stalwart-migrator/README.md

# stalwart-migrator

In-place upgrade tool for Stalwart Mail Server, 0.15.5 → latest: no data
loss, a checkpoint at every step so an interrupted run resumes instead of
restarting, and automated validation that the server still works afterwards.

**Recovery from a failed migration is your own snapshot or backup — this
tool does not undo a migration.** See [Recovery is your
job](#recovery-is-your-job) before using it on anything you care about.

Go, standard library only — no external dependencies.

A companion to [**ihasmail**](https://github.com/Coffey-Labs/ihasmail),
a JMAP-first webmail client for Stalwart. That one is what you read your
mail in; this one is what gets the server underneath it onto a version that
speaks the protocol it needs.

**This has been used to migrate a production mail server.** On 2026-08-25 it
took a live server — nine domains, six accounts, a 2.4 GB RocksDB store —
from 0.15.5 to 0.16.19 with **8 seconds** of downtime, every phase green
including post-cutover validation, mail flowing before and after. That run
was preceded by a full dress rehearsal on a clone of the same server, which
is the practice this project most recommends copying: see [Rehearse on a
clone first](#rehearse-on-a-clone-first).

**It has also been used at a scale well past that, by someone else.**
[@kaya-eu](https://github.com/Coffey-Labs/stalwart-migrator/issues/1)
reported three successful 0.15.5 → 0.16.19 migrations on three servers: a
testing and a production instance, each 16 domains, 55 accounts and roughly
**221 GB** of real mail, and an arm64 home server of 1.4 GB across 181
folders. Read that with two qualifications. They ran a commit predating the
automated Docker cutover, so they performed the cutover by hand — what those
runs exercised is preflight, the dumps, the settings conversion and the
recovery-mode store migration, not the container cutover. And they hit
things worth knowing about before you follow them: an arm64 binary that was
fetched for the wrong architecture (fixed), a store migration that needed
one more recovery-mode boot than the tool performs (**not fixed** — see [The
store migration may need one more recovery
boot](#the-store-migration-may-need-one-more-recovery-boot)), and data loss
from booting recovery mode again *after* a completed
migration (see [Do not boot recovery mode
again](#do-not-boot-recovery-mode-again-afterwards)). Their full report is
[issue #1](https://github.com/Coffey-Labs/stalwart-migrator/issues/1).

## Before you start: two things you must fix on the server

Neither is something this tool can do for you, and both stop a migration
dead. `preflight` refuses on both, while the mail server is still running —
but they are worth knowing before you book a maintenance window, because
fixing them is a change to your directory, not a flag.

1. **Remove or collapse multi-tenancy.** v0.16 requires a tenant-scoped
   account to sit on a domain owned by that same tenant, for its primary
   domain and every alias. v0.15 imposed no such rule, so an install that is
   perfectly valid today can be unrepresentable in v0.16. Run
   `stalwart-migrate tenants` to see who owns what. Where a domain has no
   tenant of its own and only one tenant's accounts use it, the conversion
   repairs it for you; where two tenants genuinely share a domain, nothing
   can, and you must resolve it in v0.15 first — give each tenant its own
   domains, move the accounts into one tenant, or remove the tenants
   entirely.

2. **Migrate as a directory account, not the built-in admin.** A
   `[authentication.fallback-admin]` from `config.toml` authenticates
   perfectly well right up to the moment the migration finishes, and then
   stops existing — v0.16 keeps its configuration in the store, so the block
   defining it is never read again. The migration itself still succeeds; what
   you lose is the ability to verify it, recalculate quotas, or administer
   the server afterwards. See [You need a named admin
   account](#you-need-a-named-admin-account-before-you-migrate).

## Status

Roughly 14,600 lines of Go, stdlib only, of which about 6,300 are tests.
Every phase exists as a package and `run` wires them into the migration
described above.

| Command | State |
|---|---|
| `stalwart-migrate preflight` | **Works** — read-only checks and a migration plan |
| `stalwart-migrate rehearse` | **Works** — read-only; converts your settings and reports what won't carry over |
| `stalwart-migrate run` | **Works** — performs the migration; `--recovery-point-confirmed --yes`. Container deployments additionally need `--container-path-unproven` (see [Docker deployments](#docker-deployments)) |
| `stalwart-migrate tenants` | **Works** — read-only; who owns which domain, and what would block a migration |
| `stalwart-migrate status <id>` | **Works** |
| `stalwart-migrate report <id>` | **Works** — prints what validation found for a run |

**`run` performs the migration**, in the order
preflight → stage → dump → stop → convert → recovery-mode → cutover →
validate. It
needs two flags: `--yes` (intent) and `--recovery-point-confirmed` (a claim
that you have a snapshot or backup you have verified you can restore — this
tool cannot undo a migration and will not start without it).

**Start with `rehearse` first.** It is read-only, needs no maintenance
window, and tells you what `run` will and won't carry over.

### Docker deployments

A containerised Stalwart can be migrated, with two things to know first.

**The container path has never completed a migration against a real
Stalwart image.** What it inspects and what it assembles have now been
checked against one, which is how two problems were found and fixed
(#11) — but a fake `docker` still proves only that the right commands are
assembled, not that the image reads the config it is handed and comes up
as the server it was. `run` refuses a container deployment unless you pass
`--container-path-unproven`, which is there so nobody reaches it without
being told. Rehearse on a clone first; that advice goes double here.

**Two flags and one convention:**

- `--target-image` names the image in full, e.g.
  `stalwartlabs/stalwart:v0.16.14`. It is never derived from the running
  container by swapping the tag — that is wrong for a digest-pinned image,
  a mirror or a fork, and being wrong means pulling the wrong software into
  a mail server.
- `--container` names the container (default `stalwart`).
- `--data-dir` must name the path **inside** the container, since that is
  where its data actually lives. `preflight` says so if it matches none of
  the container's mounts.

**What it refuses, and why.** A container cannot be edited in place the way
a unit file can, so cutting one over means rebuilding it — and a container
rebuilt without its capabilities, its custom network or its device mappings
starts cleanly and is quietly not the server it was. So cutover carries
across what it understands (mounts, ports, environment, restart policy,
labels) and refuses outright when it finds anything else, naming what it
found — and it asks that question in `preflight`, while the server is still
running, rather than only at cutover after it has stopped. What it carries
is mounts, ports, environment, restart policy, labels, and anything the
container overrides on its image: a `--user`, an `--entrypoint`, a command
of your own. What the container merely *inherits* from its old image is
left to the new one, whose own defaults are the ones that go with it.
It also refuses a container whose data is not on a volume — an
upgrade replaces the container, and the writable layer goes with it — and
one managed by Docker Compose, because recreating it out from under compose
leaves the container and the compose file disagreeing about what is
deployed, and the next `compose up` reverts the migration. Compose
deployments are migrated by editing the image tag in the compose file and
running `compose up -d`.

**The config.** The converted v0.16 config is written into the host side of
whichever mount covers `--data-dir`, named on the container side, and the
recreated container is started with `--config` pointing at it. It cannot go
anywhere else: cutover recreates a container with the mounts it had and
cannot invent a new one. The official image's own default is
`--config /etc/stalwart/config.json`, which is a *different* volume, so a
container left to that default would come up on whatever the old version
had left there. If your container overrides its command, cutover refuses
rather than merging the two — both are the container's argv and there is no
honest way to guess.

**What it keeps.** The old container is renamed rather than removed, the old
image is never pruned, and the container's `docker inspect` is preserved as
an artifact before anything is replaced. Together those are the manual
restore path — see [Recovery is your job](#recovery-is-your-job), which
applies here exactly as it does to a binary install.

Measured on a full migration: the store converts in seconds, and the service
was down for **6 seconds** end to end. Plan the window around verification,
not data volume.

After cutover, `run` compares the migrated instance against the snapshot
preflight took, and fails the command if an account that existed before is
missing from it. A domain that no longer appears is reported as a warning
rather than a failure: the two versions do not agree on what counts as a
domain — principals on one side, `Domain` objects on the other — and failing
a migration over that difference would abort runs that lost nothing. The service is left running either way — by that
point the store has been migrated in place, so stopping it would not undo
anything; your recovery point is the way back. `report <run-id>` prints the
same finding again later. Where preflight had no admin URL to snapshot from,
validation reports itself as skipped rather than passed.

Package state:

Lines are implementation only; each package carries its tests alongside.

| Package | Lines | Tests |
|---|---|---|
| `internal/stalwartapi` | 1456 | yes |
| `internal/backup` | 1333 | yes |
| `internal/preflight` | 1035 | yes |
| `internal/applyplan` | 913 | yes |
| `internal/cutover` | 784 | yes |
| `internal/recovery` | 431 | yes |
| `internal/checkpoint` | 406 | yes |
| `internal/validate` | 382 | yes |
| `internal/stage` | 233 | yes |
| `internal/service` | 201 | yes |
| `internal/plan` | 130 | yes |
| `internal/config` | stub | — |

## Why not a shell script

Stalwart's 0.15 → 0.16 boundary is not a drop-in binary swap: settings move,
and the data directory has to be migrated rather than merely copied. The
failure mode that matters is a half-migrated mail store with no way back —
which is why backup verification and checkpointing are the design centre
rather than conveniences bolted on afterwards — and why the tool refuses to
cut over until you confirm you have a way back.

[`ARCHITECTURE.md`](ARCHITECTURE.md) covers this in full: §1 on why a thin
wrapper is insufficient, §4 on the migration phases, §5 on the checkpoint
state machine, §6 on the CLI surface, and §8 on what is still open.

## Build and test

```sh
go build ./...
go test ./...
```

Requires Go 1.26 or newer.

## Trying it safely

`preflight` is the sensible starting point — its checks against the Stalwart
installation are read-only:

```sh
sudo go run ./cmd/stalwart-migrate preflight
```

It still needs write access, because it records the run as a checkpoint
before doing anything else:

```
create run: checkpoint: create run directory:
mkdir /var/lib/stalwart-migrator: permission denied
```

That path is `checkpoint.DefaultBaseDir`, a compile-time constant with no
flag or environment override — so preflight needs either root or a
pre-created writable `/var/lib/stalwart-migrator`. (`run` takes `--work-dir`
for its scratch space, but that is a different directory and does not move
the checkpoint store.)

`rehearse` is the next step, and unlike everything else here it is worth
running today — see the next section.

## Rehearse before you migrate

```sh
stalwart-migrate rehearse --admin-url https://mail.example.com \
    --admin-user admin --target 0.16.14
```

It runs preflight, dumps your settings and principals, converts them with
Stalwart's own `migrate_v016.py`, and reports **both halves** of the result:
the apply plan of what will carry over, and the worklist of what will not.

It copies no data, clones nothing, starts no server, and never writes to the
store, so it is safe to run against production repeatedly and without a
maintenance window.

It also generates a **supplemental plan** for the part it can rebuild
automatically — currently your network listeners, which is the difference
between a migrated server that answers and one that doesn't — and reports
exactly how much of the worklist that covers (on a test instance: 24 of
3,505 keys, and it says so rather than implying more). Review it, then apply
it after `export.json`:

```sh
stalwart-cli apply --file <state-dir>/runs/<run-id>/supplement.json \
    --url https://mail.example.com
```

The worklist is long but mostly not work, and `rehearse` says which is
which. Measured against a real production instance, `migrate_v016.py`
carried **219 of 12,401 settings**. Of the 12,182 it left:

- **8,547** are runtime auto-ban state that repopulates itself
- **3,337** are stock spam-filter and lookup data v0.16 ships its own copies
  of — restoring v0.15's would revert a year of upstream updates
- **~224** were already carried another way, DKIM signatures included
- **~293** genuinely need your eyes

`server.listener` is in that third group only because this tool regenerates
it for you; without that a migrated instance answers on no ports at all. Both outputs are preserved
under `<state-dir>/runs/<run-id>/` (`export.json` and `unmigrated.txt`) even
though the rest of the scratch directory is cleaned up, because they are the
conclusions.

This replaced an earlier `run --dry-run` that cloned the data directory into
a sandbox and migrated the copy. That proved the store opens, at the cost of
copying it twice — while the half that found every real problem needed no
copy at all. ARCHITECTURE.md §4.9 has the reasoning.

## Rehearse on a clone first

`rehearse` is read-only and stops short of the half that matters: it converts
your settings but never applies them, and applying is where a real migration
fails. A clone closes that gap, and on a 2.4 GB store it costs about six
seconds of production downtime to build.

1. **Copy the data directory with the service stopped.** RocksDB is
   single-writer, so a hot copy may be torn — and a rehearsal on a torn store
   fails for reasons production never would, or passes when it should not.
   Stop, `cp -a` the data dir to the same disk, start again, archive it
   afterwards; the service is down only for the local copy.
2. **Give the guest no route off the host.** A clone of a live mail server
   will otherwise renew certificates for your real domains and deliver
   whatever is in the outbound queue. In libvirt that means a network with no
   `<forward>` element. Verify it from inside the guest rather than assuming.
3. **Run the real thing**: `preflight`, then `run` with `--target-binary`,
   `--stalwart-cli` and `--migration-script` pointed at locally staged copies,
   since an isolated guest can download nothing.
4. **Snapshot the guest while it is shut down**, so a failed attempt costs
   seconds to reset rather than a rebuild.

What that rehearsal caught, none of which the mock could express: a
multi-tenant arrangement v0.16 cannot represent; `--target-binary` never
reaching preflight, so an air-gapped host failed on a release lookup; an
`admin` account that was a config fallback-admin and stopped working the
moment the migration finished; and `tenant-admin` roles the converter does
not restore.

Two things the isolation costs, so they are not mistaken for faults: the
guest cannot fetch the v0.16 web interface, so `/account/` returns 404, and
certificate renewal cannot be exercised at all.

## Stalwart's own converter silently drops ACME

`migrate_v016.py` consumes every `acme.*` setting and emits nothing for them.
They are **not** reported as unmigrated either, so nothing warns you: the
certificate carries over, the provider that renews it does not, and TLS keeps
working until the certificate expires roughly ninety days later.

Check for `acme.*` in your dump before migrating, and recreate an
`AcmeProvider` afterwards if there was one. Note that `accountKey` is
server-set in v0.16, so the existing ACME account cannot be carried over —
the server registers a new one on first issuance.

This tool does not yet generate that object for you. It should: the
supplemental plan already does the equivalent for listeners.

## The store migration may need one more recovery boot

**This one is open, and it is the reason to rehearse on a clone.** The
recovery cycle boots the target version once, replays your settings into
it, and stops. Across three real migrations, two different failures showed
up that one extra recovery-mode boot cured:

- the settings apply failing on its very first object with
  `primaryKeyViolation`, where re-running the identical apply against a
  fresh recovery boot went straight through; or
- the next normal start panicking with *"Upgrading to version 0.16 is a
  multi-step process"*, where booting recovery mode once more, letting it
  come up and stopping it cleanly was enough.

Never both on the same run — whichever appeared, one more recovery boot
before the real start got past it. That panic is Stalwart's own, and it
suggests the store migration is not finished when the single boot exits.

The likely fix is a settle boot after the apply, and it is not in yet
because getting an extra recovery boot wrong is its own hazard — see [Do
not boot recovery mode again
afterwards](#do-not-boot-recovery-mode-again-afterwards). If you hit
either, that is the manual step. Reported by
[@kaya-eu](https://github.com/Coffey-Labs/stalwart-migrator/issues/1).

## A certificate that serves HTTPS may not serve the mail ports

A `Certificate` object carried into v0.16 with the right SAN is picked up
by the HTTP listener on its own. **IMAPS, SMTPS and POP3S are not**: they
keep serving a self-signed certificate until `defaultCertificateId` is set
on `SystemSettings` and the server is restarted.

This is the kind of thing that looks fine from a browser and surfaces as a
mail client complaining days later, so check it as part of your
post-migration verification: connect to 993 or 465 and confirm which
certificate you are handed, not just to 443.

This tool does not set it for you. Like the `AcmeProvider` above, it
should, and the supplemental plan is where it belongs. Reported by
[@kaya-eu](https://github.com/Coffey-Labs/stalwart-migrator/issues/1).

## You need a named admin account before you migrate

**A config-file fallback admin will not survive the migration.** If the only
administrator you have is an `[authentication.fallback-admin]` block in
`config.toml` — which is what `stalwart --init` sets up — you will come out
of the migration unable to administer the server.

Create a real account in the directory, with the admin role, and confirm you
can log in as it *before* migrating. Three separate reasons, all verified
against a real 0.15.5 → 0.16.14 migration:

1. **v0.16 keeps its configuration in the store, not in a file.** After the
   migration the server is started with a config that is little more than a
   pointer at the data store, so the old `config.toml` — and the
   fallback-admin block inside it — is no longer read at all. That
   credential simply stops existing.
2. **`migrate_v016.py` gives every migrated account the `User` role**,
   whatever it held before. An account that was an administrator in v0.15
   comes out authenticating normally and refused every management
   operation. `rehearse` generates the operation that restores it — but the
   account has to exist in the directory for there to be anything to
   restore.
3. **The account's local part must be unambiguous.** v0.16 identifies an
   account by local part plus domain, so if `[email protected]` and
   `[email protected]` both exist, this tool will refuse to restore either
   role rather than risk granting administrator rights to the wrong one.
   It says so rather than guessing; you then grant it by hand.

4. **The role has to survive, not just the account.** `tenant-admin` has no
   v0.16 equivalent and is not restored, so an account whose rights came
   only from it authenticates afterwards and is still refused management
   operations. Preflight cannot check this — it cannot know which roles the
   converter will carry across — so confirm on a clone, or immediately
   afterwards, that the account can still administer the server.

`preflight` refuses to proceed if the account you authenticate with is not
in the directory, so this is caught before anything is touched rather than
after the migration completes. The practical check: make sure you can
authenticate to the admin API as a directory account — not as the fallback
admin — that its local part is unique across your domains, and that it holds
admin rights through a role other than `tenant-admin`.

## Recovery is your job

**This tool does not undo a migration.** There is no `rollback` command.
Recovery from a failed migration is your own snapshot or backup, taken by
whatever method you already trust and know how to restore — a ZFS, LVM or
btrfs snapshot, a VM or volume snapshot, or a restorable backup. Choosing
that method, taking it, and verifying you can actually restore from it is
out of scope for this tool: it does not take one, does not check that one
exists, and cannot restore from one.

Cutover refuses to start until you confirm a recovery point exists. That
confirmation is an acknowledgement, not a check — nothing here can verify
your snapshot. Its only purpose is that nobody migrates a production mail
server having never been asked the question.

**Take the snapshot with the service stopped** if you want a clean one. A
snapshot of a running Stalwart is crash-consistent rather than clean; RocksDB
will usually recover from its WAL, but "usually" is doing real work in that
sentence.

### Restoring from a snapshot loses mail delivered since

Reverting to any pre-migration recovery point discards mail delivered
between taking it and restoring it. This is inherent to restoring a point in
time and this tool cannot solve it — plan your migration window with that in
mind, and consider holding inbound mail at a secondary MX for the duration
if the gap matters to you.

### What the tool does to make a manual restore easier

- **The old binary is preserved**, never deleted, next to the new one as
  `<binary>.v<old-version>` — so putting things back doesn't depend on
  re-downloading a specific old release under pressure.
- **The original service definition is preserved** as `<unit>.pre-<run-id>`
  before cutover rewrites it, so you aren't reconstructing a unit file from
  memory.
- **The settings and principals dumps, the apply plan and its supplement**
  are kept in `<state-dir>/<run-id>` — `/var/lib/stalwart-migrator/runs/<run-id>`
  unless you moved it. These four are the only files in a run that cannot be
  produced again afterwards: the dumps need a live pre-migration instance,
  and the plan is what was actually replayed into your store. They are kept
  whether or not the run succeeded and whether or not you passed
  `--keep-artifacts`.
- **Every artifact path and checksum is in the checkpoint**, and
  `stalwart-migrate status <run-id>` prints exactly which steps completed
  and which failed — which is the first thing you want when deciding what to
  restore.

None of this is a substitute for the snapshot. It's what makes the twenty
minutes after restoring one less unpleasant.

### Do not boot recovery mode again afterwards

The migration works by starting the new version once in recovery mode,
replaying your settings into it, and stopping it. That is a one-time step
in a migration, and it is worth knowing that it is not a general-purpose
maintenance mode.

An operator who booted recovery mode again — the same way the migration
does, `STALWART_RECOVERY_MODE=1` against the same data directory — for
reasons unrelated to the migration, on a server that had migrated
successfully days earlier, found that `Domain` and `Account` queries came
back empty on the next normal start. This happened twice, on two different
servers. It was not a stale read: creating a domain that had certainly
existed a moment earlier succeeded, with no `primaryKeyViolation`, so the
records were genuinely gone. Disk usage did not change.

What recovered it both times was re-applying that run's `export.json` and
`supplement.json` against a fresh recovery boot, which is why those two
files are now kept for you. If you need to change something after a
migration, use the admin API or `stalwart-cli` against the running server.

This is Stalwart's behaviour rather than this tool's, and it is reported
here because this tool is where you learned the technique. Reported by
[@kaya-eu](https://github.com/Coffey-Labs/stalwart-migrator/issues/1).

### Why it works this way

An earlier version of this tool implemented rollback itself: it restored the
filesystem backup, verified every restored file against a manifest, replayed
SQL dumps, reinstalled the old binary, and re-validated the result. It was
tested and it looked good. It was removed, because restoring bytes correctly
is not the hard part — it copied contents and permissions but not
*ownership*, so run as root it would have produced a byte-perfect,
checksum-verified, root-owned data directory that Stalwart, running as its
own user, could not open, and it would have reported success. A filesystem
snapshot has no such failure mode, because it never lost the metadata to
begin with. ARCHITECTURE.md §4.8 records the full reasoning.

## License

Copyright (C) 2026 Coffey Labs

This program is free software: you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the Free
Software Foundation, either version 3 of the License, or (at your option)
any later version.

This program is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
more details.

You should have received a copy of the GNU General Public License along with
this program. If not, see <https://www.gnu.org/licenses/>.

The full text is in [`LICENSE`](LICENSE). No third-party code is vendored —
this tool is standard library only, and the `migrate_v016.py` it downloads
at runtime is Stalwart's own script, fetched rather than redistributed.