09-19-2016 02:54 AM
when I enable HA of hdfs from cloudera manager 5.8.1 , It must select 3 or more journalnode , It's right.
but next step of Review Changes , when input JournalNode Edits Directory, if I input path "/path/to/journallog" , it alert "the other two in empty."
and there is no Add button to add other journalnode Edits Directory .
How can I ?
Thank you for any help.
09-19-2016 08:49 PM
If I understand correctly, you selected three JournalNode hosts, but the Review Changes page of the Enable High Availability wizard only presented a single text box for entering the JournalNode edits directory. Is that accurate?
You should see something like the following, with three separate text boxes for entering the edits directories:
Can you post a screenshot of what you're seeing when you get to this page of the wizard?
09-19-2016 10:19 PM
09-20-2016 08:27 AM
First, note that the paths for the JournalNode edits directories are simple absolute paths like /x/y . There should be no "node:" prefix. That is why you are getting the error about the paths not matching the regular expression.
Second, your mention of an "add button" makes me think that we might not be talking about the same thing. From your description, it sounds like you might be working through the Add Cluster wizard, as opposed to the Enable High Availability wizard.
To set up an HA cluster with Cloudera Manager, you need to execute the following steps:
1. Using the Add Cluster wizard (available from the "Add Cluster" button on the home page), create a non-HA cluster.
2. Once this cluster is set up, go to the HDFS status page.
3. Using the "Actions" pull-down menu at upper right, select "Enable High Availability."
Please post the steps you have taken so far to get to this point.
09-20-2016 10:23 PM
Thank you for your replay. I did everything like you said . And post the error page up . The problem is how can I
input a path for 3 journalnode in one textarea.
09-21-2016 09:28 AM
I notice that there is only one group showing up in the Review Changes page: "hadoop." I assume that that is not the name of a host in your cluster - is that correct?
Is there any chance that the standby NameNode and JournalNode roles already exist, or were created outside of the wizard? Or possibly left over from a previous failed run of the wizard?
Before invoking the wizard, the cluster should be in a non-HA state: that is, there should be only NameNode, and no JournalNodes.
09-21-2016 08:17 PM