Generic Postgres Upgrade Document

Task

Command

Comments

Create a new postgres instance

Choose postgres version 10 (preferably 11 ) and above and same configuration as per existing instance on Azure

You are free to have independent instances or a single instance for all services as per your need

Whitelist all the IP's as per existing postgres

Test the connection by connecting to the instance from jenkins

If on VM, create a new VM (preferably Ubuntu 18 and above) and run the Postgres provision role.

Note: This document does not cover details on how to upgrade postgres on the same / exisiting VM. Feel free to contribute the steps to this document.

New instance of same configuration should be created. We have used Postgres version 11

Update the inventory

Search and update all the variables in common.yml, secrets.yml, kubernetes.yml in KP, DP and Core

Replace the postgres instance address, postgres user and postgres password with the new details

It will be simpler if you use same username and password for all dbs on that instance. If you want to use different user name and password for each DB on an instance, then you need to ensure you create those users and roles before hand by running provision jobs against your postgres instance or do them manually.

This document does not cover how to restore user accounts. For that you can refer to Azure docs

https://docs.microsoft.com/en-us/azure/postgresql/how-to-upgrade-using-dump-and-restore

Start of down time - Stop traffic

Stop traffic, put Jenkins into maintenance mode so nobody can deploy the jobs

Cut off connection to postgres from all services except Jenkins (In azure remove all connecting subnets except Jenkins, disable Azure services connections also)

In VM you have multiple ways - Disable outside connections by editing pg_hba.conf so only localhost connections are accepted. Or stop the below services so that they dont update postgres database

List of services that are going to be affected -

- Analytics
- APIManager
- Druid
- ENC
- fusionauth
- gql
- Hawkeye Superset
- Inbound
- InternalKong
- Keycloak
- Learner
- LMS
- odk
- Orchestrator
- Outbound
- Report
- Superset
- Transformer
- UCI

Respective services should be stopped / Postgres reachability should be removed

Traffic can be stopped based on situation by removing nginx daemonset or changing service port to something else from 80 / 443

Manually list all the DB's first

export PG_HOST="" # Enter the postgres host inside the quotes
export PG_USER="" # Enter the postgres user inside the quotes
export PGPASSWORD="" # Enter the postgres password inside the quotes

psql -h $PG_HOST -U $PG_USER -d postgres -c "\l"

All DB should be listed

Run command to get all the DB's on terminal and store in a file

Run command to get all user accounts and store in a file

mkdir $PG_HOST-$(date +'%s') && cd $PG_HOST-$(date +'%s')

psql -h $PG_HOST -U $PG_USER -d postgres -c "\l" | awk 'NR>3{print $1}'| grep -v "|\|(\|^$\|template0\|template1\|azure_maintenance\|postgres\|azure_sys" | tee -a dbs.txt

Verify all db are present in the file and matches our db's
Ignore db's like postgres, template0, template1, azure_maintanance and remove other not required DB's from the file

Take backup of the DB's from current instance

while read -r line; do echo "Dumping DB $line" && pg_dump -Fd -j PG_CPU_CORES -h $PG_HOST -U $PG_USER -d $line -f $line; done < dbs.txt

Replace PG_CPU_CORES with the number of cores of the postgres insatnce.
Example: For 4 core, value will be 4

Get count of all tables from all DB's from current instance

while read -r line; do psql -h $PG_HOST -U $PG_USER -d $line -c "\dt" | awk -F "|" 'NF {print $2}' | tr -d ' ' | awk 'NF' | tail -n +2 | tee -a $line-tables.txt; done < dbs.txt

while read -r line; do while read -r inline; do echo $inline | tee -a $line-table-count.txt && psql -h $PG_HOST -U $PG_USER -d $line -c "SELECT COUNT(*) FROM \"$inline\"" | tee -a $line-table-count.txt; done < $line-tables.txt; done < dbs.txt

All tables and counts will be displayed and also written into the files

Rearrange the DBs across two instances or based on your requirement

In this case, we will be creating the following databases across two instances

Instance 1 - Keycloak, Public Kong, Private Kong, Quartz, Enc Keys
Instance 2 - Analytics, Druid, Graphite

Feel free to add any other dbs you have / want to. This is not an exhaustive list

Create two directories as below

mkdir pg11
mkdir pg11-dp

cd pg11
touch dbs.txt
Add the required db names in this file

Move all the required backup folders and files to this directory. Below is a sample command
mv loadtest-new.postgres.database.azure.com-1636348963/api_manager_loadtest_kong14* pg11

Similarly do for the other dbs and pg11-db folder also (second instance)

cd pg11-dp
touch dbs.txt
Add the required db names in this file

If you are going to use only once instance, then you can move all the backup data and other files under a single directory

Instance 1:

cat pg11/dbs.txt 
api_manager_loadtest_kong14
keycloak7
quartz
api_manager_internal
loadtest-keys

ls -lrth pg11
api_manager_internal
api_manager_internal-table-count.txt
api_manager_internal-tables.txt
api_manager_loadtest_kong14
api_manager_loadtest_kong14-table-count.txt
api_manager_loadtest_kong14-tables.txt
keycloak7
keycloak7-table-count.txt
keycloak7-tables.txt
loadtest-keys
loadtest-keys-table-count.txt
loadtest-keys-tables.txt
quartz
quartz-table-count.txt
quartz-tables.txt
dbs.txt

Instance 2:

cat pg11-dp/dbs.txt 
analytics
druid-raw
graphite
superset

ls -lrth pg11-dp
analytics
analytics-table-count.txt
analytics-tables.txt
druid-raw
druid-raw-table-count.txt
druid-raw-tables.txt
graphite
graphite-table-count.txt
graphite-tables.txt
superset
superset-tables.txt
dbs.txt

Install postgres 11 tools

sudo apt install postgresql-client-11

If you to use pg_dump and pg_restore on older postgres version, install those packages

sudo apt install postgresql-client-9.6

Create empty databases in new instance

export PG_HOST="" # Enter the new postgres host inside the quotes
export PG_USER="" # Enter the new postgres user inside the quotes
export PGPASSWORD="" # Enter the new postgres password inside the quotes
psql -h $PG_HOST -U $PG_USER -d postgres -c "\l"

while read -r line; do echo "Creating DB $line" && psql -h $PG_HOST -U $PG_USER -d postgres -c "CREATE DATABASE $line"; done < dbs.txt
psql -h $PG_HOST -U $PG_USER -d postgres -c "\l"

Empty databases should be created on new postgres instance

Add double quotes if you have hyphen in DB name "DB-NAME" or else you will receive below error

Creating DB loadtest-keys
ERROR: syntax error at or near "-"
LINE 1: CREATE DATABASE loadtest-keys

Restore the DB's to the new instance

while read -r line; do echo "Restoring DB $line" && pg_restore -O -j PG_CPU_CORES -h $PG_HOST -U $PG_USER -d $line $line; done < dbs.txt

Replace PG_CPU_CORES with the number of cores of the postgres insatnce.
Example: For 4 core, value will be 4

Ignore errors like
ERROR: role "postgres" does not exist

Ignore errors like
pg_restore: [archiver (db)] Error while PROCESSING TOC:
pg_restore: [archiver (db)] Error from TOC entry 5; 2615 2200 SCHEMA public azure_superuser
pg_restore: [archiver (db)] could not execute query: ERROR: schema "public" already exists
Command was: CREATE SCHEMA public;

Analyze the db's on new instance

while read -r line; do psql -h $PG_HOST -U $PG_USER -d $line -c "ANALYZE VERBOSE"; done < dbs.txt

Verbose logs will be displayed

Get count of all tables from all DB's from new instance

while read -r line; do while read -r inline; do echo $inline | tee -a $line-table-count-new.txt && psql -h $PG_HOST -U $PG_USER -d $line -c "SELECT COUNT(*) FROM \"$inline\"" | tee -a $line-table-count-new.txt; done < $line-tables.txt; done < dbs.txt

All tables and counts will be displayed and also written into the files

Compare the row counts of both the instances

while read -r line; do echo "Diff of $line-table-count.txt and $line-table-count-new.txt" && diff $line-table-count.txt $line-table-count-new.txt; done < dbs.txt

This will not display any output. Which means files are identical

If there are differences use the below to comapre
diff -y file1 file2

Remove all connections to old Postgres

Remove all subnets and enable deny public network acces on Azure portal

If on VM, stop the VM / postgres service

To ensure no service can connect to old DBs

Clear unnecessary data

Clear offline session from Keycloak DB

truncate offline_client_session

truncate offline_user_session

If we have millions of rows in these tables, Keycloak will not start

End of down time - Redeploy services

Redeploy the follwing services under Deploy/Kubernetes directory

- Analytics
- APIManager
- Provision/DataPipeline/Druid (Choose all in service except java)
- ENC
- fusionauth
- gql
- HawkeyeSuperset
- Inbound
- Deploy/atapipeline/InternalKong
- Keycloak
- Learner
- LMS
- odk
- Orchestrator
- Outbound
- Report
- Superset
- Transformer
- UCI

The configmaps and configuration files will be updated with new data. Do verify it by checking them manually for all these services.