canic-cli

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

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

Install

Install from a checkout:

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

Install from crates.io after a release:

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

For a full local development setup, including dfx, helper tools, canic-cli, and canic-installer, use the install script in the root README.

First Commands

Show local demo canisters that already have ids:

canic list --network local

By default, canic list checks Canic's fixed demo canister roster and prints a box-drawing canister-id tree for entries that have local dfx ids. Once a project has Canic fleet state, plain canic list reads the installed root registry instead. Use --root <name-or-principal> to point at a specific installed root, --fleet <name> to use a saved fleet without switching, 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, then use canic list --network local to read the installed root registry.

Install and bootstrap the local fleet:

canic install

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

canic install root
canic install uxrrr-q7777-77774-qaaaq-cai
canic install --root uxrrr-q7777-77774-qaaaq-cai
canic install --fleet demo --config canisters/demo/canic.toml

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

canic install uses canisters/canic.toml when that project default exists. If it does not, and other canic.toml files are present, the command prints a small choices table and requires --config <path>.

Successful installs write .canic/<network>/fleets/<fleet>.json with the root target, resolved root principal, build target, config path, and release-set manifest path. canic list uses the current fleet when --root and --fleet are not provided; pass --fleet <name> to query another saved fleet or --root <name-or-principal> to override it.

List and switch saved fleets:

canic fleets --network local
canic use demo --network local

Run command-specific help when you need exact flags:

canic <command> help

Print the installed CLI version with canic --version. The flag is accepted at any command depth, so canic backup preflight --version reports the binary version instead of running the command.

Happy Path

Capture a canister and its direct registered children:

canic snapshot download \
  --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.

dfx creates snapshots only for stopped canisters. Pass --stop-before-snapshot --resume-after-snapshot when the CLI should perform that local lifecycle step around each captured artifact.

Run the standard post-capture smoke wrapper:

canic backup smoke \
  --dir backups/<run-id> \
  --out-dir smoke/<run-id> \
  --require-design \
  --require-restore-ready

Smoke is no-mutation. It writes the preflight report bundle, renders restore operations, creates a restore apply journal, previews the native runner path, and records the readiness flags in smoke-summary.json.

For the release smoke path, use the canonical checklist: docs/operations/0.30-backup-restore-smoke.md.

Backup Checks

Use these commands after capture and before restore planning:

• canic manifest validate checks manifest shape, topology hash inputs, backup units, and design conformance.
• canic backup status summarizes resumable download journal progress.
• canic backup inspect compares manifest and journal metadata without reading artifact bytes.
• canic backup provenance reports source, topology, unit, member, snapshot, code, and artifact provenance.
• canic backup verify reads durable artifacts and verifies checksums.
• canic backup preflight runs the standard no-mutation validation bundle and emits restore planning/status reports.
• canic backup smoke runs preflight plus restore dry-run and runner preview.

The stricter flags intentionally write their reports before returning a nonzero exit code. That lets CI and operators inspect the failure artifact that explains why a backup is not ready.

Restore Planning

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

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

Create the initial restore status:

canic restore status \
  --plan restore-plan.json \
  --out restore-status.json

Render operations and create an apply journal:

canic restore apply \
  --plan restore-plan.json \
  --status restore-status.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 dfx:

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

Execute a cautious one-step batch:

canic restore run \
  --journal restore-apply-journal.json \
  --execute \
  --network local \
  --max-steps 1 \
  --updated-at 2026-05-05T12:03:00Z \
  --out restore-run.json \
  --require-no-attention

The native runner checks journal readiness, claims the next operation, runs the generated dfx 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.

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

canic restore run \
  --journal restore-apply-journal.json \
  --unclaim-pending \
  --updated-at 2026-05-05T12:10:00Z \
  --out restore-run-recovery.json

Restore Journal Tools

These commands inspect the journal produced by restore apply --dry-run:

• canic restore apply-status summarizes progress, blocked work, pending claims, failed operations, and completion counts.
• canic restore apply-report writes an operator-focused report for the work needing attention.

canic restore run is 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 design-conformance readiness before execution.
• Runner summaries and journals are durable audit artifacts; failures still write status before returning a nonzero exit code.