[Bug] RaftEngineReadIndexTest fails now and then with UnfinishedStubbingException: the alive-peers refresher can call the mocked raft node while a test stubs it #3384

Bug Type (问题类型)

others (please edit later)

Before submit

• I have confirmed and searched that there are no similar problems in the historical issue and documents

Environment (环境信息)

• Server Version: PD tests, hg-pd-test pd-core-test profile (PDCoreSuiteTest). Seen on the branch of feat(pd): commit schema keys and a per-graph record as one raft entry #3376 at 171760d25, which does not change RaftEngineReadIndexTest or the alive-peers refresher of RaftEngine; both are as on master 86c05ef9c (the test from refactor(pd): make PD reads and elections safe #3368, the refresher from feat(pd): add quorum-aware /v1/ready endpoint and raft gauges #3185)
• Backend: none (unit suite)
• OS: Linux, 8 cores, JDK 17
• Data Size: N/A

Expected & Actual behavior (期望与实际表现)

RaftEngineReadIndexTest.testStoppedReadServiceClosureThatRunsFirstFailsAsNotLeader should pass every run. It failed once in three runs of the same commit on 2026-10-20, with Mockito's unfinished-stubbing error while the test stubs readIndex:

[ERROR] org.apache.hugegraph.pd.raft.RaftEngineReadIndexTest.testStoppedReadServiceClosureThatRunsFirstFailsAsNotLeader -- Time elapsed: 0.012 s <<< ERROR!
org.mockito.exceptions.misusing.UnfinishedStubbingException:
Unfinished stubbing detected here:
-> at org.apache.hugegraph.pd.raft.RaftEngineReadIndexTest.answerShutdown(RaftEngineReadIndexTest.java:377)
	at org.apache.hugegraph.pd.raft.RaftEngineReadIndexTest.answerShutdown(RaftEngineReadIndexTest.java:382)
	at org.apache.hugegraph.pd.raft.RaftEngineReadIndexTest.testStoppedReadServiceClosureThatRunsFirstFailsAsNotLeader(RaftEngineReadIndexTest.java:367)

The runs, all mvn -B -ntp test -pl hugegraph-pd/hg-pd-test -am -P pd-core-test on 171760d25:

Likely cause, not proven: the test swaps a Mockito mock into the RaftEngine singleton (Whitebox.setInternalState(..., "raftNode", mockNode)) and stubs it, while the engine's pd-raft-alive-peers refresher thread (startAlivePeersRefresher, every ALIVE_PEERS_REFRESH_MS) calls node.listAlivePeers() on whatever node the singleton holds whenever its state machine reports leader. An earlier suite test can leave the singleton started and leading. A call on the mock from that thread while doAnswer(...).when(mockNode).readIndex(...) is in progress is the case Mockito reports as unfinished stubbing.

A fix could stop the refresher for these tests (or give the refresher a node snapshot taken under the same lock the tests use), or stub listAlivePeers on the mock before it is installed.

Vertex/Edge example (问题点 / 边数据举例)

N/A

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

N/A