playwithsql

implementing as immutable as possible data modelization and benchmarking it on different platforms.

Disclaimer

the benchmark comparison is for very specific use case:

• Golang 1.8.3
• Containerized DBs, latest versions
• Specific schemas
• Used configurations
• GCP as cloud provider (or local, until docker for GCP is allowing experimental)
• n1-standard-1 as machine type

Hence, they can't be used to affirm that this or this db is better. The context matters!

Results

Just check them here!

Tested dbs

• MySQL
• MariaDB
• PerconaDB
• Postgres
• CockroachDB
• SQLite
• Microsoft SQL Server
• Reference hosted MySQL on GCP
• Reference hosted Postgres on GCP
• Vitess
• Cloud Spanner

Tested setups

• local swarm cluster - local solo db container
• swarm cluster - gcloud solo db container
• kubernetes cluster - gcloud solo db container
• swarm cluster - gcloud cluster db containers
• kubernetes cluster - gcloud cluster db containers

Rebuilding the docker image

./build-docker-cmd-status.sh

Launch tests

./status/run-test.sh

Launch local status benches

./bench/status/swarm/run-all.sh (islatest or lateststatus or history)

Launch remote tests on kubernetes

./bench/status/kubernetes/run-all.sh (islatest or lateststatus or history)

TODO list

• Bench 1000000 loops and get the best of 5 runs for each db
• Test different storage (mounted standard disk, mounted ssd, local ssd)
• Test High Availability
• Leverage new go 1.8 capabilities (named queries, remove sqlx?)
• Vendor deps (github.com/kardianos/govendor)