[ https://issues.apache.org/jira/browse/HIVE-17457?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16159992#comment-16159992 ]
Eugene Koifman edited comment on HIVE-17457 at 11/10/17 6:09 PM: ----------------------------------------------------------------- the files in new base_x may not even be readable so likely this needs base_x_x so that the whole dir is skipped on read Another issue to think about: if 'x' aborts when does the metadata for it get cleaned? There is nothing in (COMPLETED_)TXN_COMPONENTS about it. Add logic to write to txn_components and special handling for base_x when x is aborted to skip it? Normal compaction will never produce base_x with x opened since it never includes any open txns. was (Author: ekoifman): the files in new base_x may not even be readable so likely this needs base_x_x so that the whole dir is skipped on read > Acid Insert Overwrite when the transaction fails > ------------------------------------------------ > > Key: HIVE-17457 > URL: https://issues.apache.org/jira/browse/HIVE-17457 > Project: Hive > Issue Type: Bug > Components: Transactions > Affects Versions: 3.0.0 > Reporter: Eugene Koifman > Assignee: Eugene Koifman > > HIVE-14988 adds support for Insert Overwrite for Acid tables. > once we have direct write to target dir (i.e. no move op) - how do we handle > the case where the txn running IOW aborts? See if getAcidState() does the > right thing -- This message was sent by Atlassian JIRA (v6.4.14#64029)