[Question] The Hugegraph May have stability issues of 1.5.0

[hexuxu110]

Problem Type (问题类型)

None

Before submit

• 我已经确认现有的 Issues 与 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: 1.5.0 (Apache Release Version)
• Backend:Hstore x nodes, HDD or SSD
• OS: xx CPUs, xx G RAM, Ubuntu 2x.x / CentOS 7.x
• Data Size: xx vertices, xx edges

Your Question (问题描述)

部署4PD 4Server 4HStore

直接下线某个Hstore，整个集群不可用

Vertex/Edge example (问题点 / 边数据举例)

Schema [VertexLabel, EdgeLabel, IndexLabel] (元数据结构)

[dosubot]

This is a known and reproducible issue in HugeGraph 1.5.0 distributed deployments with HStore backend: if you take an HStore node offline and the number of available Store nodes drops below the configured minimum (set by initial-store-count in PD), the entire cluster becomes unavailable. This is expected behavior based on the Raft consensus and fault tolerance design—enough nodes must be online to maintain quorum and data consistency. If your initial-store-count is set to 4, losing any one HStore node will make the cluster unavailable; lowering this value (e.g., to 3) allows one node to go down without full outage, but reduces fault tolerance. See discussion and config details.

Cluster instability and data out-of-sync are often caused by configuration inconsistencies (like mismatched PD/Store node lists, ports, or IPs), partition allocation not being triggered, or legacy parameters (such as server.role) remaining in configs. Make sure all PD and Store nodes have consistent configs, remove deprecated parameters, and check that partition allocation is triggered after all nodes register and leader election completes. Manual partition allocation can be done via the PD REST API /partition/dispatch if needed—see partition allocation troubleshooting and known cluster state issues.

There is no official documentation for distributed HStore cluster fault tolerance in 1.5.0; most troubleshooting relies on community issues. This scenario is under investigation by maintainers and is a known problem in the community.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[hexuxu110]

curl --compressed --location --request GET 'http://127.0.0.1:8081/graphs/g_rocksdb/graph/vertices'
{"exception":"class org.apache.hugegraph.HugeException","message":"Failed to serialize vertices","cause":"org.apache.hugegraph.store.client.type.HgStoreClientException: io.grpc.StatusRuntimeException: UNAVAILABLE: io exception"}

[hexuxu110]

@dosu pd:

Storage path

data-path: ./pd_data

The check cycle of automatic expansion regularly checks the number of partitions in each store and automatically balances the number of partitions

patrol-interval: 1800

The minimum number of surviving store nodes, less than which the entire cluster is unavailable

initial-store-count: 1

[dosubot]

你的 PD 配置里 initial-store-count: 1 理论上意味着只要有 1 个 Store 节点存活，集群就应该保持可用，但实际出现 UNAVAILABLE: io exception 这类错误，通常不是这个参数本身导致，而是以下常见原因：

1. PD/Store 配置不一致（如 initial-store-list、raft peers-list、端口、IP 必须完全一致），或者有节点网络不通，导致 raft 共识异常、分区分配失败，最终集群不可用。参考
2. HugeGraph 1.5.0 不会自动分配分区，所有 PD/Store 节点注册、raft 选主完成后，必须手动通过 PD 的 REST API /partition/dispatch 触发分区分配，否则数据查询会报错。参考
3. 某些遗留参数（如 server.role）在 HStore+PD 场景下已废弃，保留会导致调度异常，建议彻底移除。参考
4. 认证开启时，HStore 不会自动初始化 ~user 等内置 schema，需手动补齐，否则初始化和后续操作会异常。参考
5. gRPC “UNAVAILABLE” 还可能是 Store 节点进程挂掉、网络端口被防火墙阻断、PD/Store 版本不一致、分区表丢失等运行时问题。参考

排查建议：

• 检查所有 PD/Store 配置文件，确保 initial-store-list、raft peers-list、端口、IP 完全一致，且所有节点网络互通。
• 查看 PD 和 Store 日志，确认没有 raft 选主、注册超时、分区分配失败等报错。
• 用 PD 的 REST API /partition/dispatch 手动分配分区，确保分区状态正常。
• 移除所有 server.role、server.id 等遗留参数。
• 如果开启认证，先手动补齐 ~user schema，再初始化 store。
• 确认所有节点进程都正常运行，端口未被防火墙阻断。

只设置 initial-store-count: 1 并不能规避上述问题，集群仍可能因配置、网络、分区等原因不可用。遇到 gRPC “UNAVAILABLE” 错误，优先排查上述配置和分区分配问题。详细排查案例

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[hexuxu110]

@imbajin 他说错了吧 第二点HugeGraph 1.5.0 不会自动分配分区，所有 PD/Store 节点注册、raft 选主完成后，必须手动通过 PD 的 REST API /partition/dispatch 触发分区分配，否则数据查询会报错