You write:
Each customer can have multiple sites, but only one should be
displayed in this list.
Yet, your query retrieves all rows. That would be a point to optimize. But you also do not define which site is to be picked.
Either way, it does not matter much here. Your EXPLAIN shows only 5026 rows for the site scan (5018 for the customer scan). So hardly any customer actually has more than one site. Did you ANALYZE your tables before running EXPLAIN?
From the numbers I see in your EXPLAIN, indexes will give you nothing for this query. Sequential table scans will be the fastest possible way. Half a second is rather slow for 5000 rows, though. Maybe your database needs some general performance tuning?
Maybe the query itself is faster, but "half a second" includes network transfer? EXPLAIN ANALYZE would tell us more.
If this query is your bottleneck, I would suggest you implement a materialized view.
After you provided more information I find that my diagnosis pretty much holds.
The query itself needs 27 ms. Not much of a problem there. "Half a second" was the kind of misunderstanding I had suspected. The slow part is the network transfer (plus ssh encoding / decoding, possibly rendering). You should only retrieve 100 rows, that would solve most of it, even if it means to execute the whole query every time.
If you go the route with a materialized view like I proposed you could add a serial number without gaps to the table plus index on it - by adding a column row_number() OVER (<your sort citeria here>) AS mv_id.
Then you can query:
SELECT *
FROM materialized_view
WHERE mv_id >= 2700
AND mv_id < 2800;
This will perform very fast. LIMIT / OFFSET cannot compete, that needs to compute the whole table before it can sort and pick 100 rows.
pgAdmin timing
When you execute a query from the query tool, the message pane shows something like:
Total query runtime: 62 ms.
And the status line shows the same time. I quote pgAdmin help about that:
The status line will show how long the last query took to complete. If
a dataset was returned, not only the elapsed time for server execution
is displayed, but also the time to retrieve the data from the server
to the Data Output page.
If you want to see the time on the server you need to use SQL EXPLAIN ANALYZE or the built in Shift + F7keyboard shortcut or Query -> Explain analyze. Then, at the bottom of the explain output you get something like this:
Total runtime: 0.269 ms
Best Answer
This seems like a rather esoteric thing, which is probably why it doesn't get recommended all that much. I do recommend it when the issue comes up, which just isn't all that often. The size of one index is unlikely to be all that meaningful in the context of an entire database, so making it smaller usually isn't worth worrying about a great deal.
I don't think there are any downsides which are specific to this being a foreign key. The automatically-generated queries that are used to maintain the many-valued side of the constraint can use the GIN index just as well as the Btree index.
I'd turn "fastupdate" to off, unless you do relevant benchmarking which shows you want it on. Since the index is over a scalar using "btree_gin", you don't get the explosion in index page writes for each inserted row, like you do with a normal GIN index, so the need for fastupdate is less.
Since this index type is much less used than regular btree indexes, there are more likely to be undiscovered bugs lurking in it (especially if fastpdate = on). I wouldn't (and don't) let that put me off from using it if it really provided me with something of value, but it keeps me from just blindly using "btree_gin" everywhere I conceivably could.
One thing I've noticed is that the replay of updates/inserts to GIN indexes seem pretty slow (compared to what I naively expected), which could be relevant for PITR, recovery, or streaming replication. If any of those things are important to you, make sure you include them in your test.