Fix the sqlboiler benchmark

[aarondl]

Did not update the readme numbers since I don't have the same hardware and
don't have the other ORMs set up properly.

BenchmarkKallaxInsertWithRelationships-8      	    1000	   1536213 ns/op	   10991 B/op	     225 allocs/op
BenchmarkSQLBoilerInsertWithRelationships-8   	    1000	   1646870 ns/op	    7328 B/op	     215 allocs/op

This patch significantly speeds up the InsertWithRelationships benchmark. Typically transactions should be being used in calls like this. The API that SQLBoiler has makes it a concious choice when to use them, props to Kallax for having these happen automatically when appropriate.

Also fixes an error where the benchmarks no longer run (go vet now runs automatically as of Go 1.10).

[roobre]

Hello,

Looks fine to me and I certainly can merge this right now, but benchmark results on README.md probably wouldn't get updated until the next release (which honestly can take a while). In order to be fair, I'm perfectly OK if you want to add a note to the benchmark results on README.md (like *This results are outdated as of 2018-05-28 and will get updated soon, or something along those lines).

[aarondl]

Thanks @roobre. I did take the liberty! I hope the changes I've made are acceptable and thanks for being fair about it and allowing me to make the disclaimer.

It was a lot of fun profiling kallax and sqlboiler (did some flame graphs) and also seeing the different approaches we both took in the APIs.

I'd also like to share something I found interesting about the way kallax works too. If you look at it's use of the squirrel cache, that cache is never cleared. Prepared statements use memory on the database and the client and are meant to have Close() called on them when they're finished being used, and in Kallax this grows unbounded depending on how many queries you have in your app. I also found in my benchmarking that it barely affects performance at all, and is almost surely a detriment if you're only executing a query once or infrequently. Food for thought anyway.

Thanks :)

[roobre]

I hope the changes I've made are acceptable and thanks for being fair about it and allowing me to make the disclaimer.

They look indeed good to me. Of course, it's the least I can do :D

[...] that cache is never cleared. Prepared statements use memory on the database and the client and are meant to have Close() called on them when they're finished being used, and in Kallax this grows unbounded depending on how many queries you have in your app.

I was aware of memory being leaked (while investigating #263), but not about the exact place. Thanks for sharing your findings!

[Statement caching] barely affects performance at all, and is almost surely a detriment if you're only executing a query once or infrequently.

That's certainly something I did not know. As I only miantain kallax recently, I don't know why prepared statements where chosen on the first place. Perhaps for the free sql sanitizing? What I do know is squirrel was a temporary dependency, and kallax was meant to generate its own sql in the future. That plan was unfortunately abandoned, since the original authors shifted their efforts away from kallax.

A fix for that memory leak is something that indeed needs addressing. I will try to get some free time and actually work on a fix.

Anyway, I'm merging this to get the README updated. I'll hopefully rerun the benchmarks and get them a full refresh soon. Thanks for your effort, @aarondl!

[aarondl]

Thanks for the merge. A couple more notes on caching/performance/prepared statements:

I benchmarked 10,000 inserts with a for loop, prepared statements, and a transaction. They took 8.6s, 8.2s and 800ms respectively. I also have a hunch that the sql library itself uses prepared statements for every single query by default for the sql sanitizing, so creating one manually doesn't affect performance much. It seems a bit improbable so I just re-checked and those results are consistent.

There's a way to turn off the cache so you can test easily https://godoc.org/gopkg.in/src-d/go-kallax.v1#Store.DisableCacher (not sure if you know since you aren't one of the original authors) which affects this: https://github.com/src-d/go-kallax/blob/v1.3.4/store.go#L138