The HortonWorks HDP, could be implemented in two ways:
Sandbox (VM)
Manual Installation.
I would like to understand, whether HDP SandBox, or the manual installation is preferred in the production environment. The choice could be made on obvious reasons like performance, but I would like to understand whether there are any other considerations?
The Hortonworks Sandbox allows to try out the features and functionality in Hadoop and its' ecosystem of projects. That's all.
If you want to go to production, you have three installation type:
Automated with Ambari
Manual
Cloud with Cloudbreak
Regards,
Alain
performance. hadoop is about parallel processing. Can't do that with a single node.
storage. hadoop uses a distributed file system. With a single node your storage space is very limited.
redundancy. if this node dies, everything is gone. Normal hadoop configuration include a redundancy factor (of 3 by default) so that when some nodes or disks go down, all of the data is still reachable. Similarly with a standby namenode.
There are a few other points, but these are the main ones IMO.
Single node hadoop only makes sense for proof of concept, and experimentation. Not for providing production level value.
Related
I was reading the document of Hadoop, and I found this:
"Both standalone mode and pseudo-distributed mode are provided for the purposes of small-scale testing".
I have 2 questions.
First, how big is considered as small-scale, more specifically, I'm going to use at most 32 nodes, is this ok for me to run it in the pseudo-distributed mode?
Second, even for small-scale, is there any performance difference between Pseudo-Distributed and Fully-distributed mode? Since, I'm running hadoop on my Mac, and it's kind difficult for me to find a really cluster system. Anything that I have to pay attention?
at most 32 nodes, is this ok for me to run it in the pseudo-distributed mode?
Pseudo distributed specifically means you only have one node. It means all Hadoop services are capable of talking to each other as if they were on an external interface (not all localhost) connection, and using HDFS, not just the local filesystem.
In order to create a "distributed mode" cluster, you can add additional nodes to your single node by using the correct configurations. Tip: Apache Ambari would make this process much easier.
However, HDFS will want to be able to replicate blocks at least three times by default, and in order to accommodate for downtime in these services, 5 nodes is a good minimum. I also recommend that you setup High Availability in your cluster using a standalone installation of 3-5 Zookeeper servers
I am in planning phase of a multi-node Hadoop cluster in a Docker based environment. So it should be based on a lightweight easy to use virtualized system.
Current architecture (regarding to documentation) contains 1 master and 3 slave nodes. This host machine uses HDFS filesystem and KVM for virtualization.
The whole cloud is managed by Cloudera Manager. There are several Hadoop modules installed on this cluster. There is also a NodeJS data upload service.
This time I should make architecture Docker based.
I have read several tutorials and have some opinions, but also open questions.
A. What do you think, is https://github.com/Lewuathe/docker-hadoop-cluster a good base for my project? I have found also an official image, but it is single-node.
B. How will system requirements change if I would like to make this in a single container? It would be great, because this architecture should work in different locations, so changes can be easily transferred between these locations. Synchronization between these so called clones would be important.
C. Do you have some other ideas, maybe best practices?
As of September 2016 there is no quick answer.
https://github.com/Lewuathe/docker-hadoop-cluster does not seem like a good start, as it should be universal for your B. option
Keep an eye on https://github.com/sequenceiq/hadoop-docker and https://github.com/kiwenlau/hadoop-cluster-docker
To address your question C., you may want to check out BlueData's software platform: http://www.bluedata.com/blog/2015/06/docker-containers-big-data-clusters
It's designed to run multi-node Hadoop clusters in a Docker-based environment and there is a free version available for download (you can also run it in an AWS EC2 instance).
This work has already been done for you, actually:
https://hub.docker.com/r/cloudera/clusterdock/
It includes a pre-packaged multi-node CDH cluster, with Cloudera Manager as an optional component for cluster management et al.
I like to study about Hadoop multinode setup and installation, by referring the above tutorial I understand that single node cluster environment can be used as node for the multinode cluster
http://bigdatahandler.com/hadoop-hdfs/hadoop-multi-node-cluster-setup/
Currently I am learning Hadoop using Horton sandbox, can we use a sandbox system as a single node environment?
If not what is the difference between sandbox and traditional Hadoop cluster installation
The sandbox images (from Hortonworks and Cloudera) provide the user with a pre-configured development environment with all the usual tools already available and installed (pig, hive etc.). Since the image is a single "system" it is set-up such that the hadoop cluster is single-node: i.e. everything - HDFS, Hadoop map-reduce etc. - is local to that image. That is a massive benefit, as anyone who has set up a hadoop cluster will tell you! It allows you to get up-and-running with very little operational overhead.
What these sandboxes do not provide, however, is realistic cluster behaviour as you have only one node. But there other possibilities - tools such as Vagrant and Docker - that would allow you to do this (I have not tried it myself).
The big data handler link you shared seems to be about combining several of these standalone, inherently single-node "clusters" so that you have something more realistic. But I would guess setting this up so that YARN, Zookeeper and other services are not duplicated comes with a not insignificant challenge.
I'm familiar with the infrastructure or architecture of Cloudera:
Master Nodes include NameNode, SecondaryNameNode, JobTracker, and HMaster.
Slave Nodes include DataNode, TaskTracker, and HRegionServer.
Master nodes should all be on their own nodes (unless its a small cluster, than SecondaryNameNode, JobTracker, and HMaster may be combined, and even the NameNode if its a really small cluster).
Slave Nodes should always be colocated on the same node. The more slave nodes, the merrier.
SecondaryNameNode is a misnomer, unless you enable it for High Availability.
Does MapR maintain this setup? How is it similar and how is it different?
Good information by #JamCon in his reply, but there are some things worth clarifying:
The comment regarding patches is not accurate. MapR packages a broad range of Hadoop projects in its distribution so you don't have to separately compile anything. And MapR has the same APIs as any other distro, meaning their packages are not about compatibility but are simply bug fixes / enhancements from the community. There's typically no extra work required to get Hadoop ecosystem projects to run on MapR. And they release ecosystem updates at least once a month, as far as I can tell, to keep current with new enhancements.
Regarding the inclusion of YARN, we've been running MapR on YARN across large clusters since July '14! I believe MapR has their own ecosystem project vetting process, and they graduate MapR packaged versions to GA once they determine a project is ready for enterprise support.
MapR deviates from the vanilla Hadoop & CDH distributions a bit. It keeps most of the services and structure (Job Tracker, Data Nodes, HBase Master & Region, MR, etc), but there are some significant differences.
One of the defining items about MapR's distribution is that it doesn't use HDFS. It has its own custom FS, which features HA and operates without Name Nodes (via distributed metadata). It also allowed them to enable NFS access years ahead of the rest of the Hadoop distros, as well as snap shotting.
The custom FS does complicate their distribution a bit, though ... for example, when you want to run products or services, you often need to install the MapR specific patches. When you want to run mahout, you need to compile it with the MapR patches from https://github.com/mapr/mahout. But it also gives them an opportunity to incorporate better security at the FS level, as seen by the implementation of "Access Control Expressions" and Cluster/Job/Volume ACLs.
Overall, it's a well structured product. My biggest concern is they've deviated so far from the norm that when new innovations are adopted, they're slow to adapt, because it has to be incorporated into their highly modified environment. YARN is a perfect example ... they haven't released it yet, even though their competitors have.
From an architecture stand point with MapR there are no master nodes. The functions that the master nodes provide in a typical Hadoop architecture are instead distributed and performed within the "data nodes" of MapR.
https://www.mapr.com/why-hadoop/why-mapr/architecture-matters
MapR doesn't have master node, inbuilt mechansim but in Cloudera have master node, secondary name node and resource manager
http://commandstech.com/mapr-vs-cloudera-vs-hortonworks/
I'm beginner programmer and hadoop learner.
I'm testing hadoop full distribute mode using 5 PC(has Dual-core cpu and ram 2G)
before starting maptask and hdfs, I knew that I must configure file(etc/hosts on Ip, hostname and hadoop folder/conf/masters,slaves file) so I finished configured that file
and when debating on seminar in my company, my boss and chief insisted that even if hadoop application running state, if hadoop need more node or cluster, automatically, hadoop will add more node
Is it possible? When I studied about hadoop clusturing, Many hadoop books and community site insisted that after configuration and running application, We can't add more node or cluster.
But My boss said to me that Amazon said adding node on running application is possible.
Is really true?
hadoop master users on stack overflow community, Please tell me detail about the truth.
Yes it indeed is possible.
Here is the explanation in hadoop's wiki.
Also Amazon's EMR enables one to add 100s of nodes on-the-fly in an alreadt running cluster and as soon as the machines are up they are delegated tasks(unstarted mapper and/or reducer tasks) by the master.
So, yes, it is very much possible and is in use and not just in theory.