system: validate installed rollback workflow

This commit is contained in:
2026-04-05 01:39:24 +02:00
parent b3b1ba2489
commit 9dae4e5c84
9 changed files with 1170 additions and 75 deletions

View File

@@ -11,9 +11,10 @@ This document defines the current canonical Fruix workflow for:
- installing a declarative system onto an image or disk
- booting through installer media
- rolling forward to a candidate system
- rolling back to an earlier declared system
- switching an installed system to a staged candidate generation
- rolling an installed system back to an earlier recorded generation
This is the Phase 19.1 operator-facing view of the system model already implemented in earlier phases.
This is now the Phase 19 operator-facing view of the system model as validated through explicit installed-system generation switching and rollback.
## Core model
@@ -168,6 +169,34 @@ Use this when you want to:
- install from an ISO-attached Fruix environment
- test the same install model on more realistic VM paths
### Installed-system generation commands
Installed Fruix systems now also ship a small in-guest deployment helper at:
- `/usr/local/bin/fruix`
Current validated in-guest commands are:
```sh
fruix system status
fruix system switch /frx/store/...-fruix-system-...
fruix system rollback
```
Current intended usage:
1. build a candidate closure on the operator side with `./bin/fruix system build`
2. ensure that candidate closure is present on the installed target's `/frx/store`
3. run `fruix system switch /frx/store/...` on the installed system
4. reboot into the staged candidate generation
5. if needed, run `fruix system rollback`
6. reboot back into the recorded rollback generation
Important current limitation:
- `fruix system switch` does **not** yet fetch or copy the candidate closure onto the target for you
- it assumes the selected closure is already present in the installed system's `/frx/store`
## Deployment patterns
### 1. Build-first workflow
@@ -261,7 +290,7 @@ Installed Fruix systems now record an explicit first-generation deployment layou
- `/var/lib/fruix/system`
Current validated shape:
Initial installed shape:
```text
/var/lib/fruix/system/
@@ -275,6 +304,24 @@ Current validated shape:
install.scm # present on installed targets
```
After a validated in-place switch, the layout extends to:
```text
/var/lib/fruix/system/
current -> generations/2
current-generation
rollback -> generations/1
rollback-generation
generations/
1/
...
2/
closure -> /frx/store/...-fruix-system-...
metadata.scm
provenance.scm
install.scm # deployment metadata for the switch operation
```
Installed systems also now create explicit GC-root-style deployment links under:
- `/frx/var/fruix/gcroots`
@@ -284,7 +331,9 @@ Current validated shape:
```text
/frx/var/fruix/gcroots/
current-system -> /frx/store/...-fruix-system-...
rollback-system -> /frx/store/...-fruix-system-...
system-1 -> /frx/store/...-fruix-system-...
system-2 -> /frx/store/...-fruix-system-...
```
Important detail:
@@ -294,7 +343,9 @@ Important detail:
## Roll-forward workflow
The current Fruix roll-forward model is declaration-driven.
The current Fruix roll-forward model now has two validated layers.
### Declaration/deployment roll-forward
Canonical process:
@@ -314,47 +365,61 @@ Canonical process:
5. boot or install the candidate
6. validate the candidate closure in the booted system
The important property is that the candidate closure appears beside the earlier one in `/frx/store` rather than mutating it in place.
### Installed-system generation roll-forward
When the candidate closure is already present on an installed target:
1. run `fruix system switch /frx/store/...candidate...`
2. confirm the staged state with `fruix system status`
3. reboot into the candidate generation
4. validate the new active closure after reboot
The important property is still that the candidate closure appears beside the earlier one in `/frx/store` rather than mutating it in place.
## Rollback workflow
The current canonical rollback workflow is also declaration-driven.
The current canonical rollback workflow also now has two validated layers.
Today, rollback means:
### Declaration/deployment rollback
You can still roll back by redeploying the earlier declaration:
1. retain the earlier declaration that produced the known-good closure
2. rebuild or rematerialize that earlier declaration
3. redeploy or reboot that earlier artifact again
Concretely, the usual rollback choices are:
Concretely, the usual declaration-level rollback choices are:
- rebuild the earlier declaration with `fruix system build` and confirm the old closure path reappears
- boot the earlier declaration again through `fruix system image`
- reinstall the earlier declaration through `fruix system install`, `installer`, or `installer-iso` if the deployment medium itself must change
This rollback story has already been validated at the closure/image/deployment level:
### Installed-system generation rollback
- side-by-side base-version coexistence in `/frx/store`
- roll-forward to a candidate closure
- rollback by rebuilding and booting the earlier declaration again
- validation on both local QEMU and the approved XCP-ng VM path
When an installed target already has both the current and rollback generations recorded:
1. run `fruix system rollback`
2. confirm the staged state with `fruix system status`
3. reboot into the rollback generation
4. validate the restored active closure after reboot
This installed-system rollback path is now validated on local `QEMU/UEFI/TCG`.
### Important scope note
This is not yet the same thing as a first-class installed-system generation switch command.
This is still not yet the same thing as Guix's full `reconfigure`/generation UX.
Current rollback is:
Current installed-system rollback is intentionally modest:
- **redeploy the earlier declaration again**
What still remains for later Phase 19 work is making rollback itself operator-driven at the installed-system layer, rather than only declaration/redeploy driven.
- it switches between already-recorded generations on the target
- it does not yet fetch candidate closures onto the machine for you
- it does not yet expose a richer history-management or generation-pruning policy
Still pending:
- previous-generation tracking beyond the initial explicit generation-1 layout
- an explicit rollback target link distinct from `current`
- an operator-facing installed-system rollback workflow
- generation switching without full redeploy
- operator-facing closure transfer or fetch onto installed systems
- multi-generation lifecycle policy beyond the validated `current` and `rollback` pointers
- a fuller `reconfigure`-style installed-system UX
## Provenance and deployment identity
@@ -375,16 +440,16 @@ Operators should retain metadata from successful candidate and current deploymen
## Current limitations
The deployment workflow is now coherent, but it is not yet the final generation-management story.
The deployment workflow is now coherent, and Fruix now has a validated installed-system switch/rollback path, but it is still not the final generation-management story.
Not yet first-class:
- a dedicated `switch` or `reconfigure` command
- an installed-system rollback command that moves between generations in place
- multi-generation retention and previous-generation tracking beyond generation 1
- generation switching policy independent of full redeploy
- host-side closure transfer/fetch onto installed systems as part of `fruix system switch`
- a fuller `reconfigure` workflow that builds and stages the new closure from inside the target environment
- multi-generation lifecycle policy beyond the validated `current` and `rollback` pointers
- generation pruning and retention policy independent of full redeploy
Those are the next logical steps after the current explicit-generation layout.
Those are the next logical steps after the current explicit-generation switch/rollback model.
## Summary
@@ -395,6 +460,8 @@ The current canonical Fruix deployment model is:
- **materialize** the artifact appropriate to the deployment target
- **boot or install** that artifact
- **identify deployments by closure path and provenance metadata**
- **roll back by rebuilding/redeploying the earlier declaration**, not by mutating the current closure in place
- on installed systems, **switch** to a staged candidate with `fruix system switch`
- on installed systems, **roll back** to the recorded rollback generation with `fruix system rollback`
- still use declaration/redeploy rollback when the target does not already have the desired closure staged locally
That is the operator-facing workflow Fruix should document and use while installed-system generation switching remains more limited than Guix's mature in-place system-generation workflow.
That is the operator-facing workflow Fruix should document and use while its installed-system generation UX remains simpler than Guix's mature in-place system-generation workflow.