Fixed Issues in Apache HDFS

Review the list of HDFS issues that are resolved in Cloudera Runtime 7.1.8.

OPSAPS-62062: From releases 7.1.8 onwards, increased the value of property dfs.client.block.write.locateFollowingBlock.retries from 7 to 10.

OPSAPS-61810: When an HDFS incremental replication fails the last successfully replicated snapshot is deleted to make the next run of the replication policy start a new full bootstrap replication. This can be now disabled with the new flag of deleteLatestSourceSnapshotOnJobFailure, by setting it to false.

CDPD-33801: Fixed bug where in rare cases Erasure Coded blocks may be removed permanently.

CDPD-27239: This commit reverts HDFS-13671, which was meant to reduce NameNode heap consumption, but it occasionally causes severe performance problems for some users, and that it does not go away without restarting the process. The revert will stabilize the NameNode performance for these users.