSTER_DATADIR"
#exit 1
fi
echo "WARNING: You are about to:"
{% if postgres_pgbouncer_uri is defined and postgres_pgbouncer_uri %}
echo "WARNING: * PAUSE connections to $DATABASE_NAME database."
{% endif %}
echo "WARNING: * STOP the primary database in $OLD_CLUSTER_VERSION version (and standbys)"
echo "WARNING: * dry-run the pg_upgrade from $OLD_CLUSTER_VERSION to $NEW_CLUSTER_VERSION"
echo "WARNING: * run the pg_upgrade in place with hardlinks"
echo "WARNING: * START the primary database in $NEW_CLUSTER_VERSION version (and standbys)"
{% if postgres_pgbouncer_uri is defined and postgres_pgbouncer_uri %}
echo "WARNING: * RESUME connections to $DATABASE_NAME."
{% endif %}
echo "================"
read -p "Are you sure? (y/n)" -n 1 -r
echo "================"
if [[ -z $DRY_RUN ]] && [[ $REPLY =~ ^[Yy]$ ]]; then
{% if postgres_pgbouncer_uri is defined and postgres_pgbouncer_uri %}
# Pause the databases;
psql $PGBOUNCER_PG_URI --command "PAUSE $DATABASE_NAME;"
{% endif %}
# Stop the $OLD_CLUSTER_VERSION server
pg_ctlcluster $OLD_CLUSTER_VERSION $OLD_CLUSTER_NAME stop -m fast
# Get latest checkpoint location of primary
echo "INFO: Primary PG controldata with latest checkpoint location"
/usr/lib/postgresql/$OLD_CLUSTER_VERSION/bin/pg_controldata $OLD_CLUSTER_DATADIR | head -n8
# Wait for standbys to catch up latest checkpoint
# TODO: how long to wait?
echo "Waiting for 9 seconds..."
sleep 9
echo "INFO: Standby PG controldata with latest checkpoint location (after shutdown)"
ssh $STANDBY_SERVER pg_ctlcluster $OLD_CLUSTER_VERSION $STANDBY_OLD_CLUSTER_NAME stop -m fast
ssh $STANDBY_SERVER /usr/lib/postgresql/$OLD_CLUSTER_VERSION/bin/pg_controldata $STANDBY_OLD_CLUSTER_DATADIR | head -n8
echo "WARNING: 'Latest Checkpoint location' values should match on primary and on standby"
echo "================"
read -p "Do they match? PLEASE ANSWER (y/n)" -n 1 -r
echo "================"
if [[ $REPLY =~ ^[Yy]$ ]]; then
# Test to upgrade the data
time /usr/lib/postgresql/$NEW_CLUSTER_VERSION/bin/pg_upgrade --check --link \
-b /usr/lib/postgresql/$OLD_CLUSTER_VERSION/bin \
-B /usr/lib/postgresql/$NEW_CLUSTER_VERSION/bin \
-d $OLD_CLUSTER_DATADIR \
-D $NEW_CLUSTER_DATADIR \
-o " -c config_file=/etc/postgresql/$OLD_CLUSTER_VERSION/$OLD_CLUSTER_NAME/postgresql.conf" \
-O " -c config_file=/etc/postgresql/$NEW_CLUSTER_VERSION/$NEW_CLUSTER_NAME/postgresql.conf"
# Really upgrade the data
time /usr/lib/postgresql/$NEW_CLUSTER_VERSION/bin/pg_upgrade --link \
-b /usr/lib/postgresql/$OLD_CLUSTER_VERSION/bin \
-B /usr/lib/postgresql/$NEW_CLUSTER_VERSION/bin \
-d $OLD_CLUSTER_DATADIR \
-D $NEW_CLUSTER_DATADIR \
-o " -c config_file=/etc/postgresql/$OLD_CLUSTER_VERSION/$OLD_CLUSTER_NAME/postgresql.conf" \
-O " -c config_file=/etc/postgresql/$NEW_CLUSTER_VERSION/$NEW_CLUSTER_NAME/postgresql.conf"
## Upgrade standby server
### Making sure new datadir is empty
if [ -z "$(ssh $STANDBY_SERVER ls -A $STANDBY_NEW_CLUSTER_DATADIR)" ]; then
echo "INFO: New cluster datadir is empty on standby server. Good."
else
echo "WARNING: New cluster datadir is NOT empty on standby server. Deleting the content of $STANDBY_NEW_CLUSTER_DATADIR on $STANDBY_SERVER now..."
ssh $STANDBY_SERVER rm -rf $STANDBY_NEW_CLUSTER_DATADIR
fi
### Save configuration files
TMPDIR=$(ssh $STANDBY_SERVER mktemp -d -t pg_upgrade_XXXX)
ssh $STANDBY_SERVER mv $STANDBY_OLD_CLUSTER_DATADIR/recovery.conf $TMPDIR
### Upgrade (via rsync)
rsync --archive --delete --hard-links --size-only --no-inc-recursive $OLD_CLUSTER_DATADIR $NEW_CLUSTER_DATADIR $STANDBY_SERVER:$(dirname $STANDBY_NEW_CLUSTER_DATADIR)
### Restore saved configured files
ssh $STANDBY_SERVER mv $TMPDIR/recovery.conf $STANDBY_NEW_CLUSTER_DATADIR/
ssh $STANDBY_SERVER rm -rf $TMPDIR
# Start the $NEW_CLUSTER_VERSION primary server
pg_ctlcluster $NEW_CLUSTER_VERSION $NEW_CLUSTER_NAME start
# Test local primary is accepting connections
psql --port=$DATABASE_PORT --dbname=$DATABASE_NAME --command="SELECT NOW();"
{% if postgres_pgbouncer_uri is defined and postgres_pgbouncer_uri %}
# Resume the connexions
psql $PGBOUNCER_PG_URI --command "RESUME $DATABASE_NAME;"
{% endif %}
# Start the $NEW_CLUSTER_VERSION standby server
ssh $STANDBY_SERVER pg_ctlcluster $NEW_CLUSTER_VERSION $STANDBY_NEW_CLUSTER_NAME start
echo "DONE. Upgrade of PG cluster from $OLD_CLUSTER_VERSION/$OLD_CLUSTER_NAME to $NEW_CLUSTER_VESION/$NEW_CLUSTER_NAME is finished! Well done!"
else
# Start the $OLD_CLUSTER_VERSION primary server
pg_ctlcluster $OLD_CLUSTER_VERSION $OLD_CLUSTER_NAME start
# Start the $OLD_CLUSTER_VERSION standby server
ssh $STANDBY_SERVER pg_ctlcluster $OLD_CLUSTER_VERSION $OLD_CLUSTER_NAME start
# Test local primary is accepting connections
psql --dbname=$DATABASE_NAME --command="SELECT NOW();"
{% if postgres_pgbouncer_uri is defined and postgres_pgbouncer_uri %}
# Resume the connexions
psql $PGBOUNCER_PG_URI --command "RESUME $DATABASE_NAME;"
{% endif %}
fi
else
echo "INFO: You didn't want to continue or was in 'dry-run' mode"
fi
|