Thread: The Problem of Applying Point-in-time Recovery

I am not sure if is it proper to post my problem here. If not, please forgive my ignorance and tell me where should I post to.

Recently, I am applying point-in-time recovery with Debian and postgres 8.3 (due to some reason, I have no chance to upgrade) but I encountered some problems. I followed the instruction from the official document step by step. 

1. First, I modified the postgres.conf to enable WAL arching and restart the postgres. 

2. Then I simply tar the whole data in the cluster data directory, ${PG_DATA} to be the base backup. During this step, I called pg_start_backup('label') and pg_stop_backup() before and after the tar procedure separately. 

3. After that, I inserted some data into the database. 

4. Next, I simulated that the database is corrupted and need to perform recover.

    4.0. stop postgres

    4.1. moved the WALs from ${PG_DATA}/pg_xlog to another directory

    4.2. untared the  base backup and moved the data to ${PG_DATA} (overwrite it)

    4.3. created recovery.conf, following is my configuration: (Note, I stored the WALs to a remote host)

               restore_command = 'rsync -a host_user@host_ip:/path/to/remote/host/wal/%f %p'

               recovery_target_time = 'YYYY-mm-dd HH:MM:SS'

               recovery_target_timeline = 'value'          

    4.4. restarted postgres

At first, everything was fine. I could perform recover successfully. I could see from log that postgres did restore the WALs and I could see the data which i inserted in step 3 in database, too. But when I performed recover repeatedly (that is I repeatedly performed from step 4.0 to step 4.4). I got very high possibility that postgres could fail to recover. Here is the error message: 

I do not know what caused the problem exactly. Is the problem happened because I performed recover repeatedly? Please give me some suggestion.

Yours faithfully