Hi,
I've got following online-backup setup (v. 8.1.8):
- on master
-- archive_command = 'mv %p /<wals_dir>/%f
-- rsyncd with access to DB data and <wals_dir>
- on slave
-- rsync client running every 10sec. to sync <pg_xlog_dir> and <wals_dir>
directories to slave:
rsync -a --delete master::<wals_dir> <slave_wals_dir>
rsync -a --delete master::<pg_xlog_dir> <slave_pg_xlog_dir>
-- pg_start_backup via ssh/rsync DB dir/pg_stop_backup via ssh/old wals
cleanup every hour:
ssh master --command "\"SELECT pg_start_backup('${BKNAME}');\""
rsync -a --exclude=pg_xlog/ --delete main::<db_dir> <slave_db_dir>
ssh master psql --command "\"SELECT pg_stop_backup();\""
-- sql dump every 2 hours:
create LVM snapshot from <slave_db_dir> partition, start postgresql server
with pgdata set to snapshot mountpoint, pg_dumpall, stop, remove snapshot
Setup works perfectly fine, but I' not sure about several things:
1. Is it ok to move instead of copy in archive_command? I use move since
<wals_dir> is on same partition as <pg_xlog_dir>.
2. I just rsync pg_xlog directories instead of hunting down active WAL file
and copying it. Any problems possible because of that?
3. Should I worry about using only "rsync -a" (i.e. checking only mtime,
instead of checksum to see if files differ).
4. during "old wals cleanup" I remove all wal files upto last WAL.backup
(excluding that), afaik thats always ok?
Thanks.