Repository navigation
fix(server): CLI commands no longer migrate the database of a running server - #26763
Conversation
…#26696) The test holds the T3 home's ownership lock and seeds the database one migration behind the build. It then builds EnvironmentAuth.layerRuntime, the storage layer that t3 auth, pair, connect, thread and project share. On main the layer applies the pending migration behind the owner. The test expects the layer to fail with a "server owns" error and the ledger to stay where it was. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…er owns `t3 auth`, `pair`, `connect`, `thread`, `project` and `environment` open the T3 home's database through the server's persistence layer, which applies this build's pending migrations unconditionally. A newer CLI next to an older running server therefore rewrote the schema that server runs on, and the next older build refused to start. CLI commands now use `SqlitePersistence.layerCliConfig`. When this build has migrations to run (no ledger, an id above the highest recorded, or the V2-preview reconciliation), it holds the T3 home's ownership lock for the migration and fails with `ServerOwnsDatabaseError` if a running server holds it. With nothing pending it takes no lock, so a starting server is never refused. The server and the offline `t3 project` runtime, which already own the home, keep the plain layer. `holdServerOwnershipLock` is the lock half of `acquireServerOwnership`, without the runtime-state claim and pre-lock heuristics. Importing it from persistence closed a module cycle through `serverRuntimeState`'s use of `startupAccess`, so the two host-string helpers it needed move to `@t3tools/shared/hostClassification`. Fixes #26696 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review follow-up: with nothing pending, run the read-only migration path so divergent histories are still reported, and name any T3 Code process as the possible lock holder. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ApprovabilityVerdict: Approved at Macroscope's review found this PR approvable — CLI commands now require database ownership before applying pending migrations, preventing schema changes behind a running server while preserving current-schema access. Existing server migration behavior remains unchanged, and the reported runtime checks cover both refusal and normal-access paths. You can add or adjust custom eligibility rules. Learn more. |
Thread transfer impact✅ Thread transfer remains within every enforced ceiling.
Baseline: Scenario and decoded snapshot size10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.
Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed. |
|
No actionable comments were generated in the recent review. Walkthrough
Merge Risk:
|
# Conflicts: # apps/server/src/auth/EnvironmentAuth.test.ts # apps/server/src/cli/environment.ts # apps/server/src/environment/DirectEndpoints.ts # apps/server/src/persistence/Sqlite.ts # apps/server/src/serverOwnership.ts
The project CLI added on main opened the database through the server's persistence layer, so it could still migrate a home a running server owns. Move the CLI ownership tests to Sqlite.test.ts against layerCliConfig itself, each with its own T3 home. AI-assisted; verified with the tests and tools listed above.
|
The automated pipeline stopped on this pull request and will not merge it: someone else pushed to fix/issue-26696 (head is now 866a3d9); the pipeline does not build on commits it did not review It stays open for a person to finish or close. |
The offline project runtime is built after the command acquires the server ownership lock, so layerCliConfig would contend with that same lock if a migration were pending. Use the plain layer there again. The deferred-imports test no longer asserts that server-owner.sqlite is absent: the CLI's migration lock now creates it on a fresh home. The runtime-state assertion still proves nothing was cleared. AI-assisted; verified with the tests and tools listed above.
Dismissing prior approval to re-evaluate 111f112
EvidenceStarted a real server on a fresh disposable home on each commit. Matching-manifest session issuance and pairing succeeded on both. Appended harmless CLI-only migration 81, then ran session issuance, pairing, and environment listing against the running server. Finally stopped the server and attempted startup with a conflicting migration 81.
Environment: Linux, real source CLI and HTTP server, fresh synthetic homes. Images and videos render captured transcripts; they are not screen recordings. Credential output was withheld. Reproduction driver. Evidence uploaded using the ghdrop skill. AI-assisted; verified with the tests and tools listed above. Not shown:
|
**v0.1.10-nightly.20261111.4529** · 54 changes from 29 contributors - **Find in browser.** Search pages in the desktop integrated browser with Cmd+F or Ctrl+F. - **GitLab overview threads.** Resolve merge request overview threads from Summary and Timeline. - **Faster worktree updates.** Checkpoint captures and status refreshes avoid rereading every file in newly created worktrees. - **Imported session history.** Imported agent sessions now show their existing history instead of opening as empty threads. - **SSH reconnect prompts.** Expired SSH verification prompts no longer stack up and force an app restart. - **Large reasoning traces.** Opening a thread with a large collapsed reasoning trace no longer freezes the app. - **Terminal scrollback.** Terminal history retains substantially more output instead of silently discarding older lines. ### Plus 45 more bug fixes and performance improvements. <details> <summary>Full changelog</summary> ## What's Changed * fix(server): new worktrees refresh their index once, so captures and status stop re-hashing every file by @josephv123 in pingdotgg/t3code#21341 * fix(usage): Grok and Cursor usage limits show pace and the even-spend line by @wookiist in pingdotgg/t3code#21897 * fix(web): retain diff selection across navigation by @aleo113 in pingdotgg/t3code#22347 * fix(mobile): subscription widget no longer expires on a window that already reset by @juliusmarminge in pingdotgg/t3code#22993 * fix(opencode): verify CLI updates when loopback server is still running by @intrepid37 in pingdotgg/t3code#22839 * fix(usage): Claude subagent output is no longer undercounted by @rfrost-xyz in pingdotgg/t3code#20276 * fix(server): a remote update points the service unit at the runtime it committed by @Marve10s in pingdotgg/t3code#19652 * fix(server): a limited provider's queue stays held after another provider's turn succeeds by @Marve10s in pingdotgg/t3code#22425 * fix(shared): keep spreadsheet shorthand as inline code by @juliusmarminge in pingdotgg/t3code#23203 * fix(web): offer retry while an environment reconnects by @juliusmarminge in pingdotgg/t3code#23194 * fix(server): custom Codex providers no longer report native quota by @fixfon in pingdotgg/t3code#18622 * fix(desktop): Cmd+F now searches the page in the integrated browser by @armanarutiunov in pingdotgg/t3code#22960 * fix(desktop): an SSH password prompt that ended leaves the dialog instead of stacking by @Umais-Adeed in pingdotgg/t3code#34192 * fix(web): render Claude insight blocks as a callout by @macodev00 in pingdotgg/t3code#22239 * fix(scheduler): show short server timezone on fixed-time labels by @shawk08033 in pingdotgg/t3code#23111 * fix(web): focus the primary action when confirmation dialogs open by @rgrvlsk in pingdotgg/t3code#18187 * fix(web): Changes row matches the diff comparison by @whoisaldo in pingdotgg/t3code#19553 * fix(server): imported ACP sessions show their native history by @only21mil in pingdotgg/t3code#22155 * fix(preview): always return evaluated results by value by @juliusmarminge in pingdotgg/t3code#22948 * fix(provider-core): a model reply of "title" no longer replaces the thread title by @juliusmarminge in pingdotgg/t3code#25743 * fix(desktop,web): Open in runs locally for a background service sharing the desktop's T3 home by @KinanLak in pingdotgg/t3code#21661 * fix(web): question answers submit while normal sends are blocked by @t3dotgg in pingdotgg/t3code#22874 * fix(server): cancelling a background-only delegated task reports it interrupted by @only21mil in pingdotgg/t3code#21676 * fix(web): fall back to a readable server for shared repositories by @sanjaibalajee in pingdotgg/t3code#21530 * fix(web): composer follows an optionless thread instead of the settings default by @f4llenz in pingdotgg/t3code#16546 * fix(web): don't save retention age until committed by @macodev00 in pingdotgg/t3code#22208 * fix(server): one dropped probe no longer lists a live linked peer as unreachable by @juliusmarminge in pingdotgg/t3code#34208 * fix(server): preview automation reports ambiguous selectors instead of calling them invalid by @juliusmarminge in pingdotgg/t3code#34205 * fix(opencode): a context overflow that OpenCode compacts no longer fails the turn by @juliusmarminge in pingdotgg/t3code#26743 * fix(web): Ctrl + "+" zooms on keyboard layouts with a dedicated plus key by @juliusmarminge in pingdotgg/t3code#26408 * fix(ci): release version bump push retries when main moves by @t3dotgg in pingdotgg/t3code#34326 * fix(server): CLI commands no longer migrate the database of a running server by @juliusmarminge in pingdotgg/t3code#26763 * fix(web): keep saved default model when changing effort by @macodev00 in pingdotgg/t3code#22193 * fix(web): reveal expanded work rows hidden behind the composer by @ln-bodicherla in pingdotgg/t3code#21968 * fix(composer): steer instead of queueing during automatic compaction by @NikitaMGrimm in pingdotgg/t3code#24138 * feat(gitlab): resolve overview threads from Summary and Timeline by @TonybynMp4 in pingdotgg/t3code#30575 * fix(web): pull request search keeps partial-word matches after GitHub answers by @juliusmarminge in pingdotgg/t3code#26992 * fix(server): Claude subagents no longer show the parent's effort by @juansuaya17 in pingdotgg/t3code#17042 * fix(server): runs cancelled by a server restart say why in the thread by @juliusmarminge in pingdotgg/t3code#27192 * fix(client-runtime): server updates stop waiting 32 minutes when the host returns without this update's outcome by @juliusmarminge in pingdotgg/t3code#27275 * fix(web): Stop shows a pending state and repeat clicks no longer resume by @voida6 in pingdotgg/t3code#20809 * fix(web): collapsed reasoning previews no longer process the whole block by @juliusmarminge in pingdotgg/t3code#23001 * fix(server): storage cleanup skips stale worktree paths under a regular file by @juliusmarminge in pingdotgg/t3code#26594 * fix(terminal): use a byte budget for Ghostty scrollback by @zero-two-rafaeltab in pingdotgg/t3code#8766 * fix(server): storage cleanup keeps worktrees whose file listing exceeds the inspection limit by @juliusmarminge in pingdotgg/t3code#26593 * fix(server): delete temporary ACP health probe sessions by @juliusmarminge in pingdotgg/t3code#25397 * perf(client): streamed text no longer republishes thread queue and checkpoints by @juliusmarminge in pingdotgg/t3code#27425 * fix(web): convert composer list markers completed later by @juliusmarminge in pingdotgg/t3code#22997 * test: remove tests that repeat another case or restate the source by @shivamhwp in pingdotgg/t3code#34170 * fix(server): drop stale queued pull-request watch wakes by @Sinan53x in pingdotgg/t3code#21239 * fix(web): preview recordings stay decodable after a viewport resize by @MansTomb in pingdotgg/t3code#10495 * fix(server): only offer claude update for native Claude installs by @AI091 in pingdotgg/t3code#14170 * fix(server): tell outside agents which MCP tools need a thread and that t3_thread_list takes allProjects by @juliusmarminge in pingdotgg/t3code#26525 ## New Contributors * @wookiist made their first contribution in pingdotgg/t3code#21897 * @aleo113 made their first contribution in pingdotgg/t3code#22347 * @intrepid37 made their first contribution in pingdotgg/t3code#22839 * @rfrost-xyz made their first contribution in pingdotgg/t3code#20276 * @armanarutiunov made their first contribution in pingdotgg/t3code#22960 * @KinanLak made their first contribution in pingdotgg/t3code#21661 * @sanjaibalajee made their first contribution in pingdotgg/t3code#21530 * @ln-bodicherla made their first contribution in pingdotgg/t3code#21968 * @juansuaya17 made their first contribution in pingdotgg/t3code#17042 * @zero-two-rafaeltab made their first contribution in pingdotgg/t3code#8766 * @MansTomb made their first contribution in pingdotgg/t3code#10495 * @AI091 made their first contribution in pingdotgg/t3code#14170 **Full Changelog**: pingdotgg/t3code@v0.1.9-nightly.20261111.4515...v0.1.10-nightly.20261111.4529 </details> Upstream release: https://github.com/pingdotgg/t3code/releases/tag/v0.1.10-nightly.20261111.4529


Problem
CLI commands that open the T3 home's database (
t3 auth …,t3 pair,t3 thread …,t3 connect …,t3 project …,t3 environment …) ran their own build's pending migrations even while a running server owned that home. A newer CLI next to an older server moved the server's schema under it. A later server build that ships a different migration with the same id then refused to start ("would skip required schema migrations").Reproduced on main 5b89a53:
073_CliOnlyProbeto the checkout.t3 auth session issue,t3 pair --ttl 5mort3 environment listwith--base-dirset to that home.Each command exited 0 and the ledger moved from 72 to
73 CliOnlyProbe. The server was still running and still heldserver-owner.sqlite.Change
SqlitePersistence.layerCliConfig. The server and the offlinet3 projectpath, which already hold ownership, keeplayerConfig.hasPendingMigrations, the same rule the Migrator and the V2-preview reconciliation follow).ServerOwnsDatabaseError. The error says to use thet3that matches the running server or stop it, and the database is left unchanged.serverOwnership.tsgainsholdServerOwnershipLock, the lock alone, without the runtime-state claim.acquireServerOwnershipnow shares the same lock code.isWildcardHostandformatHostForUrlmove fromstartupAccess.tsto@t3tools/shared/hostClassification. Without the move,Sqlite.tsimportingserverOwnership.tscreates an import cycle back throughEnvironmentAuth.ts. Only imports change.Scope and approval
Fixes #26696
Verification
EnvironmentAuth.layer > does not migrate a T3 home behind the server that owns it, inapps/server/src/auth/EnvironmentAuth.test.ts. It holds the ownership lock, seeds a database one migration behind, and builds the CLI storage layer. It fails on main withexpected 72 to be 71, and passes here.opens a current T3 home while the server that owns it runs, in the same file. With nothing pending, the CLI still opens the database beside an owner.apps/server:EnvironmentAuth,Sqlite,Migrations,pairandprojecttests pass, 51 of 51.tsc --noEmitis clean.vp run check:changedpasses.ServerOwnsDatabaseErrorand the ledger stays at 72. The unmodified server and a build with a different 073 both start afterwards; on main the latter refused. Control: with nothing pending,auth session issueandpairexit 0 beside the running server. transcript-26696.md, driver: repro.shnpx t3). The stand-in migration exercises the same path.Claude Opus 5.5 (Claude Code), with step agents on Claude Opus 5.5 and GPT 6.1 Sol (Codex), through T3 Code.
🤖 Generated with Claude Code