0.2.1-rc.3 Mac版 工作区之前的项目与会话不见了。也无法创建新的项目与会话。

[wMango]

workspace create failed: gateway/service-unavailable: typert gateway: workspace/create: active Service "workspaceController" is unavailable

[Recursive-max]

同样的问题，更新之后就用不了了

[Recursive-max]

我是windows

[suhuacheng66-cpu]

解决了吗，我今天也遇到了这个问题

[meovv223]

将.dsh\profiles\desktop\node_modules\@deepseek-ai里面的dsh-storage-domain文件夹删除试试看。我是这么解决的，删了也没影响功能。

[suhuacheng66-cpu]

试一下把C:\Users\用户名.dsh\sessions中的对话文件全部删除，这样就可以重新用了

[suhuacheng66-cpu]

可以先备份所有对话文件，一个一个试是哪个文件出问题，这样就不会影响到其他正常的对话文件

[polaris-smart]

Root cause found (with reproduction), plus a fix PR — our machine hit the same error and we dug all the way down.

The storage-domain peer-dependency theory from #9984 did not hold on our machine (no stale copy in node_modules, and dsh-session-persistence-jsonl / dsh-workspace / dsh-session are byte-identical between rc.2 and rc.3 by MD5). The actual chain on our deployment:

1. Latent: an external Feishu bridge archives sessions because the SDK lacks resume_session. Its archiver renames session directories in place: feishu-<id>/ → feishu-<id>.archived-<ts>/, then creates a fresh <id>/ — two directories share one session-id lineage in the same project slug (left there on 10-20).
2. Dormant: the desktop app ran 6 days without restart (ledger ready event count = 0 from 10-20 to 10-25), and sessions are only scanned at startup.
3. Trigger: the rc.2→rc.3 nightly auto-update forced a relaunch (10-26 20:11) — the first full session scan hit the relocated directory.
4. Failure mechanism: JsonlSessionPersistence.assertStoredIdentity() throws a plain Error (corrupt session log "...": header id "..." and cwd identify "..."), and listArtifacts() only filters SessionFormatUnsupportedError / SessionPersistenceCorruptionError. The plain error escapes, the listing throws, dsh-workspace never activates → workspaceController unavailable → exactly the error in this thread.

Evidence:

• rc.2 vs rc.3: the three relevant packages are MD5-identical — not version-specific
• rc.2 CLI against the same session root reproduces the identical corrupt session log error
• Moving the relocated directory out of the sessions root rebuilds all 23 workspaces / 867 sessions with zero loss; session payloads are intact (824 v4 + 43 legacy, all decompress cleanly)

Fix PR: #1 in https://github.com/polaris-smart/deepseek-harness — in assertStoredIdentity, listing scans (expectedId === undefined) throw SessionPersistenceCorruptionError for cwd/identity mismatches so listArtifacts skips relocated logs like the other corruption cases; targeted loads keep the plain error. Draft (regression test + CI validation incoming).

Residual risk worth flagging: any external tool that archives by renaming directories inside sessions/ will re-create this state on every archive cycle. Either upstream adds a first-class archive API, or listArtifacts should tolerate (skip + warn) unexpected directory layouts generally.

CC: #9984 (the same error message, different proximate trigger — same fix covers it).

[polaris-smart]

Follow-up: the actual diff (12 lines) — apply-ready if maintainers want it as a PR.

Upstream has issues and PRs disabled, so leaving this here. The change makes listing scans throw the typed SessionPersistenceCorruptionError for identity/cwd mismatches (which listArtifacts already skips), while targeted loads keep the plain error:

``````diff
diff --git a/packages/session/session-persistence-jsonl/src/index.ts b/packages/session/session-persistence-jsonl/src/index.ts
index f8351fa9ee..1fe897629a 100644
--- a/packages/session/session-persistence-jsonl/src/index.ts
+++ b/packages/session/session-persistence-jsonl/src/index.ts
@@ -1608,8 +1608,17 @@ class JsonlSessionPersistence extends SessionPersistence {
     signal?: AbortSignal,
   ): Promise<void> {
     signal?.throwIfAborted()
+    // Listing scans pass no expectedId: a relocated (e.g. archived) log must be
+    // skipped there, not fail the whole workspace listing, so identity mismatches
+    // surface as the typed corruption error that listArtifacts already filters.
+    // Targeted loads (expectedId set) keep the caller-specific plain error.
+    const isListingScan = expectedId === undefined
+    const identityError = (message: string, cause?: unknown) =>
+      isListingScan
+        ? new SessionPersistenceCorruptionError(message, { cause })
+        : new Error(message, { cause })
     if (expectedId !== undefined && meta.id !== expectedId) {
-      throw new Error(`corrupt session log "${path}": requested id "${expectedId}" does not match header id "${meta.id}"`)
+      throw identityError(`corrupt session log "${path}": requested id "${expectedId}" does not match header id "${meta.id}"`)
     }
     let expectedPath: string
     try {
@@ -1621,10 +1630,10 @@ class JsonlSessionPersistence extends SessionPersistence {
         this.compression,
       )
     } catch (error) {
-      throw new Error(`corrupt session log "${path}": header id cannot name a storage path`, { cause: error })
+      throw identityError(`corrupt session log "${path}": header id cannot name a storage path`, error)
     }
     if (path !== expectedPath && !await this.sameFile(path, expectedPath, signal)) {
-      throw new Error(`corrupt session log "${path}": header id "${meta.id}" and cwd identify "${expectedPath}"`)
+      throw identityError(`corrupt session log "${path}": header id "${meta.id}" and cwd identify "${expectedPath}"`)
     }
     signal?.throwIfAborted()
   }

The branch is on our fork (polaris-smart:fix/list-survives-relocated-session-logs) — happy to open a proper PR if/when the upstream contribution flow allows it.

Relative validation on our Linux box (same env, master vs this branch): listing-relocation test failures on master no longer fail with this change (5 failed/25 passed → 2 failed/31 passed at the package level; the remaining failures are missing workspace builds in our environment, unrelated).