chad/sweatshop-legacy
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

0048 option C: purge the machine tables — migration 0055 + delete every run read path #126

Merged
chad merged 7 commits from glen/0048-machines-purge into main 2026-10-18 08:19:43 -04:00
Conversation 1 Commits 7 Files changed 43 +2623 -978
glen commented 2026-10-18 08:17:22 -04:00
Collaborator
Copy link

Implements approved doc 0048, option C — the machines purge. Contains a destructive migration (0055); it runs on deploy.

What it does

  • Migration 0055_machines_purge.sql: deletes the 12 run conversations (all 2026-10-10 e2e artifacts — the doc's live-DB evidence) with every dependent row across all 11 FK/linkage tables, sweeps ghost mr_… stage sessions from harness_events/harness_runs, and drops machine_runs + machine_stage_runs.
  • Code cuts: /r/[runId] view + run GET API + db/machines.ts deleted; run arms removed from attention (unread is message-based only), organize (purge walk, runStatus/runParent maps), sandboxes/sandbox-view (run lease arm, run sandbox links), reaper (LIVE_STATUSES), child-status (mr_ → not-found), inspect (mr_ targets, dead stage-strip), conversation-href, sidebar-tree, page-context, preview-list, agent-cards; tool descriptions drop mr_ mentions; AGENTS.md's four machine-era notes rewritten.
  • Tests: 777 → 764 — 13 run-scenario tests deleted with the feature (verified none covered surviving behavior); attach tests repointed to a conversation owner, semantics-equivalent.

Verification

  • Migration proven twice independently: by me (seeded worst-case: run + 2 stages + every dependent kind) and by the fresh-eyes review child (rebuilt through the real migrator, re-checked FK coverage against every drizzle FOREIGN KEY declaration, exercised sandbox_attachments separately). Both: zero tables left, zero orphans, clean PRAGMA foreign_key_check, innocent chats untouched.
  • The review found one should-fix, taken: stage sessions with events but no conversations row (the 0036 case) survived the conversation-keyed DELETE — fixed with OR session_id GLOB 'mr_*' (GLOB's _ is literal) and re-proven with a ghost-session test db. Remaining nits are inert historical-render arms (old transcripts' machine wake chips), kept deliberately.
  • Gates: tsc clean, eslint exactly the 2 pre-existing warnings, 764 passed / 13 skipped.

Deploy note: migration 0055 runs on merge via the Forgejo workflow — irreversible, but the 0048 doc records what's lost (12 e2e runs; the one real conversation was "should a solo developer self-host…", one question and answer). Backups exist in data/backups per the reaper's schedule if you ever want the raw rows.

Implemented by claude (first half) + in-thread (migration, remaining cuts, review fixes); reviewed by a fresh child thread with an independent migration proof.

Implements approved doc 0048, option C — the machines purge. **Contains a destructive migration (0055); it runs on deploy.** ## What it does - **Migration `0055_machines_purge.sql`**: deletes the 12 run conversations (all 2026-10-10 e2e artifacts — the doc's live-DB evidence) with every dependent row across all 11 FK/linkage tables, sweeps ghost `mr_…` stage sessions from harness_events/harness_runs, and **drops `machine_runs` + `machine_stage_runs`**. - **Code cuts**: `/r/[runId]` view + run GET API + `db/machines.ts` deleted; run arms removed from attention (unread is message-based only), organize (purge walk, runStatus/runParent maps), sandboxes/sandbox-view (run lease arm, run sandbox links), reaper (`LIVE_STATUSES`), child-status (`mr_` → not-found), inspect (`mr_` targets, dead stage-strip), conversation-href, sidebar-tree, page-context, preview-list, agent-cards; tool descriptions drop `mr_` mentions; AGENTS.md's four machine-era notes rewritten. - **Tests**: 777 → 764 — 13 run-scenario tests deleted with the feature (verified none covered surviving behavior); attach tests repointed to a conversation owner, semantics-equivalent. ## Verification - Migration proven twice independently: by me (seeded worst-case: run + 2 stages + every dependent kind) and by the fresh-eyes review child (rebuilt through the real migrator, re-checked FK coverage against every drizzle `FOREIGN KEY` declaration, exercised `sandbox_attachments` separately). Both: zero tables left, zero orphans, clean `PRAGMA foreign_key_check`, innocent chats untouched. - The review found **one should-fix, taken**: stage sessions with events but no conversations row (the 0036 case) survived the conversation-keyed DELETE — fixed with `OR session_id GLOB 'mr_*'` (GLOB's `_` is literal) and re-proven with a ghost-session test db. Remaining nits are inert historical-render arms (old transcripts' `machine` wake chips), kept deliberately. - Gates: tsc clean, eslint exactly the 2 pre-existing warnings, **764 passed / 13 skipped**. **Deploy note**: migration 0055 runs on merge via the Forgejo workflow — irreversible, but the 0048 doc records what's lost (12 e2e runs; the one real conversation was "should a solo developer self-host…", one question and answer). Backups exist in `data/backups` per the reaper's schedule if you ever want the raw rows. Implemented by claude (first half) + in-thread (migration, remaining cuts, review fixes); reviewed by a fresh child thread with an independent migration proof.
Option C (approved): the 12 historical machine-run conversations are
e2e-test artifacts with no work worth keeping. Delete /r/<runId> and its
API, the /c redirect into it, hrefFor's and the inspect card's run arms,
the sidebar run glyph render, the preview starter-id run href, and "run"
from every kind enum that routed to it (ConversationKind,
PageContextKind, isItemRunning). run-glyph.tsx and isLiveRun stay — still
live for synthetic harness rows.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
db/machines.ts (getRun, listStageRuns, LIVE_STATUSES, getRunBinding,
MachineError) is deleted — every reader is gone:

- lib/child-status.ts: mr_ ids now return the generic null not-found
  (same shape as any other unknown id).
- lib/agent/tools/inspect.ts: inspectTarget drops the stage-visit
  lookup and the run/stage branches (no conversation can carry those
  kinds again); resolveSandboxTarget and ownerWorkDir drop their
  machine_run arms.
- lib/tool-views/inspect.ts: InspectView.machine and inspectName's run
  branch are dead with it.
- lib/sandboxes/reaper.ts: busySandboxIds drops the LIVE_STATUSES
  machine-run guard; boundToSandbox's runs count is now always 0 (the
  only thing that ever populated it).
- db/sandboxes.ts, db/sandbox-view.ts: the machine_run arms in
  getLeaseProjectId and ownerLabel/ownerHref (the sandbox page's link
  to a dead /r/ route) fall through to the generic cases.
- db/attention.ts: runUnread, the "run" UnreadThread kind, and the
  settled-at unread join are gone — unread is message-based only now.
- db/organize.ts: deleteConversationForever drops the kind==="run"
  glob cascade (no run rows exist to find); loadSidebar's runStatus/
  runParent maps are gone, so a row's own columns are authoritative
  again; conversationParentEdges drops the machine_runs edge.
- db/projects.ts: the delete cascade no longer nulls machine_runs'
  sandbox columns.
- app/api/previews/route.ts: boundSandbox drops the mr_ branch.
- workflows/chat/tools/core.ts: inspect/sandbox_exec descriptions
  drop every mr_ mention.

Tests updated to match (run-scenario cases deleted, non-run cases
kept): child-status, inspect, organize, preview-context, projects,
sandbox-gc, tool-views-info, chat-ui.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
sandboxForMachineRun and sandboxForHarnessSession's run/stage arms are
dead — no conversation can carry those kinds and machine_runs is gone.
ownerLabel keeps treating historical machine_run owners generically
(the live sandboxes rows still say it; nothing new can create one).
attach-sandbox's AttachTarget drops the machine_run owner kind; the
attach tests exercise the generic machinery via a conversation owner.
Deletes every run/stage conversation with all dependent rows (same
teardown order as deleteConversationForever), then drops machine_runs
and machine_stage_runs. The live data is 12 e2e artifacts from
2026-10-10 (the 0048 doc's live-DB evidence); nothing of value is lost.
schema.ts drops both table definitions; 0054_snapshot's isUnique key
normalizes to match drizzle's current emission.

Proven against a seeded worst-case db (run + 2 stage visits + every
dependent kind): after migrate, both tables gone, zero run/stage
conversations, zero orphan mr_ rows, foreign_key_check clean, and the
innocent chat conversation survives with its message.
AGENTS.md's four machine-era notes rewritten (the 0037 phase 5 'what
remains' list, the thread-group parent edge, loadSidebar's run nesting,
the route-storage historical columns). db/schema.ts's kind comment and
db/harness-calls.ts's one-liner drop their references to the dropped
tables. Full suite green at 764.
child-status's route handler, useChildStatus, ChildStatus.href's doc,
and cancel_run's module comment still described machine runs as a kind
of background child or pointed at the deleted run view. Comment-only —
behavior already changed when 0048 purged the tables.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
fix(db,inspect): review fixes — ghost stage sessions swept, dead arms trimmed (0048)
All checks were successful
deploy / checks (pull_request) Successful in 42s
Details
deploy / deploy (pull_request) Has been skipped
Details
402f753f45 
Finding 1 (should-fix): 0055's harness_events/harness_runs DELETEs gain
an OR session_id GLOB 'mr_*' sweep — a stage visit's session id could
exist in harness_events with no conversations row (the 0036-era case),
which the conversation-keyed DELETE missed; GLOB's literal underscore
matches exactly the run-prefixed sessions.

Nit 2: resolveSandboxTarget drops the dead mr_ stage-strip (and its
'or run' error text). Nit 3: ChildStatus.kind loses its producerless
"run" member. The remaining inert arms (runEvent 'machine' chips for
old transcripts, isLiveRun's run case, page-context 'machine') stay —
they render historical rows harmlessly.
chad approved these changes 2026-10-18 08:19:36 -04:00
chad merged commit abebbce175 into main 2026-10-18 08:19:43 -04:00
Sign in to join this conversation.
Reviewers
No reviewers
chad
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
2 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
chad/sweatshop-legacy!126
No description provided.