Обсуждение: Changing column order through dump and restore
Hello All,
I have a database in which I want to re-order some of the columns to bring it into line with our standard schema. The problem is that, although the database does not have too much data, the tables I want to change have tons of dependencies on top of them (mostly layered views). First I tried to remove those dependencies, recreate the tables and then put them back again, but I started to get into a mire with all the dependencies. So to avoid having to work out all those dependencies, I thought perhaps I could do it instead by dumping the whole schema and data separately as SQL files, then tweaking CREATE TABLE commands in the schema file, then reloading schema into a fresh database and then finally reloading the data. This way I think the dump and restore ought to take care of the dependencies for me. Currently my script looks like this:
# Dump the database (schema only)
pg_dump -s my_database > my_database_schema.dmp
# Dump the database (data only) as insert statements and with commands to disable triggers
pg_dump -a --disable-triggers --column-inserts my_database > my_database_data.dmp
# Change the order of the columns in the relevant CREATE TABLE commands
vim my_database_schema.dmp
# Create a fresh database
createdb my_new_database
# Restore the corrected schema into the new database
psql my_new_database < my_database_schema.dmp > schema_import.log 2>&1
# Restore the data into the new database
psql my_new_database < my_database_data.dmp > data_import.log 2>&1
I understand the data restore will be slow (due to the inefficiency of the INSERT statements), but is this a valid way to go about column re-ordering? Will this achieve what I want (no more, no less)? If so, is there anything else I need to consider or beware of?
I would be grateful for any advice and/or reassurance. I am using PG 9.2.4 (upgrade scheduled soon!)
Regards,
Alex.
Alex Stanier <Alex.stanier@proactis.com> writes:
> # Dump the database (schema only)
> pg_dump -s my_database > my_database_schema.dmp
> # Dump the database (data only) as insert statements and with commands to disable triggers
> pg_dump -a --disable-triggers --column-inserts my_database > my_database_data.dmp
> I understand the data restore will be slow (due to the inefficiency of
> the INSERT statements), but is this a valid way to go about column
> re-ordering?
FWIW, you shouldn't need to use --column-inserts mode, because even in
default COPY output, pg_dump emits column name lists in the COPY commands.
So the right things should happen when reloading into a database with
modified column order.
Haven't thought about it in great detail, but your plan sounds generally
sane otherwise.
regards, tom lane
> -----Original Message----- > From: Tom Lane [mailto:tgl@sss.pgh.pa.us] > Sent: 27 June 2016 21:56 > To: Alex Stanier <Alex.stanier@proactis.com> > Cc: pgsql-admin@postgresql.org > Subject: Re: [ADMIN] Changing column order through dump and restore > > Alex Stanier <Alex.stanier@proactis.com> writes: > > # Dump the database (schema only) > > pg_dump -s my_database > my_database_schema.dmp # Dump the database > > (data only) as insert statements and with commands to disable triggers > > pg_dump -a --disable-triggers --column-inserts my_database > > > my_database_data.dmp > > > I understand the data restore will be slow (due to the inefficiency of > > the INSERT statements), but is this a valid way to go about column > > re-ordering? > > FWIW, you shouldn't need to use --column-inserts mode, because even in > default COPY output, pg_dump emits column name lists in the COPY commands. > So the right things should happen when reloading into a database with > modified column order. > > Haven't thought about it in great detail, but your plan sounds generally > sane otherwise. > > regards, tom lane Great thank you. Useful to know about the COPY; it has speeded the process up dramatically. Regards, Alex Stanier.