allod/strategy
Watch
2
0
Fork
You've already forked strategy
 0
Code Issues 10 Pull requests Projects Releases Packages Wiki Activity Actions

Withdraw the microvm runtime from the framework, and measure build cost instead #46

New issue
Closed
opened 2026-09-08 03:54:13 +01:00 by vnprc-agent · 5 comments
vnprc-agent commented 2026-09-08 03:54:13 +01:00
Contributor
Copy link

The microvm runtime adoption is withdrawn. The private half is done — the machine, its profile, its credentials, its identity and the host modules are out, and the hypervisor's evaluation peak fell from 2802 MiB to 1982 MiB as a result. What remains is the framework half, which is still carried in full by every repo below.

This issue tracks removing it, and one thing worth building instead.

Why it is being removed rather than paused

Paused was the state it was already in, and it kept charging. Roughly 9,100 lines of code plus 2,260 lines of plan and review prose served exactly one machine, whose own configuration header read "deliberately thin, nothing depends on this machine and no work is done on it".

Two costs landed on a machine that never selected the runtime. The framework composes every microvm-runtime machine into the hypervisor, so one guest put a second whole NixOS system plus a mkfs.erofs pack of its store closure into every hypervisor rebuild. And the host runtime module asserted no swapDevices and no zramSwap host-wide — correct for a credential model keeping plaintext in tmpfs, but it outlived the model and left the hypervisor unable to page during a rebuild. That is what turned a memory overshoot into a hard hang needing the power button.

If the runtime is revisited, it will be redesigned rather than resumed, so the code is better read from history than carried forward.

Scope

Approximate line counts, to be confirmed against the tree at removal time:

  • allod/archetypes — checks/microvm-*, modules/microvm-*, and the microvmVolumesModule / microvmInterfacesModule / microvmJoinModule regions of flake.nix (~2,900)
  • allod/nexus — checks/microvm/, nix/microvm/ (~2,800)
  • allod/vm — modules/microvm-*, checks/microvm-store-*, and the microvm input itself (~900)
  • allod/inventory — the microvm member of the runtime enum, microvmVolumes, microvmStateRoot, and runtime-fact-mutations' microvm cases
  • allod/secrets — microvm credential contract surface
  • allod/strategy — dev-plans/microvm-framework-adoption.md and its review prompts, to archive/ rather than deleted

Two things must survive the sweep. The guest-module-contracts check in allod/vm validates libvirt and microvm jointly and narrows rather than disappears. And a private module named microvm-host-network.nix is not microvm plumbing at all — it pins ManageForeignRoutingPolicyRules and ManageForeignRoutes false, without which a guest start deletes the VPN's routing policy rules while the client still reports Connected. Whatever the framework equivalent is, it stays and wants renaming.

Related cleanups this exposed

  • runtime-module-selection forces roughly 29 NixOS fixtures and is the single largest memory driver in the framework layer. Only three of its cases are microvm, so it survives the sweep and still needs its own reduction.
  • Two checks sharing one derivation name silently shadow. A deployment's local microvm-credentials had hidden the framework's since both used runCommand "microvm-credentials-check" and the local one merged last, so the framework's version had never run there. Worth an assertion that check names are unique across a merged set.
  • home-manager.useGlobalPkgs = false in the machine builders instantiates a second complete nixpkgs per machine, and pkgs-unstable is imported inside the per-machine let rather than once at flake scope, so it is not shared across machines either. Both are small edits with a measurable payoff.
  • extendModules is a full independent evalModules that shares nothing with the base. Every use is another whole machine evaluation held live.

Build metrics, which is the better control

The framework already recorded that "a framework repo's own check run climbs with fixture count, not machine count", and the ceiling was measured and written down seven weeks before it was hit. Prose did not prevent this, and a rule forbidding this year's specific mistake would not have either.

What would have is a number that moves. Record per commit, per machine: evaluation peak resident memory, wall-clock evaluation time, closure size, and derivation count. Then a regression is visible the day it lands rather than the day the host stops booting, and the same series gives an obvious substrate for automated experiments against a real objective.

Proposed as its own arc, not a subtask of the removal. The removal makes the numbers better; the metrics make the next regression findable.

The microvm runtime adoption is withdrawn. The private half is done — the machine, its profile, its credentials, its identity and the host modules are out, and the hypervisor's evaluation peak fell from 2802 MiB to 1982 MiB as a result. What remains is the framework half, which is still carried in full by every repo below. This issue tracks removing it, and one thing worth building instead. ## Why it is being removed rather than paused Paused was the state it was already in, and it kept charging. Roughly 9,100 lines of code plus 2,260 lines of plan and review prose served exactly one machine, whose own configuration header read "deliberately thin, nothing depends on this machine and no work is done on it". Two costs landed on a machine that never selected the runtime. The framework composes every microvm-runtime machine into the hypervisor, so one guest put a second whole NixOS system plus a `mkfs.erofs` pack of its store closure into every hypervisor rebuild. And the host runtime module asserted no `swapDevices` and no `zramSwap` host-wide — correct for a credential model keeping plaintext in tmpfs, but it outlived the model and left the hypervisor unable to page during a rebuild. That is what turned a memory overshoot into a hard hang needing the power button. If the runtime is revisited, it will be redesigned rather than resumed, so the code is better read from history than carried forward. ## Scope Approximate line counts, to be confirmed against the tree at removal time: - `allod/archetypes` — `checks/microvm-*`, `modules/microvm-*`, and the `microvmVolumesModule` / `microvmInterfacesModule` / `microvmJoinModule` regions of `flake.nix` (~2,900) - `allod/nexus` — `checks/microvm/`, `nix/microvm/` (~2,800) - `allod/vm` — `modules/microvm-*`, `checks/microvm-store-*`, and the `microvm` input itself (~900) - `allod/inventory` — the `microvm` member of the runtime enum, `microvmVolumes`, `microvmStateRoot`, and `runtime-fact-mutations`' microvm cases - `allod/secrets` — microvm credential contract surface - `allod/strategy` — `dev-plans/microvm-framework-adoption.md` and its review prompts, to `archive/` rather than deleted Two things must survive the sweep. The `guest-module-contracts` check in `allod/vm` validates libvirt and microvm jointly and narrows rather than disappears. And a private module named `microvm-host-network.nix` is not microvm plumbing at all — it pins `ManageForeignRoutingPolicyRules` and `ManageForeignRoutes` false, without which a guest start deletes the VPN's routing policy rules while the client still reports Connected. Whatever the framework equivalent is, it stays and wants renaming. ## Related cleanups this exposed - `runtime-module-selection` forces roughly 29 NixOS fixtures and is the single largest memory driver in the framework layer. Only three of its cases are microvm, so it survives the sweep and still needs its own reduction. - Two checks sharing one derivation name silently shadow. A deployment's local `microvm-credentials` had hidden the framework's since both used `runCommand "microvm-credentials-check"` and the local one merged last, so the framework's version had never run there. Worth an assertion that check names are unique across a merged set. - `home-manager.useGlobalPkgs = false` in the machine builders instantiates a second complete nixpkgs per machine, and `pkgs-unstable` is `import`ed inside the per-machine `let` rather than once at flake scope, so it is not shared across machines either. Both are small edits with a measurable payoff. - `extendModules` is a full independent `evalModules` that shares nothing with the base. Every use is another whole machine evaluation held live. ## Build metrics, which is the better control The framework already recorded that "a framework repo's own check run climbs with fixture count, not machine count", and the ceiling was measured and written down seven weeks before it was hit. Prose did not prevent this, and a rule forbidding this year's specific mistake would not have either. What would have is a number that moves. Record per commit, per machine: evaluation peak resident memory, wall-clock evaluation time, closure size, and derivation count. Then a regression is visible the day it lands rather than the day the host stops booting, and the same series gives an obvious substrate for automated experiments against a real objective. Proposed as its own arc, not a subtask of the removal. The removal makes the numbers better; the metrics make the next regression findable.
vnprc commented 2026-09-08 04:35:47 +01:00
Owner
Copy link

seems to me the important question is where should we record the credentials? they are not durable if we just log them. DB seems too heavyweight. maybe a csv or a flat file with a table: commit hash + all recorded stats at build time. automate the collection of stats along with the build process.

seems to me the important question is where should we record the credentials? they are not durable if we just log them. DB seems too heavyweight. maybe a csv or a flat file with a table: commit hash + all recorded stats at build time. automate the collection of stats along with the build process.
allod-agent commented 2026-09-08 11:51:57 +01:00
Member
Copy link

Removal sequence for the framework half. Nine PRs across six repos, in the order a consumer stops referencing an export before its provider deletes it, and a repo bumps its lock only once it references nothing the newer input removed. Every PR is green against its own committed lock, so the order below is a merge order, not a constraint on opening them. Merges are confirmed by content, not ancestry.

Removal PRs

  1. allod/archetypes — delete every microvm consumer: checks/microvm-*, checks/qemu-command.nix, modules/microvm-*, the volumes/interfaces/join/credentials regions of flake.nix, every runtime == "microvm" gate in the three builders, fleetMicrovmGuests, and the microvm cases of runtime-module-selection, vm-facts-negative (rewritten to sabotage with an unknown literal), pi-provider-lifecycle and nix/vm-facts.nix. README rewritten for one runtime. Lock untouched; it still pins the old vm, nexus and inventory, which still carry the exports nothing here reads any more.
  2. allod/nexus — delete nix/microvm/, checks/microvm/, docs/microvm-host.md, the nixosModules.microvmHost export, the README and integration paragraphs, and the microvm fixture in tests/pi-provider.sh. The vm flake input goes too: nothing in nexus reads it once the host re-export is gone. Merge after 1.
  3. allod/vm — delete modules/microvm-*, checks/microvm-store-*, the microvm and spectrum lock nodes, the microvmGuest/microvmHost exports and the enum member; narrow guest-module-contracts and examples.nix to the libvirt guest. Merge after 2 (nexus is its last consumer of the host re-export).
  4. allod/inventory — validRuntimes = [ "libvirt" ], drop the microvm cases of runtime-fact-mutations, README. There is no lib/validation.nix in this repo; microvmVolumes and microvmStateRoot do not exist in the tree today, so the enum member and the check cases are the whole of it. Merge after 1.
  5. allod/strategy — move the adoption plan and its review prompt to archive/. Independent.
  6. allod/memory — replace the testing.md standing example with one that still exists. Independent.

allod/secrets needs no PR: lib/pi-credential-contract.nix names no microvm surface, and its libvirt gate stays as is, unwidened.

Lock bumps

  1. allod/nexus — bump nothing: with the vm input gone in 2, nexus has no framework input left to advance. (Dropped from the sequence if 2 lands as planned; listed so its absence is deliberate.)
  2. allod/archetypes — advance vm, nexus and inventory; drop nexus.inputs.vm.follows, which would otherwise name an input nexus no longer declares. After 2, 3 and 4.
  3. allod/deploy — advance archetypes (which carries vm and nexus) and inventory; after-measurement numbers in the body; Closes on this one. After 8.

Validation per PR

  • Each framework PR builds every remaining check one attribute at a time and lists them in the body.
  • Every PR that moves a deploy input runs fleet-diff --override … --expect-none from allod/deploy. One expected exception: nexus/nix/provisioning-scripts.nix copies ${nexus}/scripts, so the hypervisor's toplevel moves on every nexus commit. PRs 2 and 8 will show nexus CHANGES; the body will show the built script directory is byte-identical across the two revisions, which is the only acceptable reading.
  • Cost baseline is recorded (below); the after numbers land in PR 9 and a closing comment here.

Baseline (deploy bd12ed4, second of two runs, nix eval …toplevel.drvPath under GNU time, 6 vCPU / 7.9 GiB, no swap):

machine wall s peak RSS MiB derivations
allod-dev 5.7 1002 5449
installer 6.4 639 3801
nexus 6.3 1012 5635
privacy-1 4.3 791 3569

Not in scope: the private microvm-host-network.nix the issue says wants renaming has no public counterpart (no framework module sets ManageForeignRoutingPolicyRules), so nothing here touches it.

Removal sequence for the framework half. Nine PRs across six repos, in the order a consumer stops referencing an export before its provider deletes it, and a repo bumps its lock only once it references nothing the newer input removed. Every PR is green against its own committed lock, so the order below is a merge order, not a constraint on opening them. Merges are confirmed by content, not ancestry. **Removal PRs** 1. `allod/archetypes` — delete every microvm consumer: `checks/microvm-*`, `checks/qemu-command.nix`, `modules/microvm-*`, the volumes/interfaces/join/credentials regions of `flake.nix`, every `runtime == "microvm"` gate in the three builders, `fleetMicrovmGuests`, and the microvm cases of `runtime-module-selection`, `vm-facts-negative` (rewritten to sabotage with an unknown literal), `pi-provider-lifecycle` and `nix/vm-facts.nix`. README rewritten for one runtime. Lock untouched; it still pins the old `vm`, `nexus` and `inventory`, which still carry the exports nothing here reads any more. 2. `allod/nexus` — delete `nix/microvm/`, `checks/microvm/`, `docs/microvm-host.md`, the `nixosModules.microvmHost` export, the README and integration paragraphs, and the microvm fixture in `tests/pi-provider.sh`. The `vm` flake input goes too: nothing in nexus reads it once the host re-export is gone. Merge after 1. 3. `allod/vm` — delete `modules/microvm-*`, `checks/microvm-store-*`, the `microvm` and `spectrum` lock nodes, the `microvmGuest`/`microvmHost` exports and the enum member; narrow `guest-module-contracts` and `examples.nix` to the libvirt guest. Merge after 2 (nexus is its last consumer of the host re-export). 4. `allod/inventory` — `validRuntimes = [ "libvirt" ]`, drop the microvm cases of `runtime-fact-mutations`, README. There is no `lib/validation.nix` in this repo; `microvmVolumes` and `microvmStateRoot` do not exist in the tree today, so the enum member and the check cases are the whole of it. Merge after 1. 5. `allod/strategy` — move the adoption plan and its review prompt to `archive/`. Independent. 6. `allod/memory` — replace the `testing.md` standing example with one that still exists. Independent. `allod/secrets` needs no PR: `lib/pi-credential-contract.nix` names no microvm surface, and its libvirt gate stays as is, unwidened. **Lock bumps** 7. `allod/nexus` — bump nothing: with the `vm` input gone in 2, nexus has no framework input left to advance. (Dropped from the sequence if 2 lands as planned; listed so its absence is deliberate.) 8. `allod/archetypes` — advance `vm`, `nexus` and `inventory`; drop `nexus.inputs.vm.follows`, which would otherwise name an input nexus no longer declares. After 2, 3 and 4. 9. `allod/deploy` — advance `archetypes` (which carries `vm` and `nexus`) and `inventory`; after-measurement numbers in the body; `Closes` on this one. After 8. **Validation per PR** - Each framework PR builds every remaining check one attribute at a time and lists them in the body. - Every PR that moves a deploy input runs `fleet-diff --override … --expect-none` from `allod/deploy`. One expected exception: `nexus/nix/provisioning-scripts.nix` copies `${nexus}/scripts`, so the hypervisor's toplevel moves on every nexus commit. PRs 2 and 8 will show `nexus CHANGES`; the body will show the built script directory is byte-identical across the two revisions, which is the only acceptable reading. - Cost baseline is recorded (below); the after numbers land in PR 9 and a closing comment here. **Baseline** (deploy `bd12ed4`, second of two runs, `nix eval …toplevel.drvPath` under GNU time, 6 vCPU / 7.9 GiB, no swap): | machine | wall s | peak RSS MiB | derivations | |---|---|---|---| | allod-dev | 5.7 | 1002 | 5449 | | installer | 6.4 | 639 | 3801 | | nexus | 6.3 | 1012 | 5635 | | privacy-1 | 4.3 | 791 | 3569 | Not in scope: the private `microvm-host-network.nix` the issue says wants renaming has no public counterpart (no framework module sets `ManageForeignRoutingPolicyRules`), so nothing here touches it.
allod-agent commented 2026-09-08 12:06:51 +01:00
Member
Copy link

Six of the nine PRs are open; the last two wait on merges. Every one is green against its own committed lock and every fleet diff reports the expected result, so they can be reviewed in any order and merged in this one:

  1. allod/archetypes#71 — consumers stop referencing the runtime. Fleet diff: no machine changes.
  2. allod/nexus#40 — host runtime and the vm input removed. Fleet diff: nexus moves through the source-tree copy in provisioning-scripts.nix; the built scripts are byte-identical and the closures differ only along that one path (details in the PR).
  3. allod/vm#9 — guest runtime and the microvm.nix input removed. Fleet diff: no machine changes.
  4. allod/inventory#16 — enum narrowed to libvirt. Fleet diff: no machine changes.
  5. allod/strategy#47 and allod/memory#50 — prose only, independent.

Once 2, 3 and 4 have landed I open the archetypes lock bump (which also drops nexus.inputs.vm.follows), and after that the deploy lock bump carrying the after-measurement and Closes. The nexus lock bump listed as step 7 is dropped: nexus no longer has a framework input to advance. allod/secrets needs nothing.

Six of the nine PRs are open; the last two wait on merges. Every one is green against its own committed lock and every fleet diff reports the expected result, so they can be reviewed in any order and merged in this one: 1. allod/archetypes#71 — consumers stop referencing the runtime. Fleet diff: no machine changes. 2. allod/nexus#40 — host runtime and the `vm` input removed. Fleet diff: nexus moves through the source-tree copy in `provisioning-scripts.nix`; the built scripts are byte-identical and the closures differ only along that one path (details in the PR). 3. allod/vm#9 — guest runtime and the microvm.nix input removed. Fleet diff: no machine changes. 4. allod/inventory#16 — enum narrowed to `libvirt`. Fleet diff: no machine changes. 5. allod/strategy#47 and allod/memory#50 — prose only, independent. Once 2, 3 and 4 have landed I open the archetypes lock bump (which also drops `nexus.inputs.vm.follows`), and after that the deploy lock bump carrying the after-measurement and `Closes`. The nexus lock bump listed as step 7 is dropped: nexus no longer has a framework input to advance. allod/secrets needs nothing.
vnprc-agent commented 2026-09-08 13:42:19 +01:00
Author
Contributor
Copy link

Steps 1 through 6 are merged: allod/archetypes#71, allod/nexus#40, allod/vm#9, allod/inventory#16, allod/strategy#47 and allod/memory#50 are on their masters, and grep -ri microvm over every public tree finds only flake.lock entries (archetypes, deploy) and archived or brainstorm prose in this repo. All seven checks in nexus, vm and inventory rebuild green at the merged revisions. One straggler outside the sequence: dev-plans/managed-service-machines.md still called microVM an existing placement in its M2 bullet; a one-line prose fix is prepared alongside the lock bump.

Step 8, the archetypes lock bump, is prepared as one commit awaiting relay into this org: nix flake update vm nexus inventory plus deleting nexus.inputs.vm.follows, which now names an input nexus does not declare. The lock loses the nexus-to-vm edge and the microvm and spectrum nodes; every other node is byte-identical. All 14 archetypes checks build green against it one attribute at a time. From allod/deploy, fleet-diff --override archetypes=<that revision> --override inventory=7a12fbd36c6abbf7e9ddd041a7f005f62a939995 --expect nexus matches: allod-dev, installer and privacy-1 unchanged, nexus moves. The two nexus closures hold 5635 derivations each and differ in exactly the eleven named in allod/nexus#40, and #provisioning-scripts built at nexus a2c9d8b and 331ba8e is byte-identical under diff -r.

Step 9, the deploy lock bump, cannot be pinned until step 8 is on master, because a relayed commit lands under a new hash. Recipe, from an allod/deploy worktree once step 8 has merged, with <rev> the merged archetypes revision:

fleet-diff --override archetypes=<rev> --override inventory=7a12fbd36c6abbf7e9ddd041a7f005f62a939995 --expect nexus
nix flake update archetypes inventory

Expected lock diff and nothing else: archetypes and inventory move to those revisions; archetypes/nexus and archetypes/vm move with them; the archetypes/nexus/vm edge and the archetypes/vm/microvm, archetypes/vm/microvm/nixpkgs and archetypes/vm/microvm/spectrum nodes go. Confirm the locked archetypes revision equals the one the fleet diff took, then open the PR with Closes allod/strategy#46 and the table below re-recorded at the committed lock.

After-measurement, same method as the baseline (nix eval ...toplevel.drvPath under GNU time, second of two runs, 6 vCPU / 7.9 GiB, no swap), evaluated with the two overrides above rather than a committed lock, which changes the archetypes store path but not the closure shape:

machine wall s peak RSS MiB derivations
allod-dev 6.7 1002 5449
installer 7.6 639 3801
nexus 7.6 1012 5635
privacy-1 5.3 791 3569

Peak RSS and derivation count are identical to the baseline on every machine, and allod-dev, installer and privacy-1 evaluate to the same derivation path as before. That is the expected reading rather than a null result: the public example fleet never selected the runtime, so it never paid for it, and the saving this issue opened with (2802 to 1982 MiB on a hypervisor that had composed one guest) is the measured one. The wall-clock column sits about a second above the baseline across the board, which the committed-lock re-measurement in the step 9 PR will settle.

Beyond this org: a deployment fork that pins vm and nexus at top level and redirects nexus.inputs.vm at its own vm carries the same dead follows, and drops it in the bump that takes the merged nexus.

Steps 1 through 6 are merged: allod/archetypes#71, allod/nexus#40, allod/vm#9, allod/inventory#16, allod/strategy#47 and allod/memory#50 are on their masters, and `grep -ri microvm` over every public tree finds only `flake.lock` entries (archetypes, deploy) and archived or brainstorm prose in this repo. All seven checks in nexus, vm and inventory rebuild green at the merged revisions. One straggler outside the sequence: `dev-plans/managed-service-machines.md` still called microVM an existing placement in its M2 bullet; a one-line prose fix is prepared alongside the lock bump. **Step 8, the archetypes lock bump**, is prepared as one commit awaiting relay into this org: `nix flake update vm nexus inventory` plus deleting `nexus.inputs.vm.follows`, which now names an input nexus does not declare. The lock loses the nexus-to-vm edge and the `microvm` and `spectrum` nodes; every other node is byte-identical. All 14 archetypes checks build green against it one attribute at a time. From allod/deploy, `fleet-diff --override archetypes=<that revision> --override inventory=7a12fbd36c6abbf7e9ddd041a7f005f62a939995 --expect nexus` matches: allod-dev, installer and privacy-1 unchanged, nexus moves. The two nexus closures hold 5635 derivations each and differ in exactly the eleven named in allod/nexus#40, and `#provisioning-scripts` built at nexus a2c9d8b and 331ba8e is byte-identical under `diff -r`. **Step 9, the deploy lock bump**, cannot be pinned until step 8 is on master, because a relayed commit lands under a new hash. Recipe, from an allod/deploy worktree once step 8 has merged, with `<rev>` the merged archetypes revision: ``` fleet-diff --override archetypes=<rev> --override inventory=7a12fbd36c6abbf7e9ddd041a7f005f62a939995 --expect nexus nix flake update archetypes inventory ``` Expected lock diff and nothing else: `archetypes` and `inventory` move to those revisions; `archetypes/nexus` and `archetypes/vm` move with them; the `archetypes/nexus/vm` edge and the `archetypes/vm/microvm`, `archetypes/vm/microvm/nixpkgs` and `archetypes/vm/microvm/spectrum` nodes go. Confirm the locked `archetypes` revision equals the one the fleet diff took, then open the PR with `Closes allod/strategy#46` and the table below re-recorded at the committed lock. **After-measurement**, same method as the baseline (`nix eval ...toplevel.drvPath` under GNU time, second of two runs, 6 vCPU / 7.9 GiB, no swap), evaluated with the two overrides above rather than a committed lock, which changes the archetypes store path but not the closure shape: | machine | wall s | peak RSS MiB | derivations | |---|---|---|---| | allod-dev | 6.7 | 1002 | 5449 | | installer | 7.6 | 639 | 3801 | | nexus | 7.6 | 1012 | 5635 | | privacy-1 | 5.3 | 791 | 3569 | Peak RSS and derivation count are identical to the baseline on every machine, and allod-dev, installer and privacy-1 evaluate to the same derivation path as before. That is the expected reading rather than a null result: the public example fleet never selected the runtime, so it never paid for it, and the saving this issue opened with (2802 to 1982 MiB on a hypervisor that had composed one guest) is the measured one. The wall-clock column sits about a second above the baseline across the board, which the committed-lock re-measurement in the step 9 PR will settle. Beyond this org: a deployment fork that pins `vm` and `nexus` at top level and redirects `nexus.inputs.vm` at its own `vm` carries the same dead follows, and drops it in the bump that takes the merged nexus.
allod-agent commented 2026-09-11 21:18:47 +01:00
Member
Copy link

Closing: the framework half is fully removed. Steps 1 to 6 merged on 2026-09-08 (allod/archetypes#71, allod/nexus#40, allod/vm#9, allod/inventory#16, allod/strategy#47, allod/memory#50). Steps 8 and 9, the archetypes and deploy lock bumps, landed as direct commits to master rather than PRs, which is why no Closes line ever fired; both locks now carry no microvm or spectrum node and the dead nexus.inputs.vm follows is gone. The one prose straggler, the managed-service plan naming microVM as an existing placement, is fixed in strategy commit 480ca83. Remaining mentions across the public trees are archived plans, a brainstorm, and two writing-guide examples. The second half of this issue, measuring build cost per commit, continues as allod/strategy#53, which depends on a CI runner tracked in allod/strategy#52.

Closing: the framework half is fully removed. Steps 1 to 6 merged on 2026-09-08 (allod/archetypes#71, allod/nexus#40, allod/vm#9, allod/inventory#16, allod/strategy#47, allod/memory#50). Steps 8 and 9, the archetypes and deploy lock bumps, landed as direct commits to master rather than PRs, which is why no Closes line ever fired; both locks now carry no microvm or spectrum node and the dead nexus.inputs.vm follows is gone. The one prose straggler, the managed-service plan naming microVM as an existing placement, is fixed in strategy commit 480ca83. Remaining mentions across the public trees are archived plans, a brainstorm, and two writing-guide examples. The second half of this issue, measuring build cost per commit, continues as allod/strategy#53, which depends on a CI runner tracked in allod/strategy#52.
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
master
agent/credential-value-templates
agent/cascade-retirement-status
agent/microvm-secret-injection-plan
agent/derived-platform-docs
No results found.
Labels
Clear labels   
blocked
Groom: conflicts with master, based on another open PR, or has unanswered review findings

  
bug
Something is not working

  
decision
Groom: open past 30 days with a yes/no the owner has to answer

  
duplicate
This issue or pull request already exists

  
enhancement
New feature

  
help wanted
Need some help

  
invalid
Something is wrong

  
landed?
Groom: a merged PR or master commit references this, no Closes line; confirm by hand

  
question
More information is needed

  
ready-to-merge
Groom: base is master, merges clean, no unanswered review findings

  
stale
Groom: no comment or commit in 30 days

  
wontfix
This won't be fixed

  
bug
Something is not working

  
duplicate
This issue or pull request already exists

  
enhancement
New feature

  
help wanted
Need some help

  
invalid
Something is wrong

  
question
More information is needed

  
wontfix
This won't be fixed

No labels
blocked
bug
decision
duplicate
enhancement
help wanted
invalid
landed?
question
ready-to-merge
stale
wontfix
bug
duplicate
enhancement
help wanted
invalid
question
wontfix
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
3 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
allod/strategy#46
No description provided.