Re: [ADMIN] pg_basebackup recovery failure

On 3/7/17 3:36 PM, Magnus Hagander wrote:

On Tue, Mar 7, 2017 at 12:31 PM, Ray Stell <stellr@vt.edu> wrote:

On this doc page are some command examples for pg_basebackup:

        https://www.postgresql.org/docs/current/static/app-pgbasebackup.html

such as: 

    pg_basebackup -h mydbserver -D /usr/local/pgsql/data

I've never gotten this to work.  When I bring up the copy on an intended standby system I always get the classic:
PANIC:  could not locate a valid checkpoint record

Failing that, I backed up with pg_start_backup/tar/pg_stop_backup as an alternative, but I was always curious what I was doing wrong. 
The wiki page on setting up replication:    https://wiki.postgresql.org/wiki/Streaming_Replication

says use the command with this options:      pg_basebackup -h 192.168.0.10 -D /srv/pgsql/standby -P -U replication --xlog-method=stream

and I stumbled into option:
     --checkpoint=fast

The standby was able to recover when I used one of these, I don't remember which, maybe both. 

Should the doc be updated or am I missing something?    

If you are using pg_basebackup without a log archive, then yes, you need --xlog-method=stream (or fetch). The checkpoint method only affects how quickly the checkpoint runs, it will always run.

If you do have a log archive, it should work fine without --xlog=stream, but you need to put your restore_command into recovery.conf manually.

Too funny not to post.   Years ago, maybe version 7 or 8 something, I remember needing to rm the backup_label from the restored/standby dir.  I did that without thinking and it seems that was the source of the problem.   I just repeated the process with a mv of the backup_label and I get a:     PANIC:  XX000: could not locate a valid checkpoint record I restored the backup_label file and we're happy:       LOG:  00000: started streaming WAL from primary at 0/15000000 on timeline 1 Lesson: I need to get out more often, but I must say, the error msg gave me a pretty good head fake.