GML-2078 Release 1.3.1

[chengbiao-jin]

User description

Bug fixes.

---

PR Type

Bug fix, Enhancement

---

Description

• Harden TigerGraph query and ingest flows

  • Reinstall community queries when needed
  • Verify queries and loading jobs post-run
  • Auto-recreate missing loading jobs
  • Parse load stats for accurate counts

• Fix chat and connection stability

  • Handle early WebSocket disconnects safely
  • Check graph existence via getVertexTypes
  • Use listGraphs() over deprecated LS USER
  • Prevent token revoke JSON destructor errors

• Reduce graph payloads and duplication

  • Fetch only description for entities
  • Fuzzy-deduplicate near-duplicate descriptions
  • Limit stream_community output to IDs
  • Remove deprecated query metadata usage

• Expand file processing support

  • Add .png and .gif extraction support
  • Clarify JSONL copy versus conversion logs

---

Diagram Walkthrough

flowchart LR
  A["GraphRAG pipeline"] -->|"installs"| B["Required queries"]
  A -->|"reinstalls before community step"| C["Community queries"]
  A -->|"verifies after run"| D["Queries and loading jobs"]
  E["Server ingestion"] -->|"ensures exists"| F["Loading job recreation"]
  E -->|"parses"| G["Load statistics"]
  H["Chat websocket"] -->|"handles"| I["Early disconnects"]
  J["Entity lookup"] -->|"fetches only"| K["Description attribute"]
  K -->|"deduplicates with"| L["Fuzzy matching"]

Loading

File Walkthrough

	Relevant files
Bug fix	8 files tg_proxy.py Fix token revocation payload and destructor safety              +14/-10  util.py Scope required queries and optimize vertex checks                +27/-23  workers.py Harden query install checks and deduplicate descriptions  +24/-7    util.py Improve install error detection and vertex fetching            +6/-2      workers.py Add fuzzy description deduplication for entities                  +25/-8    agent_graph.py Detect SupportAI via schema and remove metadata calls        +5/-15    ui.py Handle websocket disconnects and replace deprecated graph listing +17/-13  supportai.py Recreate loading jobs and parse ingestion statistics          +62/-11
Enhancement	3 files text_extractors.py Add image formats and clearer preparation logging                +3/-1      graph_rag.py Reinstall community queries and verify pipeline artifacts +41/-0    stream_community.gsql Reduce community query output to vertex IDs                            +1/-2
Error handling	1 files main.py Validate graph existence before starting ECC tasks              +7/-1
Documentation	1 files CHANGELOG.md Document release 1.3.1 fixes and improvements                        +15/-0
Configuration changes	1 files VERSION Bump application version to 1.3.1                                                +1/-1
Dependencies	1 files requirements.txt Upgrade `pyTigerGraph` dependency to supported version      +1/-1

---

[tg-pr-agent]

PR Reviewer Guide 🔍

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 4 🔵🔵🔵🔵⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Empty response The new description merge path assumes the vertex lookup always returns at least one record with a description attribute. If the REST call returns an empty list for a missing vertex, or a different payload shape, indexing into the first element will raise and break extraction instead of treating the vertex as absent. async def get_vert_desc(conn, v_id, node: Node): new_desc = node.properties.get("description", "") exists = await util.check_vertex_exists(conn, v_id) if not exists.get("error", False): existing_descs = exists["resp"][0]["attributes"]["description"] if not new_desc or _is_near_duplicate(new_desc, existing_descs): return existing_descs if existing_descs else [new_desc] return existing_descs + [new_desc] return [new_desc] Empty response Similar entity-description deduplication logic assumes the vertex lookup response contains a first result with attributes.description. This should be validated against not-found and schema-variant responses, otherwise extraction can fail on legitimate misses. async def get_vert_desc(conn, v_id, node: Node): new_desc = node.properties.get("description", "") exists = await util.check_vertex_exists(conn, v_id) if not exists.get("error", False): existing_descs = exists["resp"][0]["attributes"]["description"] if not new_desc or _is_near_duplicate(new_desc, existing_descs): return existing_descs if existing_descs else [new_desc] return existing_descs + [new_desc] return [new_desc] Recreate parsing Loading job recreation checks and post-install validation rely on substring matching against ls output. That output format is version-sensitive, so a formatting change could incorrectly report jobs as missing or recreate them unnecessarily during ingestion. def _ensure_loading_jobs(conn: TigerGraphConnection, graphname: str, load_job_id: str) -> None: """Check that the required loading job exists; recreate it if missing.""" current_schema = conn.gsql(f"USE GRAPH {graphname}\n ls") marker = f"- CREATE LOADING JOB {load_job_id} {{" if marker in current_schema: return gsql_file = _LOADING_JOB_GSQL_FILES.get(load_job_id) if not gsql_file: raise Exception(f"Loading job '{load_job_id}' not found and no GSQL template available to recreate it") logger.info(f"Loading job '{load_job_id}' missing — recreating from {gsql_file}") with open(gsql_file, "r") as f: q_body = f.read() result = conn.gsql(f"USE GRAPH {graphname}\nBEGIN\n{q_body}\nEND\n") logger.info(f"Loading job creation result: {result}") if isinstance(result, str) and ("error" in result.lower() or "failed" in result.lower()): raise Exception(f"Failed to recreate loading job '{load_job_id}': {result}")