Postgres vs Redis

karthick-கார்த்திக் — Mon, 26 Dec 2022 22:15:04 +0000

Why ?

ℹ️ Recent article from ycombinator provides an idea about using postgres as a one stop shop for all data related needs for the application , quoting from the article

Use Postgres for caching instead of Redis with UNLOGGED tables and TEXT as a JSON data type.

thought for this experiment originated from above quote

Objective

Compare query execution time between postgres unlogged table and redis using API layer built in Golang

Procedure

For readers who are interested in codified version of the
experiment can checkout the repo

steps

set up required data stores (postgres and redis)
create an API layer in Golang for store interaction
set up query performance measurement tools for stores
run benchmark and operations on store
collect and compare

pre-requisites

Thanks to docker, Both stores can be set up using official docker images available from docker hub.Below are queries used to set up the containers
pg_stat_statements extension can be used to monitor query performance as far as redis is concerned the inbuilt redis-benchmark can be used from the command line



docker run -d --name postgres \
-e POSTGRES_USER=postgres \
-e POSTGRES_PASSWORD=postgres \
-p 5432:5432 \
-v `pwd`/docker/init.sql:/docker-entrypoint-initdb.d/init.sql \
postgres:latest -c shared_preload_libraries='pg_stat_statements'

Above docker command can help to start postgres instance up and ready

init.sql contains basic setup for the database code associated with experiment

once after the server is ready and working either pgadmin or psql can be used to interact with it
init_table.sql holds the required statements for creation of a table and extension

redis setup is more straightforward



docker run -d --name redis -p 6379:6379 redis:latest

approach

naive and simple approach is being adopted for the purpose of experiment
One line summary of the approach is to build an API layer in golang and measure query statistics using golang's in-built benchmarking and testing

Code for API layer is available in repo

how it works - There are isolated adapters set up for both stores, All required operations for populating data into store and reading required data are carried using go test

Benchmark for get opeartion(postgres) and put(redis) are used to get query stats for the corresponding store

To execute store operation - go test -run ./... -tags=integration which sets up required data in the stores

To measure performance of query a quick load test is carried out using the benchmark functions in test, to get load test and benchmark

postgres API query latency benchmark can be observed using

go test -bench . ./adapter/store/pgsql -count 10

in similar manner redis benchmark can be calculated using
go test -bench . ./adapter/store/redis -count 10

postgres query performance can be monitored via pg_stat_statements extension using the SQL query



select calls,total_exec_time,min_exec_time,max_exec_time,mean_exec_time,rows,query
from pg_stat_statements where query='select * from public.test';

redis bench mark can be observed using inbuilt tool for redis

redis-benchmark -h localhost -p 6379 -n 10000

Query latency from both data layer and API layer can be measured using extensions and benchmark

observation

max exec time	db	no of rows
0.013056668	postgres-unlogged	27570000
0.029535	redis	27570000

conclusion

From original artical covering multiple aspects of postgres followed by the naive experiment between pg and redis.It is clearly evident and this acts as additional proof postgres can be one stop shop for all data related needs for any application

Bibiliography

dockerhub
postgres
redis

Some Self promotion

twitter, mastodon, github