Switch to a custom allocator by default

[LesnyRumcajs]

Issue summary

Once we have the benchmark results comparing mimalloc, jemalloc and the default allocator, we should decide which one to use by default. Most likely, based on the feedback from @elmattic and @hanabi1224 it would be one of the custom ones.

Other information and links

[hanabi1224]

Here's the snapshot import benchmark result I got on DO droplet

branch	allocator	db size	peak rss	time
hm/track-peak-rss	default	100GB	12.02GB	4485s
hm/track-peak-rss	mimalloc	100GB	10GB	4909s
hm/track-peak-rss	jemalloc	100GB	8.92GB	3873s

Block validation metrics on mainnet

branch	allocator	total time	total blocks	avg
main	default	5384.7s	271	19.87s
main	mimalloc	8174.3s	435	18.8s
main	jemalloc	7042.6s	371	18.98s

[hanabi1224]

So let's ship jemalloc? @elmattic @LesnyRumcajs @lemmih

[lemmih]

Sounds good to me.

[LesnyRumcajs]

Yeah, sounds reasonable.

[elmattic]

@hanabi1224 what about validation time?

[hanabi1224]

@elmattic I did not run validation but I think the bottleneck of validation is not database but fvm

[elmattic]

Agree, but memory allocator should have an impact on the fvm, so validation time too, no?

[hanabi1224]

@elmattic OK, I can collect BLOCK_VALIDATION_TIME metrics with different allocators

[elmattic]

I will run the ruby benchmark script on my side on a Fedora 37.

[hanabi1224]

I will run the ruby benchmark script on my side on a Fedora 37.

@elmattic Sounds good, thx! At the same time, I will run forest with different allocators and fetch metrics with the below command

wget -O - http://localhost:6116/metrics | grep apply_blocks_time_ | tail -n 2

# output
apply_blocks_time_sum 163.53027957700002
apply_blocks_time_count 6

[hanabi1224]

@elmattic block validation metrics updated, no obvious gain or loss is observed.