Things you should definitely do in a greenfield application

A lot of these are Rails specific, but can be adapted to other stacks.

• Steal some things from https://github.com/discourse/discourse
• QueryTags in query logs, including the code namespace and function (Job, Controller+Action, etc) as well as the trace id
• Tests
  • Use minitest
  • Use fixtures, not factories
  • Postgres
    • Use tablespaces and put tables on tmpfs (RAM)
      • https://gajus.com/blog/setting-up-postgre-sql-for-running-integration-tests
    • Use unlogged tables
    • fsync=off
    • full_page_writes=off
    • synchronous_commit=off
    • vacuum=off
    • checkpoint_timeout=60m
    • wal_level=minimal
    • max_wal_senders=0
  • Enforce a maximum test runtime
  • Fix or delete flaky tests
  • https://github.com/basecamp/gh-signoff and don’t use a CI
• Preconnect cross-origin domains
• Compress assets + use a CDN
• Compress response payloads (HTML/JSON)
• Sidekiq jobs
  • Leverage https://github.com/sidekiq/sidekiq/wiki/Iteration
  • SLO based queue naming (within_6_hours, within_0_seconds, etc.)
    • Idea: expand sidekiq_options to accept slo and weight to pick/generate a queue name which is automatically used
      • slo: 5.minutes, weight: 0.5
  • Manual load shedding
    • Short-circuit all jobs of class X with argument[0] = Y
  • Implement Sidekiq batch invalidation by default for every job
  • Provide fairness between tenants
    • TODO: How? Sidekiq Limiter?
• Avoid using UUIDs at all
  • Use bigint primary keys, for public identifiers use sqids with a secret
• RMP in all environments
• RUM in browser
• Autoscale web and worker
• Instrument request queue time
• Automated alerts/SLOs in Terraform
• Prosopite/strictloading in tests
• jemalloc
• Turbo/DataStar/hx-boost
• Mise Tasks
• Code formatting
  • Standardrb
  • Enforce with git hooks https://github.com/sds/overcommit
• https://evilmartians.com/chronicles/gemfile-of-dreams-libraries-we-use-to-build-rails-apps
• Enforce a zero-bug policy
• All “Repository” operations must be batched
• Setup a CSP policy from day one
• Set a low statement_timeout in Postgres
• Track performance regressions
  • Metric A: What I want to monitor, e.g., # of requests that took longer than 5 seconds, grouped by controller action
  • Metric B: Same, but timeshifted week ago
  • Alert: B/A > 2
• Feature flags
  • Support a tree-like structure, since some feature flags are nested by nature. Enabling an child node should enable all parent nodes
    • This can be done simply with a parent_feature_ids array column
  • Notify when feature flags are fully rolled out or fully disabled, every week
    • As a reminder to delete dead code branches
• Setup span/slog fields for metrics/signals/monitoring, per request/job:
  • Tenant ID
  • User ID
  • Feature Flags
  • Error code
  • Number of DB queries
  • Number of DB tables queries
  • Number of DB services hit
  • Number of Elastic queries
  • Number of Elastic services hit
  • Number of cache reads
  • Number of cache writes
  • Number of cache services hit
  • Number of object allocations
  • Number of HTTP requests
  • Number of HTTP request retries
  • Number of HTTP requests uniqued by domain
  • Time spent in DB
  • Time spent reading from DB
  • Time spent writing to DB
  • Time spent in cache
  • Time spent reading from cache
  • Time spent writing to cache
  • Time spent in view layer
  • Time spent in HTTP requests
  • Time spent
  • CPU wall time spent