Time taken by VACUUM FULL to reclaim space

Question

After receiving some insightful guidance in a previous post here, I'm going to run VACUUM FULL on 4 PostgreSQL 9.3.10 tables. Table sizes are:

1) links_publicreply: ~30M rows, 9 columns, 3 indexes (types: int, timestamp, char, bool)

2) links_reply: ~25M rows, 8 columns, 6 indexes (types: int, text, timestamp, char)

3) links_link: ~8M rows, 14 columns, 3 indexes (types: int, text, dbl precision, timestamp, char bool)

4) links_user_sessions: ~2M rows, 7 columns, 4 indexes (types: int, text, timestamp, inet)

This is my first attempt at reclaiming disk space. It's a busy server of a local social networking website. No time is actually "downtime". But the least busy is ~4:00 AM, so that the window I'll use.

Speaking from experience, can you guys form any opinion on how long VACUUM FULL would take for the 4 tables I pointed out? I'd like to put up a "under maintenance till xx:xx:xx" message on the website while it's happening. I know no one can be sure, but is this deterministic enough for you to form a ballpark opinion?

Secondly, just so that we're on the same page, the commands I'd be running on psql are simply VACUUM (FULL, VERBOSE, ANALYZE) link_publicreply; (and so on), correct? Don't want to screw it up.

Answer 1

It's a busy server of a local social networking website. No time is actually "downtime".

So VACUUM FULL is going to be a problem since it takes an exclusive lock on each table it processes. Consider the community tool pg_repack instead which achieves the same without exclusive locks.

Related:

• VACUUM returning disk space to operating system

None of this affects the size of backups, since those do not include dead rows to begin with.