[FLINK-27954] JobVertexFlameGraphHandler does not work on standby Dispatcher

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.15.0
• Fix Version/s: 2.0.0
• Component/s: Runtime / Coordination, Runtime / REST
• Labels: pull-request-available stale-assigned
• Assignee: Weijie Guo
• Reporter: Chesnay Schepler
• Created: 08/Jun/22 10:55
• Updated: 10/Jul/24 06:57

Description

The JobVertexFlameGraphHandler relies internally on the JobVertexThreadInfoTracker which calls
ResourceManagerGateway#requestTaskExecutorThreadInfoGateway to get a gateway for requesting the thread info from the task executors. Since this gateway is not serializable it would categorically fail if called from a standby dispatcher.
Instead this should follow the logic of the MetricFetcherImpl, which requests addresses instead and manually connects to the task executors.

Issue Links

• relates to: FLINK-27933 Savepoint status cannot be queried from standby jobmanager
• links to: GitHub Pull Request #22042