Thread: New pg_upgrade data directory inside old one?
Someone on IRC reported that if they had run the pg_upgrade-created delete_old_cluster.sh shell script it would have deleted their old _and_ new data directories. (Fortunately they didn't run it.) I was confused how this could have happened, and the user explained that their old cluster was in /u/pgsql/data, and that they wanted to switch to a per-major-version directory naming schema, so they put the new data directory in /u/pgsql/data/9.5. (They could have just moved the directory while the server was down, but didn't.) Unfortunately, there is no check for having the new cluster data directory inside the old data directory, only a check for tablespace directories in the old cluster. (I never anticipated someone would do this.) The attached patch adds the proper check. This should be backpatched to all supported Postgres versions. -- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB http://enterprisedb.com + As you are, so once was I. As I am, so you will be. + + Roman grave inscription +
Attachment
On Mon, Feb 15, 2016 at 6:29 PM, Bruce Momjian <bruce@momjian.us> wrote:
Someone on IRC reported that if they had run the pg_upgrade-created
delete_old_cluster.sh shell script it would have deleted their old _and_
new data directories. (Fortunately they didn't run it.)
I was confused how this could have happened, and the user explained that
their old cluster was in /u/pgsql/data, and that they wanted to switch to
a per-major-version directory naming schema, so they put the new data
directory in /u/pgsql/data/9.5. (They could have just moved the
directory while the server was down, but didn't.)
Unfortunately, there is no check for having the new cluster data
directory inside the old data directory, only a check for tablespace
directories in the old cluster. (I never anticipated someone would do
this.)
Interesting - I definitely wouldn't have expected that either. And it definitely seems like a foot-gun we should protect the users against.
On Mon, Feb 15, 2016 at 06:32:06PM +0100, Magnus Hagander wrote: > > > On Mon, Feb 15, 2016 at 6:29 PM, Bruce Momjian <bruce@momjian.us> wrote: > > Someone on IRC reported that if they had run the pg_upgrade-created > delete_old_cluster.sh shell script it would have deleted their old _and_ > new data directories. (Fortunately they didn't run it.) > > I was confused how this could have happened, and the user explained that > their old cluster was in /u/pgsql/data, and that they wanted to switch to > a per-major-version directory naming schema, so they put the new data > directory in /u/pgsql/data/9.5. (They could have just moved the > directory while the server was down, but didn't.) > > Unfortunately, there is no check for having the new cluster data > directory inside the old data directory, only a check for tablespace > directories in the old cluster. (I never anticipated someone would do > this.) > > > Interesting - I definitely wouldn't have expected that either. And it > definitely seems like a foot-gun we should protect the users against. Patch applied back through 9.3 where delete script tests were added. -- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB http://enterprisedb.com + As you are, so once was I. As I am, so you will be. + + Roman grave inscription +