canic-cli 0.33.0

Operator CLI for Canic fleet backup and restore workflows
Documentation
# canic-cli

`canic-cli` publishes the `canic` operator binary. It is the command-line
surface for building Canic artifacts, installing local Canic fleets, selecting
fleet configs, capturing canister snapshots, validating backup artifacts, and
preparing guarded restores.

The CLI wraps ICP CLI for live snapshot and restore mutations. Canic
owns the topology selection, manifests, journals, readiness checks, restore
ordering, and runner state around those `icp` calls.

`canic-cli` intentionally keeps a narrow Rust library surface: external callers
should treat the installed `canic` binary as the operator interface. Host-side
build/install/fleet helpers live in `canic-host`, and backup/restore contracts
live in `canic-backup`.

## Install

Install from a checkout:

```bash
cargo install --locked --path crates/canic-cli
canic help
```

Install from crates.io after a release:

```bash
cargo install --locked canic-cli --version <version>
```

For a full local development setup, including ICP CLI, helper tools, and the
`canic` CLI, use the install script in the root README.

## First Commands

Show local test-fleet canisters that already have ids:

```bash
canic list test --network local
```

`canic list <name>` reads the installed root registry for that fleet.
Use `--root <name-or-principal>` to point at a specific installed root, or
`--from <name-or-principal>` to print one subtree with that node as the rendered
root.
Live list sources call `canic_ready` for each listed canister and include a
`READY` column with `yes`, `no`, or `error`.

If the list only shows the `root` row, the project has reserved a local root id
but has not installed the tree. Run `canic install test`, then use
`canic list test --network local` to read the installed root registry.

Install and bootstrap the local fleet:

```bash
canic install test
```

Build one Canic canister artifact through the same public CLI surface used by
ICP CLI build hooks:

```bash
canic build root
```

`canic install` defaults to the `root` ICP canister name. You may pass either a
project canister name or an IC principal as the root target:

```bash
canic install test
canic install test root
canic install test uxrrr-q7777-77774-qaaaq-cai
canic install test --root uxrrr-q7777-77774-qaaaq-cai
canic install test --config fleets/test/canic.toml
```

When the root target is a principal, the CLI still builds the conventional
`root` canister artifact by default. Use `--root-build-target <name>` only
when the local root canister is named differently in `icp.yaml`.

When no `--config` is provided, `canic install <name>` uses
`fleets/<name>/canic.toml`.

The selected install config must include a fleet identity:

```toml
[fleet]
name = "test"
```

Successful installs write `.canic/<network>/fleets/<fleet>.json` with the root
target, resolved root principal, build target, config path, and staging
manifest path. `canic config <name>` shows the selected fleet declaration,
while `canic list <name>` queries the deployed root registry for that fleet.
Commands use network `local` unless you pass
`--network <name>`.

List saved fleet configs:

```bash
canic fleet list
canic fleet delete demo
canic fleet list --network ic
```

Create a new root-plus-app fleet:

```bash
canic fleet create my_app --yes
canic install my_app
```

Diagnose the named fleet, replica reachability, saved config path, and root
readiness:

```bash
canic medic test
```

Run command-specific help when you need exact flags:

```bash
canic <command> help
```

The installed CLI version is visible in top-level help and from `canic
--version`. The version flag is accepted at any command depth, so `canic backup
verify --version` reports the binary version instead of running the command.

## Happy Path

Capture a canister and its direct registered children:

```bash
canic snapshot download test \
  --canister <canister-id> \
  --root <root-canister-id> \
  --include-children \
  --out backups/<run-id>
```

Use `--recursive` instead of `--include-children` to include all descendants.
Use `--dry-run` to compute the target set without creating or downloading
snapshots.

Non-dry-run captures recompute the selected topology immediately before
snapshot creation and fail if the topology hash changed since discovery. This
keeps subtree backups from silently crossing a registry change.

ICP CLI creates snapshots only for stopped canisters. Canic stops each canister
before snapshot creation; pass `--resume-after-snapshot` when the CLI should
start each canister again after its artifact is captured.

Verify the captured backup directory:

```bash
canic backup verify \
  --dir backups/<run-id>
```

Verification is no-mutation. It validates the manifest, journal agreement,
durable artifact paths, and checksums before restore planning.

## Backup Checks

Use these commands after capture and before restore planning:

- `canic manifest validate` checks manifest shape, topology hash inputs,
  and backup units.
- `canic backup status` summarizes resumable download journal progress.
- `canic backup verify` validates the backup layout and artifact checksums.

For deeper no-mutation restore checks, use `canic restore plan`,
`canic restore apply --dry-run`, and `canic restore run --dry-run` directly.

## Restore Planning

Restore starts from a manifest, not from loose snapshot files:

```bash
canic restore plan \
  --backup-dir backups/<run-id> \
  --mapping restore-map.json \
  --out restore-plan.json \
  --require-verified \
  --require-restore-ready
```

Planning performs no mutations. It validates mapping, identity mode, snapshot
provenance, verification coverage, artifact checksums when requested, and
restore ordering. Plans include operation counts and parent-before-child
ordering metadata so operators can see the intended restore sequence before any
target is touched.

Render operations and create an apply journal:

```bash
canic restore apply \
  --plan restore-plan.json \
  --backup-dir backups/<run-id> \
  --dry-run \
  --out restore-apply-dry-run.json \
  --journal-out restore-apply-journal.json
```

`restore apply` currently requires `--dry-run`; direct mutation through that
command is intentionally disabled. The generated journal is the input to the
guarded runner.

## Guarded Runner

Preview the maintained runner path without calling `icp`:

```bash
canic restore run \
  --journal restore-apply-journal.json \
  --dry-run \
  --network local \
  --out restore-run-dry-run.json
```

Execute a cautious one-step batch:

```bash
canic restore run \
  --journal restore-apply-journal.json \
  --execute \
  --network local \
  --max-steps 1 \
  --out restore-run.json \
  --require-no-attention
```

The native runner checks journal readiness, claims the next operation, runs the
generated `icp` command, marks the operation completed or failed, and persists
the journal after each transition. `--max-steps 1` is the safest operational
mode while validating a new restore path. Snapshot load operations first run
`icp canister status` and fail before loading unless the target is visibly
stopped.

If a previous runner stopped after claiming work, release the pending operation
back to ready:

```bash
canic restore run \
  --journal restore-apply-journal.json \
  --unclaim-pending \
  --out restore-run-recovery.json
```

## Restore Journal Tools

Use `canic restore run --dry-run` to inspect the journal produced by
`restore apply --dry-run`. The runner preview includes progress, blocked work,
pending claims, failed operations, completion counts, and the next command
preview.

`canic restore run` is also the only maintained command for advancing a restore
journal. It owns command preview, claiming, execution, completion/failure
records, and pending-operation recovery.

## Safety Model

- Directory data may select a root, but topology defines membership.
- Captures fail closed when the selected topology hash changes before snapshot
  creation.
- Backup manifests carry topology, unit, identity, snapshot, artifact,
  provenance, and verification metadata.
- Restore planning is no-mutation and must prove mapping, ordering, checksum,
  verification, and snapshot-restore readiness before execution.
- Runner summaries and journals are durable audit artifacts; failures still
  write status before returning a nonzero exit code.