The following bug has been logged on the website:
Bug reference: 13450
Logged by: Leo
Email address: galaxyshih@gmail.com
PostgreSQL version: Unsupported/Unknown
Operating system: Debian
Description:
Hello,
Recently, I am applying point-in-time recovery and I tried to combine it
with full backup(simply tar the whole data). However I encountered some
problems. Here is my scenario:
T1: simply tar a base file(named b1)(/path/to/cluster_data, at this time, I
do not enable WAL)
T2: enable WAL (modify postgres.conf and restart postgres)
T3: tar a base file again (named b2)
T4, T5: insert some data to my database
T6: perform restore (untar the base, overwrite the files to
/path/to/cluster_data, create recover.conf if needed)
Everything is fine if I restore to T2-T5. But here is my problem: if I
restore to T1 first(untar b1 and overwrite the files to
/path/to/cluster_data), then I cannot successfully restore to T2-T5. I will
receive some message in the log:
2015-06-15 20:57:44 GMT+8 LOG: restored log file
"00000001000000000000002E.00000020.backup" from archive
2015-06-15 20:57:44 GMT+8 LOG: restored log file "00000001000000000000002E"
from archive
2015-06-15 20:57:44 GMT+8 LOG: unexpected pageaddr 0/2A000000 in log file
0, segment 46, offset 0
2015-06-15 20:57:44 GMT+8 LOG: invalid checkpoint record
2015-06-15 20:57:44 GMT+8 FATAL: could not locate required checkpoint
record
2015-06-15 20:57:44 GMT+8 HINT: If you are not restoring from a backup, try
removing the file "/home/genie/db_mount_point/backup_label".
2015-06-15 20:57:44 GMT+8 LOG: startup process (PID 30139) exited with exit
code 1
2015-06-15 20:57:44 GMT+8 LOG: aborting startup due to startup process
failure
I do not know the problem exactly. Is that my way of applying it is
completely wrong? I thought that I could use these two ways of restore. When
I try to restore to T1, it should extract the files from b1. The same
procedure applies to restore T2 except the part of replaying the WALs. Can
somebody give me some suggestion, please.