MigrationNew

    Before You Press Upgrade: A Supabase Postgres Pre-Flight Check

    Lovable designs your schema and runs your migrations. It does not own your database. That makes the Postgres upgrade, and everything that can block it, your calendar rather than the platform's.

    The one line in Lovable's docs that decides this

    Lovable's Supabase documentation is unusually clear about where the boundary sits: the integration connects a Supabase project you own to your Lovable project. Lovable does not provision the database for you. The stated prerequisite is that you already have a Supabase project, or an organisation where you can create one. From then on Lovable designs your schema, runs migrations, deploys edge functions and wires your app's UI to your data - all against a project on your own account, with your billing, your backups and your instance sizing.

    That is the right design. It also means one specific button is yours and only yours: the Postgres version upgrade. No build package presses it for you, and no support window makes it not happen.

    What pressing it actually does

    Supabase's upgrade guide describes the mechanics plainly. Your project goes offline. A new instance is created on the latest version, your data is copied across with pg_upgrade, and if something fails the original database is brought back. On success a pg_basebackup is taken before the project comes back online.

    Downtime is a function of database size and disk throughput. Supabase notes that a project is deployed with a GP3 disk type by default, which gives roughly 100Mbps when upgrading. That is the number to do arithmetic with when somebody asks how long the maintenance window is.

    The extensions that stop a Postgres 17 upgrade

    The most common reason an upgrade to Postgres 17 stops is an extension that no longer exists there. Supabase's guide is explicit that deprecated or unsupported extensions must be dropped before the upgrade runs.

    ExtensionStatus in Postgres 17What it means for your project
    plv8Not supportedAny JavaScript written as database functions has to move into application code or an edge function first
    timescaledbNot supportedTime-series tables need a plan before the upgrade window, not after it
    plcoffeeNot supportedSame class of problem as plv8, and rarely enabled deliberately
    pllsNot supportedSame class of problem as plv8, and rarely enabled deliberately
    pgjwtNot supportedWas enabled by default on every Supabase project up until Postgres 17; Supabase's own page says the platform creates and handles JWTs for you, so if you were not explicitly using it, it is most likely safe to disable

    pgjwt is the row worth reading twice. It was on by default, which means a project that never knowingly used it can still be carrying it, and it is the one most likely to be sitting in a Lovable-built database that nobody has audited.

    The four blockers that stop the upgrade outright

    Beyond extensions, Supabase's guide lists conditions that have to be resolved before an upgrade can run at all.

    Read replicas. Projects with read replicas cannot be upgraded. The replicas have to be deleted and recreated after the upgrade completes.

    The reg* data types. pg_upgrade does not support upgrading databases containing reg* data types that reference system OIDs, so the data has to be modified not to use them.

    Logical replication slots. These must be dropped. If anything downstream depends on one, that consumer needs a plan first.

    Deprecated extensions. As above.

    None of these is difficult on its own. All of them are discovered at the worst possible moment, which is after you have already announced a maintenance window.

    The one that breaks after a successful upgrade

    This is the failure mode worth planning for, because the upgrade reports success and then things quietly stop working.

    Supabase is moving password hashing from md5 to scram-sha-256. Managed role passwords are migrated automatically. Custom roles are not: the guide states you will need to manually migrate the passwords of any custom roles you have created, or you will not be able to connect using them after the upgrade. If a background worker, an analytics tool or a scheduled job connects as a role you created yourself, that connection fails after the upgrade, and the cause is not obvious from the error.

    The backups documentation compounds it. Daily backups do not store passwords for custom roles, so a restore does not rescue you either. Write those credentials down somewhere that is not the database before you start.

    Dates that pause rather than warn

    Two dated items from Supabase's changelog are worth knowing, because one of them has teeth.

    Supabase support for Postgres 14 ended on 1 July 2026. Projects still on that version are upgraded automatically - but projects using extensions that were removed, specifically timescaledb, plv8 and pgjwt, are paused instead. A paused project is not a degraded project. It is a down project.

    Separately, the self-hosted Supabase default database image moved from Postgres 15 to Postgres 17 on 17 June 2026, which matters if any part of your stack runs Supabase outside their cloud.

    The pre-flight check

    1. Know your current Postgres version and your target version, and read the release notes for the target and every intermediate version. Supabase notes that breaking changes are generally only present in major version upgrades of Postgres and PostgREST.
    2. List your enabled extensions and check each one against the target version's bundle. Back up any data that depends on an extension you will have to drop.
    3. Check specifically whether pgjwt is enabled, and whether anything in your project actually calls it.
    4. Confirm you have no read replicas, or plan to delete them and recreate them afterwards.
    5. Drop logical replication slots - after telling whoever owns the downstream consumer.
    6. Inventory every custom role and store its password somewhere outside the database.
    7. Know your recovery position before you start. Daily backup retention is the last 7 days on Pro, the last 14 days on Team and up to 30 days on Enterprise. Point-in-time recovery is a paid add-on and, once enabled, replaces daily backups. Database backups also exclude objects stored through the Storage API, because the database only holds their metadata.
    8. Plan your own application-level validation for afterwards. Supabase says as much directly: there might be changes you did not anticipate.

    Where we land on it

    Our own figures, for the record: 80 jobs on Upwork, 100% Job Success and a 4.9 rating from 30 reviews. A Strategy Audit is $499 and is credited toward any build. Founding Client is $2,999 and carries a 14-day build guarantee with a full refund if the deadline is missed. Publishable MVP is $4,999. Both build tiers include 90-day support until your first paying customer, and the revenue guarantee is first paying customers within 90 days of launch or a full refund.

    None of those tiers includes pressing your Postgres upgrade button in month eleven, and we would rather say so than let you discover it. A version upgrade is a scoped piece of work with a maintenance window, a blocker list and a rollback position, which is why we keep it in ongoing maintenance and support rather than folding it into a fixed build price. If you want to know which of the blockers above your project is carrying right now, that is a cheap thing to establish before it becomes urgent.

    Book your free 20-minute intro call.

    Sources

    Related posts