Thank you for your response. Based on HIVE-28366, it appears that data correctness can be ensured through HMS Lock when 'Insert' and 'IOW' operations are executed concurrently. If both operations are performed by Hive, my understanding is that the problem is resolved regardless of the 'hive.txn.manager' setting.
However, I encountered persistent issues in a case where an IOW operation is executed by Hive, and the Insert operation is run by a different engine, specifically, Trino. To resolve this issue, it seems necessary to use 'DbTxnManager'. Could you kindly share your thoughts on this matter? Am I approaching this problem correctly, or is there another solution you recommend? Thank you! 2025년 3월 26일 (수) 오후 2:42, Butao Zhang <zhangbu...@apache.org>님이 작성: > Hi, IMO, 'org.apache.hadoop.hive.ql.lockmgr.DbTxnManager' is used for hive > native acid table, not for Iceberg table. > The lock mechanism in Iceberg table is controlled by HMS Catalog or other > Iceberg Catalog(JDBC catalog, etc). So > 'org.apache.hadoop.hive.ql.lockmgr.DbTxnManager' should not be used in your > case(insert overwrite iceberg table). > > Thanks, > Butao Zhang > > On 2025/03/26 04:54:33 박근형 wrote: > > I am experiencing a possible deadlock issue while working with Hive > version > > 4.1.0. The issue occurs when an 'insert overwrite' query is executed on > an > > Iceberg table. Specifically, I have found that if the 'hive.txn.manager' > > setting is 'org.apache.hadoop.hive.ql.lockmgr.DbTxnManager', a deadlock > > seems to happen. > > > > I would like to know whether this is a bug or a limitation. Is it that > > 'org.apache.hadoop.hive.ql.lockmgr.DbTxnManager' should not be used in > this > > case? I greatly appreciate any guidance or information on this issue. > > >
