Just let Discourse handle it. This requires 3 times the disk space. So if your DB is 100GB, you would need an additional 200GB free to do the upgrade. Obviously a huge problem for people with large installs.
Follow their “manual update” procedure. This requires 2 times the disk space, so if your DB is 100GB you would need an additional 100GB free. This is also a big problem for some.
In this post, @Falco suggested using the --link option to do the upgrade in-place using hard links. The docker container they suggest using supports that argument, but Discourse devs don’t suggest using it in the post.
So my question is this, should option 3 be:
Run the command below, which will require a very small amount of additional disk space. So if your DB is 100GB, it might require, say, an additional 10GB? And if so, is this a recommended procedure by the Discourse devs, and has anyone actually done it before and lived to tell the tale?
New command to upgrade in-place:
docker run --rm \
-v DIR:/var/discourse/shared/standalone/postgres_data:/var/lib/postgresql \
tianon/postgres-upgrade:12-to-13 \
--link
Compared to the old command to upgrade into a new directory (requiring double the space):
P.S.: I would have just replied to that PG13 upgrade thread, but it deletes posts after 7 days. Why do you have it configured that way? I know there was a lot of discussion when this first came up that would have been useful for reference.
If they have, they didn’t mention it here. Mostly instrucions here try to be as foolproof as possible and require as little system adminstration knowledge as possible. Most people here woud rather do something the safest, most tested way possible than some way designed to save a very few dollars.
If it works for you, you can update PostgreSQL 13 update accordingly, but before you do, do you feel comfortable recommending to someone who doesn’t know what bash is that they do it that way? You’re sure that it won’t hose their database and their site will be ruined forever?
The idea is that if some other good information is presented that it be added to the OP rather than asking people to read through year’s worth of posts that are likely to be unhelpful or wrong.
No I’m not sure, I don’t have much experience with postgres and was hoping one of the discourse devs could provide some assurances it would work.
Even if it does work I also wouldn’t recommend it as the default upgrade procedure as the old way keeps a separate copy of the DB for rollback. If it works it would be a great option for space-constrained environments though.
Another easy way is to spin up a new server, migrate the data, and turn off the old one. If you must use the old one, do the upgrade on a temporary server, so a fresh install on the original server (which probably needs an OS upgrade) and move it back.
That’s safe, easy, and well documented. Hundreds of people have done that.
Yes, but that would take a day or two. During that time we could either a) tell users their posts during this period will be lost or b) set the forum read-only. Neither is a great solution.
I don’t think that the server would be down a whole lot longer than during the rebuild. And if you move to the new server and stay there, you can leave the old server in read only mode while you make the move. If downtime is your concern then moving to a new server will be much, much better.
We have a pretty big forum, but I’ve never tried restoring a backup so I don’t know how long it would take. We would indeed stay on the new host if we did it. I would like to avoid that due to the extra work/annoyance if possible.
Another way to solve the upgrade with limited space issue is to make a backup, rm - r the postgres directory, rebuild, and then restore the backup. I did this on a site last week.
Nope, never got it upgraded. Deleting the DB and restoring the backup sounds pretty risky. We need the upgrade in-place to work, basically.
We’re running ubuntu 18.04 which hits desupport in 2023 so I figure at that point we’ll have no choice but to migrate to a new host anyway, and plan to bite the bullet then, build-out a new host running 22.04 LTS, and restore from backup.
Hmm. Could be a wash. I think with the backup model one of the copies is compressed,which could make a difference? And the site that I did it on had backups on S3. And it was a test site so the stakes were low if there was a problem.
Except backups are used lots more often in lots more situations than the upgrade in place. I consider it much safer.
Perhaps, but I don’t have a lot of expertise with postgres and don’t feel comfortable doing it. Restoring the whole site from backup on an entirely different VM, that I do feel comfortable doing, however it would mean losing posts for however many hours is takes to restore so I’m not super enthusiastic about that either. But since 18.04 is being desupported I won’t have much choice next year.
Unless your database is tens of gigabytes, it won’t take hours. And you’ll put the forum in read only mode before you backup and restore, so you won’t lose any posts. It’s not that hard to do with virtually no down time, just read only time.
root@forum-app:/shared/postgres_data# du -sh
97G .
I wouldn’t put it in read-only, I would put up a banner telling people their posts today are ephemeral. Better to let them chat about it even if those posts would be lost, in my view.
By then you will have access to built-in Discourse chat as well, that’s a feature which will ship in 2.9 (possibly default off, but in beta and supported for use).