There a few solutions out there for windows hadoop client, I haven't tried any of them so, will defer to the community at large for specifics. One elegant/interesting approach that I will point out is the possibility of a flume agent for windows.    Tried and true method: Off the top of my head, you can use file transfer method of your choice to get the files from the windows machine to a linux machine (SFTP, Samba, etc), and then use your favorite HDFS loading command/process to get the files into HDFS (hdfs dfs -copyFromLocal, flume, etc.)   ... View more

The latter. Active NN will write every edit to all available journal nodes. A majority is confirm is required before the transaction is committed - that is, 2 out of your 3 journal nodes will have to have accepted the write.   In the case of the standby NN reading from the journal nodes, it will again look for a majority and discard the discrepancy.   I think if one out of three JN are down, and the final two don't agree, it will accept the most up to date JN's answer. But I could be completely wrong on this. ... View more

HAProxy is extremely lightweight. You should put it somewhere that is highly available or use two in a HA configuration. ... View more

This is actually a logging regression issue, that was introduced in CDH5.7.0, but fixed in CDH5.7.1, and the messages are benign, but can fill up the logs quickly. The recommendation would be to upgrade to CDH5.7.1 or higher where this logging regression was fixed. Alternatively, you can suppress these INFO messages by adding the following to the"Solr Server Logging Advanced Configuration Snippet (Safety Valve) ": log4j.logger.org.apache.solr.servlet.SolrDispatchFilter=WARN -pd ... View more