[ https://issues.apache.org/jira/browse/HADOOP-15485?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16482658#comment-16482658 ]
Steve Loughran commented on HADOOP-15485: ----------------------------------------- alternatively: once a stream read has succeeded once, it can't ever fail again. Note that ti's the default "failure = 1.0f" value which is trouble here > reduce/tune read failure fault injection on inconsistent client > --------------------------------------------------------------- > > Key: HADOOP-15485 > URL: https://issues.apache.org/jira/browse/HADOOP-15485 > Project: Hadoop Common > Issue Type: Sub-task > Components: fs/s3, test > Affects Versions: 3.1.0 > Reporter: Steve Loughran > Priority: Major > > If you crank up the s3guard directory inconsistency rate to stress test the > directory listings, then the read failure rate can go up high enough that > things read IO fails,. > Maybe that read injection should only happen for the first few seconds of a > stream being created, to better model delayed consistency, or least limit the > #of times it can surface in a stream. (This woluld imply some kind of > stream-specific binding....) > Otherwise: provide a way to explicitly set it, including disable -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: common-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: common-issues-h...@hadoop.apache.org