[IMPALA-12167] TestJvmMemTracker.test_jvm_mem_tracking crashing impalad during startup

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: Impala 4.3.0
• Fix Version/s: Impala 4.3.0
• Component/s: Backend
• Epic Color: ghx-label-12
• Assignee: Michael Smith
• Reporter: David Rorke
• Created: 25/May/23 22:43
• Updated: 16/Jul/24 23:56
• Resolved: 05/Jun/23 17:05

Description

test_jvm_mem_tracking is failing and causing the impalad to abort/crash early during startup.  Minidump is attached.

The test starts impala with the following arguments:

start-impala-cluster.py '--state_store_args=--statestore_update_frequency_ms=50 --statestore_priority_update_frequency_ms=50 --statestore_heartbeat_frequency_ms=50' --cluster_size=1 --num_coordinators=1 --log_dir=/data/jenkins/workspace/impala-asf-master-exhaustive/repos/Impala/logs/custom_cluster_tests --log_level=1 '--impalad_args=--mem_limit_includes_jvm=true --codegen_cache_capacity=0 ' '--state_store_args=None ' --jvm_args=-Xmx1g --impalad_args=–default_query_options=

This test is unusual in setting --mem_limit_includes_jvm=true in the startup flags, so it seems likely the failure is related to use of this flag.  There was also recent change to the Java startup options in IMPALA-11260 (to "add add-opens to JAVA_TOOL_OPTIONS on startup") that could be related.

Issue Links

• is caused by: IMPALA-11260 Catalog cache item sizes of CatalogdMetaProvider are underestimated on Java9+