DevOps Technical Debt

Document Status

Ported from legacy documentation on November 7, 2025 This document has been migrated from Almanac and is pending human review for accuracy and completeness. Some items may have been addressed since original writing.

DevOps Choices We Should Revisit as We Grow

Aka - where the skeletons are buried and when should we take care of them?

Fly

• Protection against accidental deletion: We should up the protection on accidental deletion of a fly app by any dev.

  As of 2024-03-01, there are clunky ways to do it (see threads), and fly is considering a better feature.

  When to tackle it? Tough one. Arbitrary target - when we hit 10 devs in the team?

• Alternative: Production organization: The other alternative is to create a new organization for production, with limited access. This might be quite painless.

Pipeline

• Use --json where possible when calling fly for more parseable output
• S3 robustness with boto: E.g. if the copy command fails, are we guaranteed that an exception is thrown

Dashlane

We have Dashlane individual, not Dashlane Business. Might be needed for SOC-2 (SSO, easier offboarding).

Doppler

• Similar to fly, our current developer plan gives too many folks access to too many secrets. Upgrading to Team ($18/m/u) would give us per user access control.

• We could also create a second doppler account for production with limited access, and create a script to sync keys across environments. May not lose much.

• We may also want to tackle removing non-secrets from Doppler and baking them into fly configs (the env section) instead.

• Another approach would be to switch to using doppler API from the apps to retrieve the environment rather than syncing doppler to fly.

Build process

• 2024-07-24: Enlightn is run alongside PHPStan. It might make sense to get rid of PHPStan once we’re comfortable with Enlightn, see comment here.

HyperDX

• We’ve briefly played with HyperDX, but then the integration broke and nobody really cared. On 2024-03-20 we decided to pull the plug for now, but it might become more relevant later if Sentry doesn’t give us all we need. See slack thread.