Experiment/use mashumaro

[drewbanin]

Draft PR - do not merge this.

Example script to assess perf gains:

import dbt
import dbt.adapters.postgres.relation
import time

Relation = dbt.adapters.postgres.relation.PostgresRelation

N = 10000

t_create = 0
t_to_dict = 0
t_from_dict = 0
for i in range(N):
    step_1 = time.time()
    rel = Relation.create(database='debug', schema='debug', identifier='debug')
    step_2 = time.time()
    as_dict = rel.to_dict()
    step_3 = time.time()
    from_dict = Relation.from_dict(as_dict)
    step_4 = time.time()

    t_create += step_2 - step_1
    t_to_dict += step_3 - step_2
    t_from_dict += step_4 - step_3

total = t_create + t_to_dict + t_from_dict

print(f"create: {t_create:0.2f} ({t_create / N:0.4f} per node)")
print(f"to_dict: {t_to_dict:0.2f} ({t_to_dict / N:0.4f} per node)")
print(f"from_dict: {t_from_dict:0.2f} ({t_from_dict / N:0.4f} per node)")
print(f"total: {total:0.2f} ({total / N:0.4f} per node)")

Timing

Branch	Cycles	Time
experiment/use-mashumaro	100	0.00s
experiment/use-mashumaro	1000	0.04s
experiment/use-mashumaro	10000	0.23s
kiyoshi-kuromiya	100	0.36s
kiyoshi-kuromiya	1000	3.54s
kiyoshi-kuromiya	10000	33.95s

kiyoshi-kuromiya avg: 3.4ms / iteration
experiment/use-mashumaro avg: 0.023ms / iteration

So that's like a ~170x improvement :)

[jtcohen6]

Closing on account of #3034 :)