Skip to content

Latest commit

 

History

History
54 lines (48 loc) · 1.92 KB

RECOVERY.md

File metadata and controls

54 lines (48 loc) · 1.92 KB
Raw

Free space requirements:

  1. For base backup itself, you can check size with barman show-backup

  2. Space for WALs

    • Take into account that barman stores WALs in compress form. You can calcuate required space by checking once again barman show-backup. There you will find disk usage and compression rate.
    • During recovery barman put restore_command = cp... which means that you have to have at least double of WAL size. Upod recovery completion duplicates wil be removed.

Recovery steps:

  1. Stop pg cluster, but retain containers. IMPORTANT start with slaves 
    touch /var/run/recovery.lock
killall repmgrd
gosu postgres pg_ctl stop
  2. Check sshd is running on master, if not start it: 
    SSH_ENABLE=1 /home/postgres/.ssh/entrypoint.sh
  3. Connect to barman container and select appropriate base backup 
    barman list-backup all
  4. Recover to required point, for example: 
    barman recover pg_cluster 20170815T041301 /var/lib/postgresql/data/ --target-time "2017-08-15 04:11:26.5" --remote-ssh-command="ssh pgmaster" -j$(nproc)
  5. Connect to master node, start recovery and check DB consistency, then finish recovery 
    gosu postgres pg_ctl start -o "-c listen_addresses='localhost'"
psql -U monkey_user monkey_db -c "SELECT * from t"
psql -U postgres -c "SELECT pg_xlog_replay_resume()"
  6. Restart master container 
    rm /var/run/recovery.lock
  7. Wait until master will become fully functional and restart every standby node 
    rm /var/run/recovery.lock
  8. Recreate replication slot for barman 
    barman receive-wal --create-slot $UPSTREAM_NAME

Caveat

You need manually rewind 2nd-tier replicas to master

gosu postgres pg_rewind --target-pgdata /var/lib/postgresql/data/ --source-server 'host=pgmaster user=replication_user password=replication_pass dbname=replication_db'