pg_dump and restore a db in a container

Tue, 20 Dec 2016 11:00:00 +0000

I’m running containerized ManageIQ appliances for development. To dump your postgres db running inside such a container you simply exec pg_dump. In my case not directly, because pg_dump is a some obscure place inside the container.

docker exec miq_latest bash -lc 'pg_dump --format custom vmdb_production' > /tmp/pg_dump.pgdata

Start a vanilla postgres container to host your development db:

docker run --name miq_pg -e POSTGRES_USER=root -e POSTGRES_DB=vmdb_development -d -p 5432:5432 postgres:9.5

Now you can restore the dump

docker exec -i miq_pg pg_restore --dbname=vmdb_development --verbose --clean < /tmp/pg_dump.pgdata

Beware Mr. T

You might be used to run docker exec -ti - but don’t do this to dump a db. The -t, --tty flag will also append a carriage return character.

❯ head /tmp/pg_dump.pgdata_dos | head -1 | cat -A
PGDMP^A...@9.5.2^@M-r^M$

❯ head /tmp/pg_dump.pgdata | head -1 | cat -A
PGDMP^A...@9.5.2^@M-r$

Better yet, pg_restore will just segfault on the wrong format

❯ docker exec -i miq_pg pg_restore --dbname=vmdb_development --verbose --clean < /tmp/pg_dump.pgdata_dos

❯ echo $?
139

❯ docker run --rm -ti --link=miq_pg -v /tmp:/tmp postgres:9.5 bash
root@9ce662730c82:/# pg_restore --dbname=vmdb_development --verbose --clean /tmp/pg_dump.pgdata_dos
Segmentation fault

init

Wed, 14 Dec 2016 16:08:04 +0000

cat /dev/urandom | hexdump