Details
-
Bug
-
Status: Resolved
-
Critical
-
Resolution: Fixed
-
Impala 4.0.0, Impala 4.1.0, Impala 4.2.0, Impala 4.1.1, Impala 4.1.2
-
None
-
ghx-label-3
Description
When running in LocalCatalog mode, INVALIDATE METADATA under unloaded db will hit the following DCHECK:
F0423 17:12:50.566385 7565 catalog-op-executor.cc:88] 9c4c9d5f5d7974bc:b97e6dae00000000] Check failed: !obj.db.__isset.metastore_db Minimal database TCatalogObject should have empty metastore_db
*** Check failure stack trace: ***
@ 0x37ce3ed google::LogMessage::Fail()
@ 0x37d0324 google::LogMessage::SendToLog()
@ 0x37cddcc google::LogMessage::Flush()
@ 0x37d0849 google::LogMessageFatal::~LogMessageFatal()
@ 0x195389d VerifyMinimalResponse()
@ 0x195809c impala::CatalogOpExecutor::Exec()
@ 0x16ce391 impala::ClientRequestState::ExecDdlRequestImpl()
@ 0x16d713c boost::detail::function::void_function_obj_invoker0<>::invoke()
@ 0x18cdb52 impala::Thread::SuperviseThread()
@ 0x18ce95b boost::detail::thread_data<>::run()
@ 0x23cd927 thread_proxy
@ 0x7f196963f6db start_thread
@ 0x7f19663b761f clone
This is a bug introduced by IMPALA-9936 in which we added the DCHECK. This use case is still sending back the full DB object to the coordinator.
For reproducing the issue:
# Start Impala cluster with --hms_event_polling_interval_s=0 bin/start-impala-cluster.py --catalogd_args="--catalog_topic_mode=minimal --hms_event_polling_interval_s=0" --impalad_args=--use_local_catalog # Create new db and new table in Hive beeline -u "jdbc:hive2://localhost:11050" > create database mydb; > create table mydb.mytbl (i int); # Run IM in Impala impala-shell.sh > invalidate metadata mydb.mytbl; # The coordinator will crash at hitting the DCHECK
Attachments
Issue Links
- is caused by
-
IMPALA-9936 Only send invalidations in DDL responses to LocalCatalog coordinators
- Resolved