pgBaseBackup

pgBaseBackup Examples

The script assumes you are going to backup the primary container created in the first example, so you need to ensure that container is running. This example assumes you have configured storage as described in the Storage Configuration documentation. Things to point out with this example include its use of persistent volumes and volume claims to store the backup data files.

A successful backup will perform pg_basebackup on the primary container and store the backup in the $CCP_STORAGE_PATH volume under a directory named $CCP_NAMESPACE-primary-backups. Each backup will be stored in a subdirectory with a timestamp as the name, allowing any number of backups to be kept.

The backup script will do the following:

  • Start up a backup container or Kubernetes job named backup
  • Run pg_basebackup on the container named primary
  • Store the backup in the /backup volume
  • Exit after the backup

When you are ready to restore from the backup, the restore example runs a pgbasebackup-restore container or Kubernetes job in order to restore the database into a new pgdata directory, specifically using rsync to copy the backup data into the new directory. A crunchy-postgres container then deployed using the restored database, deploying a new PostgreSQL DB that utilizes the original backed-up data.

The restore scripts will do the following:

  • Start up a container or Kubernetes job named restore
  • Copy the backup files from the previous backup example into /pgdata
  • Deploy a crunchy-postgres container using the restored database

To shutdown and remove any artifacts from the example, run the following under each directory utilized when running the example:

./cleanup.sh

Docker

Perform a backup of primary using pg_basebackup by running the following script:

cd $CCPROOT/examples/docker/pgbasebackup/backup
./run.sh

When you’re ready to restore, a restore example is provided. In order to run the restore, a path for the backup must be provided to the pgabasebackup-restore container using the BACKUP_PATH environment variable. To get the correct path, check the logs for the backup container:

docker logs backup | grep BACKUP_PATH
Fri May 10 01:40:06 UTC 2019 INFO: BACKUP_PATH is set to /pgdata/primary-backups/2019-05-10-01-40-06.

BACKUP_PATH can also be discovered by looking at the backup mount directly (if access to the storage is available to the user).

When you are ready to restore from the backup, first update the $CCPROOT/examples/docker/pgbasebackup/full/run.sh script used to run restore example with the proper BACKUP_PATH:

--env BACKUP_PATH=primary-backups/2019-05-09-11-53-32 \

Then run the restore:

cd $CCPROOT/examples/docker/pgbasebackup/full
./run.sh

Once the restore is complete, the restored database can then be deployed by running the post restore script:

./post-restore.sh

You can then test the restored database as follows:

docker exec -it pgbasebackup-full-restored psql

Kubernetes and OpenShift

Perform a backup of primary using pg_basebackup by running the following script:

cd $CCPROOT/examples/kube/pgbasebackup/backup
./run.sh

This runs a Kubernetes job, which deploys a pod that performs the backup using pg_basebackup, and then exits. You can view the status of the job by running the following command:

${CCP_CLI} get job

When you’re ready to restore, a restore example is provided. In order to run the restore, a path for the backup must be provided to the pgabasebackup-restore container using the BACKUP_PATH environment variable. To get the correct path, check the logs for the backup job:

kubectl logs backup-txcvm | grep BACKUP_PATH
Fri May 10 01:40:06 UTC 2019 INFO: BACKUP_PATH is set to /pgdata/primary-backups/2019-05-10-01-40-06.

BACKUP_PATH can also be discovered by looking at the backup mount directly (if access to the storage is available to the user).

When you are ready to restore from the backup, first update $CCPROOT/examples/kube/pgbasebackup/full/restore.json with the proper BACKUP_PATH:

{
    "name": "BACKUP_PATH",
    "value": "primary-backups/2019-05-08-18-28-45"
}

Then run the restore:

cd $CCPROOT/examples/kube/pgbasebackup/full
./run.sh

Once the restore is complete, the restored database can then be deployed by running the post restore script:

./post-restore.sh

You can then test the restored database as follows:

kubectl exec -it pgbasebackup-full-restored-7d866cd5f7-qr6w8 -- psql

Point in Time Recovery (PITR)

PITR (point-in-time-recovery) is a feature that allows for recreating a database from backup and log files at a certain point in time. This is done using a write ahead log (WAL) which is kept in the pg_wal directory within PGDATA. Changes made to the database files over time are recorded in these log files, which allows it to be used for disaster recovery purposes.

When using PITR as a backup method, in order to restore from the last checkpoint in the event of a database or system failure, it is only necessary to save these log files plus a full backup. This provides an additional advantage in that it is not necessary to keep multiple full backups on hand, which consume space and time to create. This is because point in time recovery allows you to “replay” the log files and recover your database to any point since the last full backup.

More detailed information about Write Ahead Log (WAL) archiving can be found here.

By default in the crunchy-postgres container, WAL logging is not enabled. To enable WAL logging outside of this example, set the following environment variables when starting the crunchy-postgres container:

ARCHIVE_MODE=on
ARCHIVE_TIMEOUT=60

These variables set the same name settings within the postgresql.conf file that is used by the database. When set, WAL files generated by the database will be archived to the /pgwal mount point.

A full backup is required to do a PITR. The crunchy-backup container is utilized to perform the backup in the example below, specifically running pg_basebackup to backup the database.

After the backup is complete, a restore can then be performed. The restore is performed using the pgbasebackup-restore container, which uses rysnc to copy a pg_basebackup backup into a new pgdata directory.

There are two volume mounts used when performing the restore.

  • /backup - The volume containing the backup you would like to restore from.
  • /pgdata - The volume containing the restored database

The following environment variables can be used to manipulate the point in time recovery logic when performing the restore using the pgbasebackup-restore container, specifically configuring the proper recovery target within the recovery.conf file:

  • RECOVERY_TARGET_NAME - Used to restore to a named restore point
  • RECOVERY_TARGET_TIME - Used to restore to a specific timestamp
  • RECOVERY_TARGET_XID - Used to restore to a specific transaction ID

If you would rather restore to the end of the WAL log, the RECOVERY_REPLAY_ALL_WAL environment varibale can be set to true. Please note that if this enviornment variable is set, then any recovery targets specified via the environment variables described above will be ignored.

The RECOVERY_TARGET_INCLUSIVE environment variable is also available to let you control the setting of the recovery.conf setting recovery_target_inclusive. If you do not set this environment variable the default is true.

Once you recover a database using PITR, it will be in read-only mode. To make the database resume as a writable database, run the following SQL command:

postgres=# select pg_wal_replay_resume();
If you’re running the PITR example for PostgreSQL versions 9.5 or 9.6, please note that starting in PostgreSQL version 10, the pg_xlog directory was renamed to pg_wal. Additionally, all usages of the function pg_xlog_replay_resume were changed to pg_wal_replay_resume.

It takes about 1 minute for the database to become ready for use after initially starting.

To shutdown and remove any artifacts from the example, run the following:

./cleanup.sh

Docker

Create a database container as follows:

cd $CCPROOT/examples/docker/pgbasebackup/pitr
./run-pitr.sh

Next, we will create a base backup of that database as follows:

./run-backup-pitr.sh

This will create a backup and write the backup files to a persistent volume (specifically Docker named volume pitr-backup-volume). Additionally, WAL segment files will be created every 60 seconds under the pgwal directory of the running pitr container that contain any additional database changes.

Next, create some recovery targets within the database by running the SQL commands against the pitr database as follows:

./run-sql.sh

This will create recovery targets named beforechanges, afterchanges, and nomorechanges. It will create a table, pitrtest, between the beforechanges and afterchanges targets. It will also run a SQL CHECKPOINT to flush out the changes to WAL segments. These labels can be used to mark the points in the recovery process that will be referenced when creating the restored database.

Next, now that we have a base backup and a set of WAL files containing our database changes, we can shut down the pitr database to simulate a database failure. Do this by running the following:

docker stop pitr

Now get the BACKUP_PATH created by the backup-pitr example by viewing the containers logs:

docker logs backup-pitr | grep PATH
Thu May 10 18:07:58 UTC 2018 INFO: BACKUP_PATH is set to /pgdata/pitr-backups/2018-05-10-18-07-58.

Edit the run-restore-pitr.sh file and change the BACKUP_PATH environment variable using the path discovered above:

--env BACKUP_PATH=pitr-backups/2018-05-10-18-07-58 \

Next, we restore prior to the beforechanges recovery target. This recovery point is before the pitrtest table is created.

Open file run-restore-pitr.sh, and edit the environment variable to indicate we want to use the beforechanges recovery point:

--env RECOVERY_TARGET_NAME=beforechanges \

Then run the following to restore the database:

./run-restore-pitr.sh

Once the restore is complete, the restored database can then be deployed by running the post restore script:

./post-restore.sh

As a result of the recovery.conf file configured using RECOVERY_TARGET_NAME when performing the restore, the WAL segments are read from the WAL archive and applied up until the beforechanges named restore point. At this point you should therefore be able to verify that the database was restored to the point before creating the test table:

docker exec -it pgbasebackup-pitr-restored psql -c 'table pitrtest'

This SQL command should show that the pitrtest table does not exist at this recovery time. The output should be similar to:

ERROR: relation "pitrtest" does not exist

PostgreSQL allows you to pause the recovery process if the target name or time is specified. This pause would allow a DBA a chance to review the recovery time/name/xid and see if this is what they want or expect. If so, the DBA can run the following command to resume and complete the recovery:

docker exec -it pgbasebackup-pitr-restored psql -c 'select pg_wal_replay_resume()'

Until you run the statement above, the database will be left in read-only mode.

Next, run the script to restore the database to the afterchanges restore point. Update the RECOVERY_TARGET_NAME to afterchanges in run-restore-pitr.sh:

--env RECOVERY_TARGET_NAME=afterchanges \

Then run the following to again restore the database:

./run-restore-pitr.sh

Once the restore is complete, the restored database can then once again be deployed by running the post restore script:

./post-restore.sh

After this restore you should be able to see the test table, and will still be required to make the database writable:

docker exec -it pgbasebackup-pitr-restored psql -c 'table pitrtest'
docker exec -it pgbasebackup-pitr-restored psql -c 'select pg_wal_replay_resume()'

Lastly, start a restore to the end of the WAL log. This will restore the database to the most current point possible. To do so, edit run-restore-pitr.sh and remove RECOVERY_TARGET_NAME, and then set environment variable RECOVERY_REPLAY_ALL_WAL to true:

--env RECOVERY_REPLAY_ALL_WAL=true \

Then run the following to again restore the database:

./run-restore-pitr.sh

Once the restore is complete, the restored database can then once again be deployed by running the post restore script:

./post-restore.sh

At this point, you should be able to create new data in the restored database and the test table should be present. When you recover the entire WAL history, resuming the recovery is not necessary to enable writes.

docker exec -it pgbasebackup-pitr-restored psql -c 'table pitrtest'
docker exec -it pgbasebackup-pitr-restored psql -c 'create table foo (id int)'

Kubernetes and OpenShift

Start by running the example database container:

cd $CCPROOT/examples/kube/pgbasebackup/pitr
./run-pitr.sh

This step will create a database container, pitr. This container is configured to continuously archive WAL segment files to a mounted volume (/pgwal).

After you start the database, you will create a pg_basebackup backup using this command:

./run-backup-pitr.sh

This will create a backup and write the backup files to a persistent volume made available to the crunchy-backup container via Persistent Volume Claim backup-pitr-pgdata.

Next, create some recovery targets within the database by running the SQL commands against the pitr database as follows:

./run-sql.sh

This will create recovery targets named beforechanges, afterchanges, and nomorechanges. It will create a table, pitrtest, between the beforechanges and afterchanges targets. It will also run a SQL CHECKPOINT to flush out the changes to WAL segments.

Next, now that we have a pg_basebackup backup and a set of WAL files containing our database changes, we can shut down the pitr database to simulate a database failure. Do this by running the following:

${CCP_CLI} delete deployment pitr

Next, we will create 3 different restored database containers based upon the base backup and the saved WAL files.

First, get the BACKUP_PATH created by the backup-pitr example by viewing the pods logs:

${CCP_CLI} logs backup-pitr-8sfkh | grep PATH
Thu May 10 18:07:58 UTC 2018 INFO: BACKUP_PATH is set to /pgdata/pitr-backups/2018-05-10-18-07-58.

Edit the restore-pitr.json file and change the BACKUP_PATH environment variable using the path discovered above (note: /pgdata/ is not required and should be excluded in the variable):

{
    "name": "BACKUP_PATH",
    "value": "pitr-backups/2018-05-10-18-07-58"
{

Next, we restore prior to the beforechanges recovery target. This recovery point is before the pitrtest table is created.

Edit the restore-pitr.json file, and edit the environment variable to indicate we want to use the beforechanges recovery point:

{
    "name": "RECOVERY_TARGET_NAME",
    "value": "beforechanges"
{

Then run the following to restore the database:

./run-restore-pitr.sh

Once the restore is complete, the restored database can then once again be deployed by running the post restore script:

./post-restore.sh

After the database has restored, you should be able to perform a test to see if the recovery worked as expected:

kubectl exec -it pgbasebackup-pitr-restored-5c5df7894c-hczff -- psql -c 'table pitrtest'
kubectl exec -it pgbasebackup-pitr-restored-5c5df7894c-hczff -- psql -c 'create table foo (id int)'

The output of these commands should show that the pitrtest table is not present. It should also show that you can not create a new table because the database is paused in read-only mode.

To make the database resume as a writable database, run the following SQL command:

kubectl exec -it pgbasebackup-pitr-restored-5c5df7894c-hczff -- psql -c 'select pg_wal_replay_resume()'

It should then be possible to write to the database:

kubectl exec -it pgbasebackup-pitr-restored-5c5df7894c-hczff -- psql -c 'create table foo (id int)'

You can also test that if afterchanges is specified, that the pitrtest table is present but that the database is still in recovery mode.

Lastly, you can test a full recovery using all of the WAL files by removing any recovery targets, and setting RECOVERY_REPLAY_ALL_WAL to true.

The storage portions of this example can all be found under $CCP_STORAGE_PATH/$CCP_NAMESPACE-restore-pitr.