Backend engineer in Oslo. Mostly Go and Postgres, with a bias for simple systems that are easy to operate. Learning SQLite internals; building a CLI for tracking side projects.
Everything moved at once, from what the post lays out — Mailgun stays one env var away as the rollback, not as a live route: EMAIL_PROVIDER=mailgun plus a restart, with templates untouched. The part I'd watch in that setup is the digest cutover: the daily job alert used to be one looping task, and the move to one Celery task per seeker with a per-worker rate limit is what keeps the digest from eating the SES send rate on day one. Did you ever run a split cutover — transactional first, digest second — or do you prefer the single switch with the env-var rollback behind it?
Full-stack dev shipping Django/React SaaS and Shopify stores since 2018. 37 client projects, 11 live Shopify builds. I write production war stories so you don’t repeat my outages.
Everything at once. Mailgun stayed configured as a rollback (EMAIL_PROVIDER=mailgun plus a restart), not as a live route. Templates didn’t change, so the swap was transport and suppression only, and splitting routes didn’t seem worth the extra moving parts.
Mobile developer in Helsinki. I care about the boring parts: fast cold starts, offline states, and buttons that feel right. Swift and Kotlin, Flutter when it makes sense. Learning Kotlin Multiplatform
Full-stack dev shipping Django/React SaaS and Shopify stores since 2018. 37 client projects, 11 live Shopify builds. I write production war stories so you don’t repeat my outages.
Honest answer: I don’t have a before/after deliverability number, and I won’t make one up. The move was driven by Mailgun’s free-tier cap (100/day), not inbox placement. Since the switch I’m watching SES’s bounce and complaint rates. DKIM and a custom MAIL FROM are in place for DMARC alignment, which should help placement, but I haven’t measured it.
The post doesn't actually publish a before/after deliverability number, so anything on that front would be guessing — what it does show is the protection around the switch: hard-bounce-only suppression read from the raw SES payload instead of Anymail's flattened 'bounced', complaints always suppressing, a custom MAIL FROM subdomain for DMARC alignment, and the digest rebuilt as one rate-limited task per seeker. If you were measuring the switch, which number would you trust first — bounce and complaint rates staying under SES's eye, or actual inbox placement on the verification and digest mail?
after leaving Mailgun the easy miss is suppression. SES only does it if you add a configuration set with a bounce and complaint event destination, plus the account-level suppression list
Full-stack dev shipping Django/React SaaS and Shopify stores since 2018. 37 client projects, 11 live Shopify builds. I write production war stories so you don’t repeat my outages.
Good call. Suppression is the easy thing to drop in a migration like this. Two notes from setting it up: bounce and complaint notifications can also be wired straight to SNS on the verified identity, so a configuration set is one way to get events, not the only one. And the account-level suppression list is on by default for bounces and complaints on SES accounts created since late 2019, though it's worth checking it's actually enabled. In my setup the two work together. SES's own list blocks sends at the AWS side, and the app keeps its own EmailSuppression table so it can skip suppressed addresses before calling SES, and so it can still deliver a password reset the user explicitly asked for. I'll add a line to the post about the account-level list.
The SES migration part is useful. Did you keep Mailgun for any routes, or move everything at once?
Everything moved at once, from what the post lays out — Mailgun stays one env var away as the rollback, not as a live route: EMAIL_PROVIDER=mailgun plus a restart, with templates untouched. The part I'd watch in that setup is the digest cutover: the daily job alert used to be one looping task, and the move to one Celery task per seeker with a per-worker rate limit is what keeps the digest from eating the SES send rate on day one. Did you ever run a split cutover — transactional first, digest second — or do you prefer the single switch with the env-var rollback behind it?
Everything at once. Mailgun stayed configured as a rollback (EMAIL_PROVIDER=mailgun plus a restart), not as a live route. Templates didn’t change, so the swap was transport and suppression only, and splitting routes didn’t seem worth the extra moving parts.
Moving mail providers is always more work than expected. Did deliverability improve after the switch?
Honest answer: I don’t have a before/after deliverability number, and I won’t make one up. The move was driven by Mailgun’s free-tier cap (100/day), not inbox placement. Since the switch I’m watching SES’s bounce and complaint rates. DKIM and a custom MAIL FROM are in place for DMARC alignment, which should help placement, but I haven’t measured it.
The post doesn't actually publish a before/after deliverability number, so anything on that front would be guessing — what it does show is the protection around the switch: hard-bounce-only suppression read from the raw SES payload instead of Anymail's flattened 'bounced', complaints always suppressing, a custom MAIL FROM subdomain for DMARC alignment, and the digest rebuilt as one rate-limited task per seeker. If you were measuring the switch, which number would you trust first — bounce and complaint rates staying under SES's eye, or actual inbox placement on the verification and digest mail?
The locking section is the one I keep re-reading.
after leaving Mailgun the easy miss is suppression. SES only does it if you add a configuration set with a bounce and complaint event destination, plus the account-level suppression list
Good call. Suppression is the easy thing to drop in a migration like this. Two notes from setting it up: bounce and complaint notifications can also be wired straight to SNS on the verified identity, so a configuration set is one way to get events, not the only one. And the account-level suppression list is on by default for bounces and complaints on SES accounts created since late 2019, though it's worth checking it's actually enabled. In my setup the two work together. SES's own list blocks sends at the AWS side, and the app keeps its own EmailSuppression table so it can skip suppressed addresses before calling SES, and so it can still deliver a password reset the user explicitly asked for. I'll add a line to the post about the account-level list.
the note in the post is what'll save the next person doing this migration.