[DRILL-3192] TestDrillbitResilience#cancelWhenQueryIdArrives hangs

• Type: Bug
• Status: In Progress
• Priority: Critical
• Resolution: Unresolved
• Fix Version/s: Future
• Component/s: Execution - RPC
• Assignee: Sudheesh Katkam
• Reporter: Sudheesh Katkam
• Created: 27/May/15 00:28
• Updated: 24/Jun/16 21:50

Description

TestDrillbitResilience#cancelWhenQueryIdArrives (previously named cancelBeforeAnyResultsArrive) hangs when the test is run multiple times.

(Will add more information)
Configuration: BIT_SERVER_RPC_THREADS = 1

The remote RPC thread with a cancel signal is waiting for the fragment to start accepting external events. The fragment is waiting for a ControlConnection to the Foreman node (through ReconnectingConnection). The Foreman node is waiting for the remote node to accept the connection, which happens through an RPC thread. Distributed deadlock.

DRILL-3242 should solve this problem.

Issue Links

• depends upon: DRILL-3242 Enhance RPC layer to offload all request work onto a separate thread.