Which is more better way in aspect of control, optimisations and easy to change. Installation is not an issue we have done that thing before also last time in a mobile analytics system we have used manual install of hortonworks cause that time there were lot of bugs in Amabri ..
Now this time we are creating a advertising product and installing new Hadoop cluster. So exploring Ambari also. so need some feedbacks
- is it stable now ?
- is this impact cluster in any way or create any overhead ?
- If we say we have technical expertise on Hadoop than what you will suggest Ambari or manual ?
I have listen from some peers you lose control over configurations for more customise tweaks, is it right ?
Depends there are some really big clusters out there managed by Ambari. There are a couple things that can make problems and require workarounds which might be easier to manage when done manually. So I suppose its a tradeoff. What size are we talking about? And secondly what were the big issues you expect or encountered that made it easier to use the manual approach? We could tell you if that is still needed with Ambari.
Some nice things that were added recently:
Express upgrade: which solves the problem of upgrading a very large cluster that could have taken days in the rolling upgrade
Support for kerberos in flume/kafka will come soon.
Some other things might require some workarounds. For example big clusters often use 5 journalnodes instead of 3 which is possible using the curl api but not directly from the UI. One other thing I have seen was the ability to hotswap a third namenode in for a failed one. Also something that needs some workarounds but also works with some curl commands.
I would suggest coming with a list of concerns and more details of the cluster size to our support and then we could give you more detailled information.
Perhaps some other people can provide some experience from a non-HWX side :-)
@Shashank Agarwal We have used Ambari to install the HDP 2.4 on a small 8 node cluster without any issues. We've repeated this process more than 10 times and it just seems to work. I'd strongly recommend paying a lot of attention to the node prep process.
We're using RHEL 7 and HDP 2.4.0, and going to 2.4.2 in the next couple of weeks. The nodes are all based in the AWS cloud.
Hope this helps.