[ https://issues.apache.org/jira/browse/KUDU-1936?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15922986#comment-15922986 ]
David Alves commented on KUDU-1936: ----------------------------------- Looks to me that LogReader::ReadReplicatesInRange might be the culprit. It uses a 32-bit int that gets set to 'starting_at' which is a 64 bit int. > Potential 2B max log index size > ------------------------------- > > Key: KUDU-1936 > URL: https://issues.apache.org/jira/browse/KUDU-1936 > Project: Kudu > Issue Type: Bug > Components: consensus > Affects Versions: 1.3.0 > Reporter: Mike Percy > > We have a report of a potential truncation of a 64 bit integer to 32 bits on > a log index. This effectively means we can only update a tablet 2B times > before crashing on an integer overflow. > I'm currently investigating and working on reproducing this. > Log message looks like this: > {noformat} > F0313 log_index.cc:171 Check failed: log_index > 0 (-2147483648 vs. 0) > {noformat} -- This message was sent by Atlassian JIRA (v6.3.15#6346)