Comparison
Environment parityvsEphemeral environment
Environment parity
it worked in staging and broke in production, because staging had one replica, no mesh and a different database size.
How closely lower environments resemble production in configuration, topology, data shape and scale. Perfect parity is unaffordable, so the useful discipline is knowing exactly where the differences are and which classes of bug those differences will therefore let through. Undocumented divergence is what turns a green staging run into false confidence.
Full entry →Ephemeral environment
opening the pull request created a whole running copy at its own URL, and merging it destroyed the copy.
A full environment created per branch or pull request and torn down when it is no longer needed. It removes the shared staging queue and lets non-engineers look at a change before it merges, which is usually its highest-value property. It requires the whole stack to be genuinely reproducible from code, including data seeding, which is why it is a good forcing function and a lot of work.
Full entry →