Keep a Linear sync going when one issue or document cannot be read

An issue deleted after it was listed, or comments the token cannot read,
now lose that detail with a warning instead of failing the whole sync; an
unreadable document is skipped.

docsgpt/parser/remote/linear_loader.py

@@ -19,7 +19,7 @@ import re
 from typing import Any, Optional
 
 from docsgpt.connectors import linear
-from docsgpt.connectors.mcp import run_connection_session
+from docsgpt.connectors.mcp import MCPToolError, run_connection_session
 from docsgpt.parser.remote.base import BaseRemote
 from docsgpt.parser.schema.base import Document
 
@@ -90,9 +90,19 @@ class LinearLoader(BaseRemote):
                     issues.setdefault(key, issue)
         documents = []
         for key, issue in issues.items():
+            # One issue Linear will not read (deleted since it was listed, or
+            # out of the token's reach) loses that detail, not the whole sync.
             if linear.is_truncated(issue.get("description")) or issue.get("descriptionTruncated"):
-                issue = await self._full_issue(session, key, issue)
-            comments = await self._comments(session, key) if selection["include_comments"] else []
+                try:
+                    issue = await self._full_issue(session, key, issue)
+                except MCPToolError as exc:
+                    logger.warning("Linear sync keeps %s as listed: %s", key, exc)
+            comments = []
+            if selection["include_comments"]:
+                try:
+                    comments = await self._comments(session, key)
+                except MCPToolError as exc:
+                    logger.warning("Linear sync leaves out the comments of %s: %s", key, exc)
             documents.append(self._issue_document(issue, comments, selection))
         if selection["include_documents"]:
             documents.extend(await self._project_documents(session, selection["projects"]))
@@ -179,11 +189,16 @@ class LinearLoader(BaseRemote):
             ):
                 content = record.get("content")
                 if (not content or linear.is_truncated(content)) and get_schema is not None and record.get("id"):
-                    full = linear.unwrap(
-                        await session.call("get_document", {linear.argument(get_schema, "id", "documentId"):
-                                                            str(record["id"])}),
-                        "document",
-                    )
+                    try:
+                        full = linear.unwrap(
+                            await session.call("get_document", {linear.argument(get_schema, "id", "documentId"):
+                                                                str(record["id"])}),
+                            "document",
+                        )
+                    except MCPToolError as exc:
+                        # A document that cannot be read is skipped, not the sync.
+                        logger.warning("Linear sync skips document %s: %s", record.get("id"), exc)
+                        continue
                     record = {**record, **{k: v for k, v in full.items() if v not in (None, "")}}
                 document = self._linear_document(record, project)
                 path = document.extra_info["file_path"]

tests/parser/remote/test_linear_loader.py

@@ -288,3 +288,40 @@ class TestLoadData:
     def test_needs_a_connection(self):
         with pytest.raises(ValueError):
             LinearLoader().load_data({"teams": ["t1"]})
+
+
+class TestOneBadItemDoesNotStopTheSync:
+    """An issue deleted since it was listed, or comments the token cannot
+    read, lose that detail; the rest of the sync still lands."""
+
+    @staticmethod
+    def _refuse(args):
+        from docsgpt.connectors.mcp import MCPToolError
+
+        raise MCPToolError("Entity not found")
+
+    def test_an_issue_that_cannot_be_read_in_full_keeps_what_was_listed(self):
+        clipped = {**ISSUE, "description": "The spinner (truncated, use get_issue to read the full description)"}
+        session = FakeLinear({
+            "list_issues": lambda args: {"issues": [clipped, {**ISSUE, "id": "ENG-2", "title": "Other"}]},
+            "get_issue": self._refuse,
+        })
+        docs = _collect(session, teams=["team-1"], include_comments=False)
+        assert [d.text.splitlines()[0] for d in docs] == ["# ENG-1310: Fix the loader", "# ENG-2: Other"]
+
+    def test_comments_that_cannot_be_read_are_left_out(self):
+        session = FakeLinear({"list_issues": lambda args: [ISSUE], "list_comments": self._refuse})
+        [doc] = _collect(session, teams=["team-1"])
+        assert "## Comments" not in doc.text
+
+    def test_a_document_that_cannot_be_read_is_skipped(self):
+        session = FakeLinear({
+            "list_issues": lambda args: {"issues": []},
+            "list_documents": lambda args: {"documents": [
+                {"id": "d1", "title": "Gone"},
+                {"id": "d2", "title": "Plan", "content": "We ship on Monday."},
+            ]},
+            "get_document": self._refuse,
+        })
+        docs = _collect(session, projects=[{"id": "project-1", "name": "Acme"}], include_documents=True)
+        assert [d.text.splitlines()[0] for d in docs] == ["# Plan"]