Re: Removing unneeded self joins

Andrei Lepikhov <a.lepikhov@postgrespro.ru>

From: Andrei Lepikhov <a.lepikhov@postgrespro.ru>

To: Robert Haas <robertmhaas@gmail.com>, Alexander Korotkov <aekorotkov@gmail.com>

Cc: Tom Lane <tgl@sss.pgh.pa.us>, Richard Guo <guofenglinux@gmail.com>, Alexander Lakhin <exclusion@gmail.com>, "Gregory Stark (as CFM)" <stark.cfm@gmail.com>, Michał Kłeczek <michal@kleczek.org>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>

Date: 2024-05-05T02:45:47Z

Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

Remove GUC_NOT_IN_SAMPLE from enable_self_join_elimination
- 717d0e8dd945 18.0 landed
Put enable_self_join_elimination into postgresql.conf.sample
- c2d329260cd8 18.0 landed
Get rid of ojrelid local variable in remove_rel_from_query()
- e167191dc146 18.0 landed
Implement Self-Join Elimination
- fc069a3a6319 18.0 cited
Revert: Remove useless self-joins
- d1d286d83c0e 17.0 landed
Replace lateral references to removed rels in subqueries
- 466979ef031a 17.0 landed
Replace relids in lateral subquery parse tree during SJE
- 489072ab7a9e 17.0 landed
Forbid SJE with result relation
- 8c441c082797 17.0 landed
Fix misuse of RelOptInfo.unique_for_rels cache by SJE
- 30b4955a4668 17.0 landed
Replace the relid in some missing fields during SJE
- a7928a57b9f0 17.0 landed
Revert 56-bit relfilenode change and follow-up commits.
- a448e49bcbe4 16.0 cited
Stabilize timetz test across DST transitions.
- 4a071afbd056 14.0 cited
Speed up finding EquivalenceClasses for a given set of rels
- 3373c7155350 13.0 cited
Fix mark-and-restore-skipping test case to not be a self-join.
- 24d08f3c0a1f 12.0 landed

On 3/5/2024 20:55, Robert Haas wrote:
> One of my most embarrassing gaffes in this area personally was
> a448e49bcbe40fb72e1ed85af910dd216d45bad8. I don't know how I managed
> to commit the original patch without realizing it was going to cause
> an increase in the WAL size, but I can tell you that when I realized
> it, my heart sank through the floor.
I discovered this feature and agree that it looks like a severe problem.
Unfortunately, in the case of the SJE patch, the committer and reviewers 
don't provide negative feedback. We see the only (I'm not sure I use the 
proper English phrase) 'negative feelings' from people who haven't 
reviewed or analysed it at all (at least, they didn't mention it).

Considering the situation, I suggest setting the default value of 
enable_self_join_removal to false in PG17 for added safety and then 
changing it to true in early PG18.

Having no objective negative feedback, we have no reason to change 
anything in the design or any part of the code. It looks regrettable and 
unusual.

After designing the feature, fixing its bugs, and reviewing joint 
patches on the commitfest, the question more likely lies in the planner 
design. For example, I wonder if anyone here knows why exactly the 
optimiser makes a copy of the whole query subtree in some places. 
Another example is PlannerInfo. Can we really control all the 
consequences of introducing, let's say, a new JoinDomain entity?

You also mentioned 2024.pgconf.dev. Considering the current migration 
policy in some countries, it would be better to work through the online 
presence as equivalent to offline. Without an online part of the 
conference, the only way to communicate and discuss is through this 
mailing list.

-- 
regards,
Andrei Lepikhov