El 21/1/20 a las 10:14, Laurenz Albe escribió:
On Mon, 2020-01-20 at 14:59 +0100, Ekaterina Amez wrote:
PS: Just in case anyone wants to know, this is part of the process of upgrade a server with 9.2 version
that has no free space in PGDATA and that can't be stopped for much time. After asking here, the strategy
to upgrade will be: replicate this DB to a path with space to grow, switch clusters (slave becoming master
and creating a new slave to have just-in-case), and pg_upgrade slave/new master with --link option.
Do not upgrade to 9.6, upgrade to at least v11.
v9.6 is target version because this is the last server that remains with
unsupported PG version, and the others have been upgraded to 9.6. After
having all of them in 9.6, I have to discuss what to do next as at least
one of the servers is installed over CentOS 32 bits (facepalm...)
Set "wal_keep_segments" high enough on the primary for safety.
I'm not sure how to figure this value . Of course I could use a big
value and forget, but as the server is running out of space I'd like to
adjust it so there is (almost) no waste of space. This is why I began to
think about wal archiving, though it seems a bit more complicated.
Run "pg_basebackup" on the standby.
Create a "recovery.conf" on the standby that has "primary_conninfo" set
and "standby_mode = on".
Start the standby and wait until it has caught up, then switch over.
Yeah, I've been testing this part and it's the only part where I feel I
know what I'm doing.
Yours,
Laurenz Albe
Regards,
Ekaterina
