SebastianGruza opened a new issue, #3219: URL: https://github.com/apache/hugegraph/issues/3219
### Bug Type (问题类型) data / logic error ### Before submit - [X] 我已经确认现有的 [Issues](https://github.com/apache/hugegraph/issues) 与 [FAQ](https://hugegraph.apache.org/docs/guides/faq/) 中没有相同 / 重复问题 (I have confirmed and searched that there are no similar problems in the historical issue and documents) ### Environment (环境信息) - Server Version: master `4f5202b8` (PD and store on master as well; the same happens with PD and stores left on 1.7.0) - Backend: HStore, 1 PD + 3 stores, `usePD=true` - Data written by the official `apache-hugegraph-incubating-1.7.0` release ### Expected & Actual behavior (期望与实际表现) After swapping only the jars (`lib/`) to master on a 1.7.0 installation (config and data directories untouched) the server lists the graph and sees the raw data, but not the schema: ``` GET /graphs/hugegraph/schema/propertykeys -> {"propertykeys": []} g.V().count() -> 29 g.V().hasLabel('person').count() -> Undefined vertex label: 'person' ``` On the same data 1.7.0 returns 13 property keys and 20 `person` vertices. With `usePD=false` the same swap works (HStore and RocksDB, every schema and data read identical to 1.7.0). ### Root cause The meta keys in PD are prefixed with `HUGEGRAPH/<cluster>/…`, and `MetaManager.connect()` binds the cluster name once per process: a later call with a different name is silently ignored. Two places supply the name: `GraphManager` from the `cluster` option of `rest-server.properties` (default `hg-test`), and `StandardHugeGraph` with the literal `"hg"`. Whoever connects first wins: - **1.7.0**: `HugeRestServer.start()` runs first; the `GraphManager` constructor with `usePD=true` calls `loadMetaFromPD()` → `initMetaManager()` → `connect("hg-test")`. The schema is written under `HUGEGRAPH/hg-test/GRAPHSPACE/DEFAULT/hugegraph/SCHEMA/…`. - **master**: #3008 moved `HugeGremlinServer.prepare()` in front of `HugeRestServer.start()`. `prepare()` puts every file from `conf/graphs` into `settings.graphs`, TinkerPop opens the `StandardHugeGraph`, and it calls `connect("hg")` first. The configured `cluster` is ignored from then on; the server reads `HUGEGRAPH/hg/…`, which only holds the system schema it created itself at startup. Key listing through `KvClient.scanPrefix("HUGEGRAPH/")` after the swap: 106 keys under `hg-test/…/hugegraph` (written by 1.7.0), 36 under `hg/…/hugegraph` (created by master at startup). Removing `hugegraph.properties` from `conf/graphs` does not help (the graph is not registered in the meta, the server then has no graph at all), and setting `cluster=hg-test` does not either (the literal wins regardless of the option). ### How to reproduce 1. 1.7.0, HStore, `usePD=true` and `pd.peers` in `rest-server.properties`; create a schema and some data. 2. Stop, replace the server `lib/` with master (PD and stores may stay on 1.7.0 or move to master, same result), start without `init-store`. 3. `GET /schema/propertykeys` → empty list, `g.V().hasLabel(...)` → Undefined vertex label. Scripts and logs from the lab run: https://github.com/SebastianGruza/hugegraph-validation/tree/master/results/upgrade-170-to-master. A fix is in the linked PR. ### Vertex/Edge example (问题点 / 边数据举例) _No response_ ### Schema [VertexLabel, EdgeLabel, IndexLabel] (元数据结构) _No response_ -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
