[CI] TransportClusterStateActionDisruptionIT class failing

[elasticsearchmachine]

Build Scans:

• elasticsearch-periodic-platform-support #7907 / ubuntu-2404-aarch64_checkpart1_platform-support-arm
• elasticsearch-periodic #7880 / openjdk21_entitlements_true_checkpart1_java-matrix
• elasticsearch-periodic-platform-support #7901 / oraclelinux-8_platform-support-unix
• elasticsearch-periodic-platform-support #7889 / oraclelinux-9_platform-support-unix
• elasticsearch-periodic #7856 / openjdk21_checkpart1_java-fips-matrix
• elasticsearch-periodic #7856 / openjdk23_entitlements_true_checkpart1_java-matrix
• elasticsearch-periodic-platform-support #7877 / opensuse-leap-15_platform-support-unix
• elasticsearch-periodic-platform-support #7871 / sles-15_platform-support-unix
• elasticsearch-periodic-platform-support #7871 / rocky-8_platform-support-unix
• elasticsearch-pull-request #69239 / part-1

Reproduction Line:

undefined

Applicable branches:
main

Reproduces locally?:
N/A

Failure History:
See dashboard

Failure Message:

undefined

Issue Reasons:

• [main] 10 failures in class org.elasticsearch.action.admin.cluster.state.TransportClusterStateActionDisruptionIT (2.4% fail rate in 414 executions)
• [main] 5 failures in pipeline elasticsearch-periodic-platform-support (45.5% fail rate in 11 executions)
• [main] 2 failures in pipeline elasticsearch-periodic (20.0% fail rate in 10 executions)

Note:
This issue was created using new test triage automation. Please report issues or feedback to es-delivery.

[elasticsearchmachine]

Pinging @elastic/es-distributed-indexing (Team:Distributed Indexing)

[elasticsearchmachine]

Pinging @elastic/es-distributed-coordination (Team:Distributed Coordination)

[JeremyDahlgren]

I was able to reproduce the null pointer exception locally if in InternalTestCluster.getMasterName() I added a sleep in between the awaitClusterState() and the client call to get the cluster state before getting the master node name.
The getMasterName() method was changed recently in #127213, where David mentions the danger of waiting for the state and then having a second call to get the state. Niels and David agreed that a test shouldn't use getMasterName() if the test expects cluster instability. The fix I'm testing ensures the master node is not null in the assertBusy() in TransportClusterStateActionDisruptionIT.runRepeatedlyWhileChangingMaster().