* Every optimization problem needs data on costs, and the more and better the data is, the better. Postgres has made some improvements here, especially with cross column statistics, but there are still massive improvements left on the table. The most glaring omission is data on syscall latencies. Reading a page from disk has dramatically different latencies from system to system, and postgres still relies on configuration values for these costs, when it could very easily be measuring them. Another omission is foreign key statistics. Joins along foreign keys should never have a bad plan, but they still occasionally do.

* Deferred or alternative scenario planning should be adopted, especially for large and expensive queries. As it is today, your plan is finalized before it is executed, even though earlier stages of execution could provide information (like rowcounts or cardinality estimates) that could dramatically improve later stage plans.

* Machine learning most definitely could be an area where improvements could be made, but I've been unimpressed with the efforts that Ive seen. Don't use machine learning for planning, use machine learning for cost discovery and estimation. Build better cost models, and then let the optimization engine work with that data.

In terms of the deferred / alternative planning, do you think adaptive query execution is a reasonable way to achieve this? It certainly allows for information early in the query execution to impact later plans. My worry with these approaches is that if you get the first couple of joins wrong (which is not uncommon), unless you have something like Yannakakis/SIPs, you still can't recover.

I am obviously biased on the whole "ML for query optimization" thing. One thing I would note is that every "ML for planning" approach I've seen does, under the hood, use ML for cost discovery/estimation. These approaches are just trying to balance the data they collect (exploration) with the quality of the plans they produce (exploitation). Interestingly, if you use ML in a way that is completely removed from planning, you actually get worse query plans despite more accurate estimates: https://people.csail.mit.edu/tatbul/publications/flowloss_vl... (again, I've got a horse in this race, so my opinion should come with a side of salt :D)

I don't want to be tuning this, it takes a lot of time to do it properly and I haven't retested this in a long time. I just set something that has worked in the past which is probably bad.

I completely agree that Postgres should be able to figure this out on its own. This is just an example, there are more such parameters which should be adjusted to the hardware but most people will leave the defaults.

Alternative planning sounds awesome. I saw a query plan the other day where it expected about 1000 rows from some subquery so did a nested loop to an index scan. In reality there’s about a billion rows. I’m not sure yet why the estimate was so bad, but being able to pivot from a nested loop to a hash join if the row count crossed some threshold would be great at avoiding some catastrophic plans.

I'm curious what you mean. Postres, like many RDBMSs, does not automatically create an index on a FK. But you have to know that already.

Is it join sequencing?

It is also true that measuring cold cache and warm cache performance can produce different results, and this experiment is certainly in the warm cache scenario. But, the cold cache scenario suffers from the problem you mention as well: an improvement to PG's B-tree that saves a few IOs will dominate any kind of CPU-based improvement (at least at the data size of the join order benchmark).

FWIW, the plan for the query with the P90 latency changes from a plan that uses loop and merge join in PG8.4 to a plan that uses hash join in PG16 (where it is no longer the P90 query), which is at least some evidence of optimizer improvements.

I don't think that it's possible to test optimizer quality in isolation -- not really, not if it's to be in any way practical. Many (most?) individual improvements to the optimizer are hopelessly intertwined with executor enhancements. This is usually fairly obvious, because the same commit changes both the optimizer and the executor together. Sometimes it's much less obvious, though, because its more a case of the optimizer and executor coevolving.

It's probably still the case that a lot of the improvements seen here are pure executor enhancements, but I can't say that I'm very confident about that. (As the main person behind those B-Tree IO savings you might expect me to have more confidence than that, but I find it horribly difficult to tease these issues apart while keeping the discussion high level/relevant.)

Turns out homebrew installed postgres without JIT support and one query on the developer machine ran in 200ms, while with JIT the query took 4-5 seconds. Took some time to figure out, since I'm not a heavy postgres user, but since then I've always disabled JIT and never looked back.

You can configure thresholds for the JIT activation in PostgreSQL as well if you want to elevate the bar from which the JIT is enabled.

It would probably be less detrimental if it could perform asynchronous compilation for future queries (which really is closer to how normal JITs actually behave, at least for optimising backends)

The inline blocking compilation step will still fuck your query.

> also add hints please

I’d much rather they allowed bypassing the planner entirely and had an interface to feed plans directly to the executor.

Only on the first execution, the long plan/jit time is usually only an issue when running a fast query over and over again.

However if plans where cached then you could also plan without jitting then background jit and update cached plan for next time which would be even smarter.

>I’d much rather they allowed bypassing the planner entirely and had an interface to feed plans directly to the executor

Other database have plan locking where you can load an existing plan from somewhere else in serialized form (XML) and force it to use that. Most of the time I prefer just a few hints in the query source solves almost all issue with the planner.

More seriously, we care a lot about the format of the 8k pages to ease upgrade paths as much as possible.

The main reason I did this was to reduce the number of versions I had to test, but I agree a more complete analysis would test each (true) major version.

Imagine what environmental impact you would have if you optimized Python's performance by 1%? How much CO2 are you removing from the atmosphere? It's likely to overshadow the environmental footprint of you, your family and all your friends combined. Hell, maybe it's the entire city you live in. All because someone spent time implementing a few bitwise tricks.

I imagine this would also increase the complexity of the python intepreter by more than 1%, which in turn would increase the number of bugs in the interpreter by more than 1%. Which would burn both manpower and CPU-Cycles on a global scale.

(I assume that optimization, that reduce complexity are already exhausted, e.g. stuff like "removing redundant function calls". )

Then if you actually go ahead and check, it turns out it's not true! It's quite a shocking revelation.

When you dig into the popular compilers/runtimes (with the exception of things like LLVM)

Many of them still have low hanging fruit of the form:

a = b + c - b

Yes, the above is still not fully optimized in the official implementations of some popular programming languages.

Also an optimization of "removing redundant function calls" isn't a binary on/off switch. You can do it better or worse. Sometimes you can remove them, sometimes not. If you improve your analysis, you can do more of that and improve performance. Same for DSE, CSE, etc...

Is it because you view the 15% to be a low number? Because it really, really, isn’t in the context. It’s certainly smaller than the 60% from your linked law especially if you do the whole 15/10 thing, but you really shouldn’t compare Postgres’s performance to hardware increases because they aren’t the same. You would need absolutely massive amounts of hardware improvements to compare to even 1% performance on what is being measured here.

I don’t think the law is as silly as other posters here, but it’s talking about programming language compile times. I wouldn’t compare something as unimportant as that to what is arguably one of the most important things in computer science… considering how much data storage and consumption means to our world.

Murphy’s law is the only comparison given. I wonder what the comparative expense is to develop faster and faster hardware, vs. to keep improving compilers. Depending on how the ROIs compare, (in dollars-per-percent gain, let’s say) maybe this “law” is worth some weight.

OTOH, the Postgres article in question seems to show diminishing returns from optimization, which belies the whole premise of the “law” which assumes the gain is consistent from year to year. And might prove Prorbstig’s implied point about optimization being a bad investment over the long term!

"Tail latency has improved by YMMV for everything else" => yes, I think that's a valid (but conservative) read. Of course, in many (most?) applications, tail latency is very important. Tail latency also tends to be the thing that optimizer engineers target (i.e., reduce the runtime of the longest running query).

From what I gather this is mostly because various different SQL queries might get transformed into the same "instructions", or execution plan, but also because SQL semantics doesn't leave much room for language-level optimizations.

As a sibling comment noted, an important decision is if you can replace a full table scan with an index lookup or index scan instead.

For example, if you need to do a full table scan, and do significant computation per row to determine if a row should be included in the result set, the optimizer might change the full table scan with a parallel table scan then merge the results from each parallel task.

When writing performant code for a compiler, you'll want to know how your compilers optimizer transforms the source code to machine instructions, so you prefer writing code the optimizer handles well and avoid writing code the optimizer outputs slower machine instructions for. After all the optimizer is programmed to detect certain patterns and transform them.

Same thing with a query optimizer and its execution plan. You'll have to learn which patterns the query optimizer in the database you're using can handle and generate efficient execution plans for.

A lot more can be done: choosing the join order, choosing the join strategies, pushing the filter predicates at the source, etc. That's the vast topic of SQL optimization.

Basically, it's theoretically possible that some old ass system running version 8 runs it faster than your current system runs version 8 (excluding hardware).

Add to that all the other usual caveats about benchmarks etc (like the fact that IO wasn't tested here but is pretty dang relevant to real world performance)