Personally I'm struggling with the exactly the same issue. There was already some kind of discussion on this topic here, earlier:
https://lists.apache.org/thread/wj898zq8q348721xf460mttqlty4v3zw Personally I have never ran a production cluster without CM before. Bigtop 3 with Ambari seemed ideal, but as there was no ambari-mpack for bigtop 3 yet, I put some energy in porting the HDP mpack to support bigtop. But I can understand that maintaining such a component under the Bigtop project is a no-go, because of Ambari's attic state, Python 2.7 and the (un)maintainability of such an mpack, so I stopped working on that. >From what I also understand from the above thread, 李帅 is already working on some light weight ambari alternative. It feels like that would be a good way forward but I don't know how much work has to be done to make this work and if it's still viable? The alternative would be running vanilla hadoop/bigtop. I don't have any experience with it but I guess the most important gaps for me will be: - initial installation (with support for Kerberos and SSL) - (rolling) service restarts - basic service monitoring (is a service running or down?) Maybe we just have to put some energy in creating puppet/Ansible scripts for this purpose and just forget about Ambari. Op wo 18 mei 2022 20:07 schreef Battula, Brahma Reddy <[email protected]>: > Supposed to ask same question @Martin Blom here. > IMO, still Ambari will be good choice for cluster management. I think, > most/some amount of people from bigtop(who are using for building the > packages) still use ambari. > > Planning to take collective opinion on bring back even with some other > name if not with same name. > > Any thoughts on this..? > > > On 18/05/22, 6:28 PM, "李帅" <[email protected]> wrote: > > Maintaining Ambari is not easy for its complex architecture. There are > many > Configuration Management Tools such as Puppet, Ansible for > alternatives. So > for me, I will try to use Puppet or Ansible to deploy and monitor > cluster. > Using Puppet and Ansible without Ambari-like web ui will be a gap for > common users. > > > > Martin Blom <[email protected]> 于 2022年5月18日周三 17:04写道： > > > > > Hi all. I'm new here. > > > > So at work we provide a service that's been running flawlessly on > top of > > HDP since at least 2014. The recent "events" (which we became aware > of only > > late last year, when it was already too late) had us panicking a bit > at > > first, since we realised we could no longer manage our cluster, but > we've > > since then been able to track down repo mirrors so the immediate > crisis is > > over (no thanks to Cloudera). > > > > However, with HDP dead and CentOS 7 nearing EOL, it's clear that we > need > > to move on. While we have evaluated both CDP, GCP Dataproc and even > > considered migrating from HBase to Bigtable, Bigtop seems like the > only > > sensible way forward. We just want a small database/MQ backend > cluster that > > keeps running forever to power our service. > > > > Which brings us to Ambari. I really liked Ambari, for cluster > management > > obviously but also so for monitoring, and it has worked great for us > all > > those years. A lot of work has obviously been put into Ambari and it > seems > > like such a waste to throw it all away. > > > > What are this Bigtop's plans for Ambari, the upstream project being > > retired and all? > > > > For us, we would like to bring us to the point were, in 2024 when > CentOS 7 > > goes EOL, we can bring up a new cluster on Rocky 8 with HDFS, Yarn, > HBase, > > Kafka and Zookeeper using Ambari to manage, monitor and keep it up > and > > running like that until at least 2029 when RL8 retires. It doesn't > seem > > *that* difficult given the HDP 3.1 vs Bigtop 3.0.1 BOMs for the > components > > we need. We would also be willing to put some time into making that > > happening. > > > > What are your thoughts on that? Is anybody here still interested in > > Ambari? Has anybody actually been using it to manage Bigtop > components? > > > > PS. Also cross-posting to user. Anybody else in my situation? What > are you > > all using to monitor your clusters once deployed? > > > > -- > > Martin Blom > > [email protected] > > > > > >
