[ https://issues.apache.org/jira/browse/HBASE-9559?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13770388#comment-13770388 ]
Liang Xie commented on HBASE-9559: ---------------------------------- Just found those comments before getRowKeyAtOrBefore() : {code} * Find the key that matches <i>row</i> exactly, or the one that immediately * precedes it. WARNING: Only use this method on a table where writes occur * with strictly increasing timestamps. This method assumes this pattern of * writes in order to make it reasonably performant. Also our search is * dependent on the axiom that deletes are for cells that are in the container * that follows whether a memstore snapshot or a storefile, not for the * current container: i.e. we'll see deletes before we come across cells we * are to delete. Presumption is that the memstore#kvset is processed before * memstore#snapshot and so on. {code} > getRowKeyAtOrBefore may be incorrect for some cases > --------------------------------------------------- > > Key: HBASE-9559 > URL: https://issues.apache.org/jira/browse/HBASE-9559 > Project: HBase > Issue Type: Bug > Reporter: Sergey Shelukhin > Priority: Minor > > See also HBASE-9503. Unless I'm missing something, getRowKeyAtOrBefore does > not handle cross-file deletes correctly. It also doesn't handle timestamps > between two candidates of the same row if they are in different file (latest > by ts is going to be returned). > It is only used for meta, so it might be working due to low update rate, lack > of anomalies and the fact that row values in meta are reasonably persistent, > new ones are only added in split. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira