archive_command vs recovery_command paths
archive_command vs recovery_command paths
От:
Craig Ringer <craig@2ndquadrant.com>
Дата:
Hi folks Another point of confusion I've been seeing a lot in users on Stack Overflow, dba.stackexchange.com, etc surrounds the meaning of paths given in archive_command and restore_command. Lots of people seem to assume that they are both relative to the master, and that the master will run the restore_command to fetch archives to send to the replica on request. (Yes, I know that's completely missing the point of archive-based replication, but it seems common). So I think docs changes are needed to the explanations of those options, and to the replication/recovery section, that better explain that we assume there's shared storage like NFS involved, and if there isn't you need to use commands like scp/rsync instead, or use tools like WAL-E. I'm not going to get time to do this one for at least a few days, but I'm posting it now partly so I don't forget about it. -- Craig Ringer http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training & Services
Re: archive_command vs recovery_command paths
On Thu, Feb 13, 2014 at 08:20:02AM +0800, Craig Ringer wrote: > Hi folks > > Another point of confusion I've been seeing a lot in users on Stack > Overflow, dba.stackexchange.com, etc surrounds the meaning of paths > given in archive_command and restore_command. > > Lots of people seem to assume that they are both relative to the master, > and that the master will run the restore_command to fetch archives to > send to the replica on request. > > (Yes, I know that's completely missing the point of archive-based > replication, but it seems common). > > So I think docs changes are needed to the explanations of those options, > and to the replication/recovery section, that better explain that we > assume there's shared storage like NFS involved, and if there isn't you > need to use commands like scp/rsync instead, or use tools like WAL-E. > > I'm not going to get time to do this one for at least a few days, but > I'm posting it now partly so I don't forget about it. I have applied the attached patch which at least clarifies this issue. -- Bruce Momjian http://momjian.us EnterpriseDB http://enterprisedb.com + Everyone has their own god. +