steveloughran commented on issue #1003: HADOOP-16384: Avoid inconsistencies between DDB and S3 URL: https://github.com/apache/hadoop/pull/1003#issuecomment-510176835 thanks for the tests. Reviewing the failures {code} testRmEmptyRootDirNonRecursive(org.apache.hadoop.fs.contract.s3a.ITestS3AContractRootDir) {code} This is precisely the tombstone problem surfacing again, somehow. if the root dir is considered entry the test rm is becomes a no-op and succeeds. {code} ITestS3AEncryptionSSEC.testRenameFile:178 ? NullPointer {code} Now, that is a regression, looks like the local test runs are doing it because there's no state. Reverted back to trunk. The DDB AncestorState now does (At debug level only) a check on the state of the result, logs if there are issues. View it as an audit of the outcome. But I'd missed that the local store used the same rename tracker without any object. Switched back to the trunk close-quietly operation, and if/when the auditing actually does real validation (or even better: goes down tagging everything in the dest dir are auth) then it can be revisited. {code} ITestPartitionCommitMRJob {code} I've filed https://issues.apache.org/jira/browse/HADOOP-16407 on this...it looks like the to move to sequential runs and FS recycling is causing this. I've not been worrying about it precisely because it's clearly a test setup problem that I believe can be done in isolation. But I can take another look...
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
