spree-upgrade

Upgrading Spree

Every Spree upgrade is the same three stages:

1. Bump the gems — bundle update the spree* gems.
2. Migrate the schema — copy new migrations out of the gems (spree:install:migrations) and run db:migrate.
3. Backfill data — rake spree:upgrade runs the version-specific data tasks listed in the upgrade manifests shipped inside spree_core.

Then a fourth, manual stage: review the upgrade guide for your hop (behavior changes, config moves, code that needs editing). The rake-runnable part never covers everything.

Going from 5.x to 6.0? That's a major, breaking hop with preconditions (you must be on 5.6 first, Rails 8.1, Gemfile swaps, roles-as-data). Read the spree-upgrade-5-to-6 skill before starting.

Pick your flavor

CLI (create-spree-app)

spree upgrade --plan            # list the backfill steps a run would execute; runs nothing
spree upgrade                   # bundle update spree* → install migrations + db:migrate → rake spree:upgrade

spree upgrade prompts before bundle update and before migrations (answer No to skip one). --yes skips the prompts for non-interactive use. The rake backfills always run.

It needs the ejected dev stack. Fresh projects run the prebuilt ghcr.io/spree/spree image with no source bind mount, so bundle update inside it can't touch your server/Gemfile.lock and copied migrations never land in server/db/migrate/. Run spree eject first (switches to the build-from-source compose), then upgrade.

The CLI refuses to run in a monorepo edge project (SPREE_PATH set in .env) — use the pnpm server:* scripts from the monorepo root there.

Native (classic Rails app, or no CLI)

bundle update

CLI flags map 1:1 to env vars on the rake task: --plan → DRY_RUN=1, --step <id> → STEP=<id>, --to <x.y> → TO=<x.y>.

How rake spree:upgrade works

• Manifests live in the gem at spree_core/lib/spree/upgrades/<from>_to_<to>/manifest.yml (dots become underscores: 5_6_to_6_0). Each lists ordered steps: id, name, task (a rake task), notes, and optionally optional: true (skipped when the task isn't defined, e.g. a provider gem you don't have).
• A normal run walks every manifest whose to is ≤ the installed minor version, oldest first. An app that skipped a backfill two versions ago catches up automatically.
• Every step must be idempotent — re-running the whole task on an already-upgraded app is a safe no-op. This is what makes it safe as a release-phase command.
• Manifests only contain data backfills. bundle update, migrations, cron/recurring jobs and code changes are never in them.

List what the installed gem ships:

Plan first

Prints every step (id, rake task, notes) that a real run would execute, in order — plan mode uses the same manifest selection as a real run (every manifest whose to ≤ the installed minor, or ≤ --to when given). Nothing is executed and the bundle/migrate pre-steps are skipped.

The plan reads manifests from the installed gem, so bump the gems first — a 5.6 bundle doesn't contain the 5.6 → 6.0 manifest.

Retry one step

After a partial failure, fix the cause and re-run only that step (skips bundle + migrate):

Step ids come from the plan output. If the same id exists in two manifests the task aborts and asks for TO= to disambiguate. Some tasks take their own env knobs (BATCH_SIZE, SKIP_INVALID_ROWS=true, …) — they're documented in each step's notes; pass them the same way (spree rake spree:upgrade STEP=… BATCH_SIZE=200, or spree exec env BATCH_SIZE=200 bin/rake …).

You can also invoke the underlying task directly — spree rake <task> / bundle exec rake <task> — which is handy for tasks that aren't in any manifest.

Cap the version

Eligible manifests become those whose to ≤ the cap. Useful when you're landing on an intermediate minor before the next hop (5.x apps must land on 5.6 before going to 6.0).

What it does not do

• Extension upgrades. spree upgrade bumps every installed spree* gem (it lists them with bundle list --name-only inside the container), but each extension's own migrations, install generators and breaking changes are yours to run — check its CHANGELOG. Extensions must have a release compatible with the target Spree version or bundle update won't resolve.
• Recurring jobs / config. New scheduled jobs (Solid Queue config/recurring.yml, or your Sidekiq/cron setup), moved settings and new env vars are listed in the upgrade guide, not the manifest.
• Your code. Decorators, subscribers, workflow hooks, dashboard plugins and storefront code referencing renamed or removed APIs have to be audited by hand. The /spree:audit-upgrade command in this plugin does a read-only readiness pass.
• SDKs. Frontend packages (@spree/sdk, @spree/admin-sdk, @spree/dashboard) are bumped through your own package manager — the CLI prints the apps/storefront @spree/sdk version in its "Next steps" panel as a reminder.

In production

Don't run spree upgrade against production — it's a dev CLI. Your deploy already does bundle install and db:migrate (the starter's Docker entrypoint runs bin/rails db:prepare on boot). Add the backfills as one more command:

Wire it as a release-phase command (Heroku), pre-deploy command (Render), init container / Job (Kubernetes) or post-deploy hook (Kamal, Capistrano). Because every step is idempotent, running it on every deploy is safe and cheap once the data is migrated.

Order matters: migrate → backfill → serve new code. Some steps are safety nets for the migration and some must run before later migrations (they say so in their notes). For big hops, run the backfills in a maintenance window or against a restored copy first to measure duration — several tasks honor BATCH_SIZE.

Checklist

1. Back up the database.
2. Read the upgrade guide for the hop: node_modules/@spree/docs/dist/developer/upgrades/<from>-to-<to>.md or https://spreecommerce.org/docs/developer/upgrades.
3. Bump gems (and extensions), run migrations.
4. spree upgrade --plan → read every step's notes.
5. Run the backfills; re-run failed steps with --step.
6. Do the manual items from the guide; run your test suite; bump SDKs.
7. Ship with bundle exec rake spree:upgrade in the release phase.

Gotchas

• --plan printing nothing (or only old manifests) means the gems aren't bumped yet — the plan comes from the installed spree_core.
• bundle update fails in the container → you're on the prebuilt image; spree eject first, or the bundle is out of sync (spree bundle install).
• A step aborts on purpose. Several 6.0 tasks refuse to continue on data they can't convert faithfully (e.g. migrate_users_to_customers, migrate_returns) and print the offending ids plus the env var that overrides. Read the message; don't blindly re-run.
• Never skip ahead. Manifests assume their from version's schema. Jumping minors is fine (the walk catches up); skipping a required landing version (5.6 before 6.0) is not.

Where to read further

• node_modules/@spree/docs/dist/developer/upgrades/ — per-hop upgrade guides
• node_modules/@spree/docs/dist/developer/cli/quickstart.md — spree upgrade reference
• https://spreecommerce.org/docs/developer/upgrades
• Related skills: spree-upgrade-5-to-6, spree-cli, spree-deployment, spree-extensions