pdurlej/platform

Fork 0

ops(rs2000-legacy): destructive gate delete legacy backups after zero deps and 48h green #557

New issue

Closed

opened 2026-05-28 01:37:57 +02:00 by codex · 3 comments

codex commented

2026-05-28 01:37:57 +02:00

Collaborator

Scope

This is the explicit destructive cleanup gate for /opt/vps-home-platform-infra/backups on RS2000.

It must not run until all validation gates pass and the operator explicitly approves the destructive action.

Current observed size on 2026-05-28

/opt/vps-home-platform-infra/backups: 213G
/opt/pdurlej-platform/backups: 5.6G

Required preconditions

All legacy bind-mount migration issues are closed or explicitly deferred with safe reason.
ops(rs2000-legacy): validate zero legacy bind mounts and green smokes is closed green.
docker inspect shows zero running container mounts under /opt/vps-home-platform-infra.
Latest platform smoke green.
Zero unhealthy containers.
48h green after zero legacy bind mounts.
Canonical backup path and retention status are verified.
W3d/restore-confidence status is accepted or explicitly waived by operator.
Operator writes exact approval phrase: m01-destructive-cleanup-approved.

Allowed destructive action

Only after the preconditions above:

delete /opt/vps-home-platform-infra/backups;
record timestamp, pre-delete size, post-delete free space, command class;
do not print secret-bearing command output.

Stop conditions

Any legacy bind mount remains.
Any unhealthy production container appears.
Any smoke fails.
Any backup/restore uncertainty remains.
Operator approval phrase is absent.

Out of scope

Deleting /opt/vps-home-platform-infra/data.
Deleting /opt/vps-home-platform-infra/env.
Deleting /opt/vps-home-platform-infra/state or retired control-plane material unless a later issue explicitly scopes it.
Pruning Docker images/volumes.

Common safety rules

Recommended executor: Gemini 3.5 Flash local for repo/doc/plan work.
Do not delete, rename, prune, or edit /opt/vps-home-platform-infra in this issue.
Do not print secrets, env values, private messages, emails, or Iskra memory.
Runtime execution requires a separate explicit operator approval in the PR/issue thread.
First step is always read-only evidence refresh.
Preserve ownership, modes, symlinks, and timestamps when copying data/config.
Update compose/module docs through a PR before recreating affected services.
Recreate only affected service(s), never the whole platform, unless a later operator gate says otherwise.

## Scope This is the explicit destructive cleanup gate for `/opt/vps-home-platform-infra/backups` on RS2000. It must not run until all validation gates pass and the operator explicitly approves the destructive action. ## Current observed size on 2026-05-28 - `/opt/vps-home-platform-infra/backups`: 213G - `/opt/pdurlej-platform/backups`: 5.6G ## Required preconditions - All legacy bind-mount migration issues are closed or explicitly deferred with safe reason. - `ops(rs2000-legacy): validate zero legacy bind mounts and green smokes` is closed green. - `docker inspect` shows zero running container mounts under `/opt/vps-home-platform-infra`. - Latest platform smoke green. - Zero unhealthy containers. - 48h green after zero legacy bind mounts. - Canonical backup path and retention status are verified. - W3d/restore-confidence status is accepted or explicitly waived by operator. - Operator writes exact approval phrase: `m01-destructive-cleanup-approved`. ## Allowed destructive action Only after the preconditions above: - delete `/opt/vps-home-platform-infra/backups`; - record timestamp, pre-delete size, post-delete free space, command class; - do not print secret-bearing command output. ## Stop conditions - Any legacy bind mount remains. - Any unhealthy production container appears. - Any smoke fails. - Any backup/restore uncertainty remains. - Operator approval phrase is absent. ## Out of scope - Deleting `/opt/vps-home-platform-infra/data`. - Deleting `/opt/vps-home-platform-infra/env`. - Deleting `/opt/vps-home-platform-infra/state` or retired control-plane material unless a later issue explicitly scopes it. - Pruning Docker images/volumes. ## Common safety rules - Recommended executor: Gemini 3.5 Flash local for repo/doc/plan work. - Do not delete, rename, prune, or edit `/opt/vps-home-platform-infra` in this issue. - Do not print secrets, env values, private messages, emails, or Iskra memory. - Runtime execution requires a separate explicit operator approval in the PR/issue thread. - First step is always read-only evidence refresh. - Preserve ownership, modes, symlinks, and timestamps when copying data/config. - Update compose/module docs through a PR before recreating affected services. - Recreate only affected service(s), never the whole platform, unless a later operator gate says otherwise.

codex added this to the 01 - RS2000 closeout and legacy cleanup milestone

2026-05-28 01:37:57 +02:00

codex added the

class/security-sensitive

labels

2026-05-28 01:37:57 +02:00

codex referenced this issue

2026-05-29 01:48:40 +02:00

ops(rs2000-legacy): validate zero legacy bind mounts and green smokes #556

codex commented

2026-05-29 01:48:41 +02:00

Author

Collaborator

M01 zero-bind work is now complete enough for the destructive gate to be the only remaining M01 cleanup decision.

Current state as of 2026-05-28T23:45Z:

Docker legacy bind mounts under /opt/vps-home-platform-infra: 0 total, 0 running, 0 exited.
Unhealthy containers: 0.
/proc/*/mountinfo references to /opt/vps-home-platform-infra: 0.
platformctl validate all --json: green, 88 modules, 0 failed.
Legacy source paths and backups have not been deleted.
Runtime rollback snapshots for the Docker API remount pass are stored under /opt/pdurlej-platform/runtime/m01-container-snapshots with mode 0600.

Still required before destructive deletion, unless operator explicitly waives the wait:

decide whether the post-zero-bind observation/quarantine period is satisfied or intentionally waived;
operator must provide exact phrase: m01-destructive-cleanup-approved.

Allowed destructive target remains only the approved legacy backup cleanup scoped in this issue. Do not delete data/env/source trees without a separate issue.

M01 zero-bind work is now complete enough for the destructive gate to be the only remaining M01 cleanup decision. Current state as of 2026-05-28T23:45Z: - Docker legacy bind mounts under `/opt/vps-home-platform-infra`: `0` total, `0` running, `0` exited. - Unhealthy containers: `0`. - `/proc/*/mountinfo` references to `/opt/vps-home-platform-infra`: `0`. - `platformctl validate all --json`: green, 88 modules, 0 failed. - Legacy source paths and backups have not been deleted. - Runtime rollback snapshots for the Docker API remount pass are stored under `/opt/pdurlej-platform/runtime/m01-container-snapshots` with mode 0600. Still required before destructive deletion, unless operator explicitly waives the wait: - decide whether the post-zero-bind observation/quarantine period is satisfied or intentionally waived; - operator must provide exact phrase: `m01-destructive-cleanup-approved`. Allowed destructive target remains only the approved legacy backup cleanup scoped in this issue. Do not delete data/env/source trees without a separate issue.

codex referenced this issue

2026-05-29 01:50:26 +02:00

docs(cutover): record M01 zero-bind validation #594

codex referenced this issue

2026-05-29 02:22:37 +02:00

docs(cutover): record M01 host-ops gate validation #595

codex commented

2026-05-29 02:23:46 +02:00

Author

Collaborator

M01 host-ops gate red-team closeout update from codex.

DeepSeek V4 Pro first pass returned YELLOW_NEED_ONE_MORE_CHECK and correctly caught that host-level systemd timers still referenced /opt/vps-home-platform-infra even though Docker/mountinfo was clean.

That gap is now closed:

cron/systemd references to /opt/vps-home-platform-infra: 0
host-ops script references to /opt/vps-home-platform-infra: 0
Docker legacy mounts: 0 total/running/exited
unhealthy containers: 0
platformctl validate all --json: exitCode 0, 88 modules, 0 failed
hp-backup-critical.service: success from new host-ops root
hp-restore-smoke.service: success from new host-ops root
hp-backup-noncritical.service: success from new host-ops root
DeepSeek V4 Pro second pass: GREEN_TO_REQUEST_OPERATOR_PHRASE

Evidence is merged in state/reports/m01-host-ops-gate-validation-2026-05-29.md via PR #595.

Still not performed: destructive cleanup. This issue remains gated on Piotr's exact phrase m01-destructive-cleanup-approved, and scope remains limited to legacy backup data covered here. Live data/env/source trees stay out of scope.

M01 host-ops gate red-team closeout update from codex. DeepSeek V4 Pro first pass returned `YELLOW_NEED_ONE_MORE_CHECK` and correctly caught that host-level systemd timers still referenced `/opt/vps-home-platform-infra` even though Docker/mountinfo was clean. That gap is now closed: - cron/systemd references to `/opt/vps-home-platform-infra`: `0` - host-ops script references to `/opt/vps-home-platform-infra`: `0` - Docker legacy mounts: `0` total/running/exited - unhealthy containers: `0` - `platformctl validate all --json`: exitCode `0`, 88 modules, 0 failed - `hp-backup-critical.service`: success from new host-ops root - `hp-restore-smoke.service`: success from new host-ops root - `hp-backup-noncritical.service`: success from new host-ops root - DeepSeek V4 Pro second pass: `GREEN_TO_REQUEST_OPERATOR_PHRASE` Evidence is merged in `state/reports/m01-host-ops-gate-validation-2026-05-29.md` via PR #595. Still not performed: destructive cleanup. This issue remains gated on Piotr's exact phrase `m01-destructive-cleanup-approved`, and scope remains limited to legacy backup data covered here. Live data/env/source trees stay out of scope.

codex referenced this issue

2026-05-29 02:23:47 +02:00

fix(maintenance): reconcile ops_maintenance_queue schema after host-ops migration #596

codex referenced this issue

2026-05-29 02:30:52 +02:00

docs(cutover): record M01 legacy backup cleanup #597

pdurlej referenced this issue from a commit

2026-05-29 02:31:20 +02:00

docs(cutover): record M01 legacy backup cleanup

codex commented

2026-05-29 02:32:46 +02:00

Author

Collaborator

Codex cleanup complete.

Sanitized evidence:

Approved target deleted: /opt/vps-home-platform-infra/backups only.
Pre-delete size: 222G.
/opt available space: ~69G before, ~290G after validation.
Post-delete checks: legacy target absent, legacy mount refs 0, cron/systemd legacy refs 0, unhealthy containers 0.
Canonical backups remained present: latest critical and non-critical backups under /opt/pdurlej-platform/runtime/host-ops/backups.
platformctl validate all --json passed after cleanup.
Evidence PR merged: #597.

Out of scope for this issue remains untouched: legacy env/data/state/control-plane remnants and the separate maintenance schema follow-up (#596).

Codex cleanup complete. Sanitized evidence: - Approved target deleted: `/opt/vps-home-platform-infra/backups` only. - Pre-delete size: 222G. - `/opt` available space: ~69G before, ~290G after validation. - Post-delete checks: legacy target absent, legacy mount refs 0, cron/systemd legacy refs 0, unhealthy containers 0. - Canonical backups remained present: latest critical and non-critical backups under `/opt/pdurlej-platform/runtime/host-ops/backups`. - `platformctl validate all --json` passed after cleanup. - Evidence PR merged: #597. Out of scope for this issue remains untouched: legacy env/data/state/control-plane remnants and the separate maintenance schema follow-up (#596).

codex closed this issue

2026-05-29 02:32:53 +02:00

codex referenced this issue

2026-05-29 08:19:23 +02:00

chore(legacy): inventory non-backup RS2000 remnants after M01 closeout #598

codex referenced this issue

2026-05-29 08:19:26 +02:00

ops(legacy): plan non-backup legacy env/data/state disposal after inventory #599

codex referenced this issue

2026-05-29 08:19:28 +02:00

fix(legacy): verify stale cleanpush env symlink is unused before cleanup #600