Re: Cluster Management: OpenSource Vendor Options
I can¹t comment on the direction of Ambari, nor CM¹s internals, but a good set of Hadoop command-line management tools, based on puppet, chef or salt, would certainly be a good addition to the community, regardless of BigTop supporting Ambari. Especially since it allows easy integration into existing data centre administration operations processes. If sufficiently mature, a thin GUI shouldn¹t be that hard to add for common operations. The question is: where does this fit? In BigTop as a sub-project? A project in and of itself? It¹s not a trivial amount of work, and Ambari might just be the Œgood enough¹ path of least resistance; at least being open source some hooks could be added to support orchestration tools. - SteveN On 2013-12-31 14:21 , Konstantin Boudnik c...@apache.org wrote: The reason for not-much activity on that JIRA is because Ambari seems to be drifting away from real-life orchestration systems like Puppet toward something else. And that's exactly why I prefer to use Puppet or Chef orchestration - you have a state machine that works in the same way on every supported platform. In 20 years of doing system and network administration as well as software development I've seen times and again how the fancy UI applications fail to deliver on their promise. Main reason is in the shifting of the focus on the bling instead of the core functionality. Cloudera's CM is a perfect example of my point, because it is doing totally heinous things with standard Linux services, their life-cycle and configurations up to the point where any sane Hadoop devops would be helpless to do anything without CM. Ambrai seems to be a bit better in this respect. However, with the development above and replacement of the Puppet I am not sure how much longer it will be the case. Happy New Year everyone! Cos On Tue, Dec 31, 2013 at 03:10PM, Steven Núñez wrote: Thanks. That issue answers pretty much all the questions. I¹d certainly give it a +1 if I had a login. That definitely seems like the right direction to move in. I don¹t know the internals, but if everyone is using BigTop for packaging, perhaps there¹s some way to read the manifest files (if that¹s what they¹re called) to produce what Ambari needs for management. From: Chris Mildebrandt ch...@woodenrhino.commailto:ch...@woodenrhino.com Reply-To: user@bigtop.apache.orgmailto:user@bigtop.apache.org user@bigtop.apache.orgmailto:user@bigtop.apache.org Date: Tuesday, 31 December 2013 7:57 To: u...@ambari.apache.orgmailto:u...@ambari.apache.org u...@ambari.apache.orgmailto:u...@ambari.apache.org Cc: user@bigtop.apache.orgmailto:user@bigtop.apache.org user@bigtop.apache.orgmailto:user@bigtop.apache.org Subject: Re: Cluster Management: OpenSource Vendor Options You may want to watch this: https://issues.apache.org/jira/browse/AMBARI-3524 and include it in your write-up for future considerations. Though there hasn't been much activity on it. On Mon, Dec 30, 2013 at 8:32 AM, Steven Núñez steven.nu...@illation.commailto:steven.nu...@illation.com wrote: Seasons Greetings All, I¹m doing a bit of a write-up on the various Hadoop distributions and would like to understand exactly what packages are installed by the Apache version of Ambari. It¹s an exciting place to be working (big data Hadoop) but the lines are blurred in many ways. The way I see the open source landscape now is something like this (from a management/installation/configuration perspective): BigTop - RPM like packaging for Hadoop Ambari - GUI management/monitoring/provisioning Looking at it from a vendor perspective, we¹ve got (I know there are others, this is just for discussion): BigTop (packaging) CDH HDP Apache Bigtop Cloudera Cloudera Manager (closed source, commercial) Hortonworks / Apache Ambari(open source) The CDH, BigTop and HDP (I assume) base distributions require a lot of manual configuration, so the best way to spin up a cluster with a reasonable set of applications (say HDFS, YARN, Hive, HCatalog, HBase, ZooKeeper, Oozie, Pig, Sqoop) is to use CDH + CM or Ambari + HDP. Is there an equivalent for Apache? If I use the kit found at ambari.apache.orghttp://ambari.apache.org to spin up a cluster, do I get Apache components, or the HDP distribution? I¹m trying to define the ŒApache distribution¹ in my mind, if there is one, and understand exactly what its capabilities are, and cluster management is rather fundamental, since not many folks have the luxury of spending time climbing the long, steep learning curve of Hadoop ecosystem configuration. Cheers, - SteveN
Re: Cluster Management: OpenSource Vendor Options
Thanks. That issue answers pretty much all the questions. I’d certainly give it a +1 if I had a login. That definitely seems like the right direction to move in. I don’t know the internals, but if everyone is using BigTop for packaging, perhaps there’s some way to read the manifest files (if that’s what they’re called) to produce what Ambari needs for management. From: Chris Mildebrandt ch...@woodenrhino.commailto:ch...@woodenrhino.com Reply-To: user@bigtop.apache.orgmailto:user@bigtop.apache.org user@bigtop.apache.orgmailto:user@bigtop.apache.org Date: Tuesday, 31 December 2013 7:57 To: u...@ambari.apache.orgmailto:u...@ambari.apache.org u...@ambari.apache.orgmailto:u...@ambari.apache.org Cc: user@bigtop.apache.orgmailto:user@bigtop.apache.org user@bigtop.apache.orgmailto:user@bigtop.apache.org Subject: Re: Cluster Management: OpenSource Vendor Options You may want to watch this: https://issues.apache.org/jira/browse/AMBARI-3524 and include it in your write-up for future considerations. Though there hasn't been much activity on it. On Mon, Dec 30, 2013 at 8:32 AM, Steven Núñez steven.nu...@illation.commailto:steven.nu...@illation.com wrote: Seasons Greetings All, I’m doing a bit of a write-up on the various Hadoop distributions and would like to understand exactly what packages are installed by the Apache version of Ambari. It’s an exciting place to be working (big data Hadoop) but the lines are blurred in many ways. The way I see the open source landscape now is something like this (from a management/installation/configuration perspective): BigTop - RPM like packaging for Hadoop Ambari - GUI management/monitoring/provisioning Looking at it from a vendor perspective, we’ve got (I know there are others, this is just for discussion): BigTop (packaging) CDH HDP Apache Bigtop Cloudera Cloudera Manager (closed source, commercial) Hortonworks / Apache Ambari(open source) The CDH, BigTop and HDP (I assume) base distributions require a lot of manual configuration, so the best way to spin up a cluster with a reasonable set of applications (say HDFS, YARN, Hive, HCatalog, HBase, ZooKeeper, Oozie, Pig, Sqoop) is to use CDH + CM or Ambari + HDP. Is there an equivalent for Apache? If I use the kit found at ambari.apache.orghttp://ambari.apache.org to spin up a cluster, do I get Apache components, or the HDP distribution? I’m trying to define the ‘Apache distribution’ in my mind, if there is one, and understand exactly what its capabilities are, and cluster management is rather fundamental, since not many folks have the luxury of spending time climbing the long, steep learning curve of Hadoop ecosystem configuration. Cheers, - SteveN
Re: Cluster Management: OpenSource Vendor Options
On Tue, Dec 31, 2013 at 7:10 AM, Steven Núñez steven.nu...@illation.com wrote: Thanks. That issue answers pretty much all the questions. I’d certainly give it a +1 if I had a login. That definitely seems like the right direction to move in. I don’t know the internals, but if everyone is using BigTop for packaging, perhaps there’s some way to read the manifest files (if that’s what they’re called) to produce what Ambari needs for management. From the Bigtop side of things -- I'd love to see a better integration between Ambari and Bigtop. As a datapoint -- I poked around HDP2 (the distro that Ambari seems to support in the best possible way) and it looked very similar to Bigtop in layout and everything else. Perhaps it won't be that big of a deal to adopt Ambari to support Bigtop distro natively. That would have an added benefit of all the Bigtop-derived distros (Cloudera, Hortonworks, Intel, Pivotal, WANDisco) getting the baseline support for free. I'd love to help from Bigtop side of things, but my Ambari foo is weak enough to request somebody from the Ambari team of developers to help. So... if there's enough interest, perhaps we can find a way? Thanks, Roman.
Re: Cluster Management: OpenSource Vendor Options
The reason for not-much activity on that JIRA is because Ambari seems to be drifting away from real-life orchestration systems like Puppet toward something else. And that's exactly why I prefer to use Puppet or Chef orchestration - you have a state machine that works in the same way on every supported platform. In 20 years of doing system and network administration as well as software development I've seen times and again how the fancy UI applications fail to deliver on their promise. Main reason is in the shifting of the focus on the bling instead of the core functionality. Cloudera's CM is a perfect example of my point, because it is doing totally heinous things with standard Linux services, their life-cycle and configurations up to the point where any sane Hadoop devops would be helpless to do anything without CM. Ambrai seems to be a bit better in this respect. However, with the development above and replacement of the Puppet I am not sure how much longer it will be the case. Happy New Year everyone! Cos On Tue, Dec 31, 2013 at 03:10PM, Steven Núñez wrote: Thanks. That issue answers pretty much all the questions. I’d certainly give it a +1 if I had a login. That definitely seems like the right direction to move in. I don’t know the internals, but if everyone is using BigTop for packaging, perhaps there’s some way to read the manifest files (if that’s what they’re called) to produce what Ambari needs for management. From: Chris Mildebrandt ch...@woodenrhino.commailto:ch...@woodenrhino.com Reply-To: user@bigtop.apache.orgmailto:user@bigtop.apache.org user@bigtop.apache.orgmailto:user@bigtop.apache.org Date: Tuesday, 31 December 2013 7:57 To: u...@ambari.apache.orgmailto:u...@ambari.apache.org u...@ambari.apache.orgmailto:u...@ambari.apache.org Cc: user@bigtop.apache.orgmailto:user@bigtop.apache.org user@bigtop.apache.orgmailto:user@bigtop.apache.org Subject: Re: Cluster Management: OpenSource Vendor Options You may want to watch this: https://issues.apache.org/jira/browse/AMBARI-3524 and include it in your write-up for future considerations. Though there hasn't been much activity on it. On Mon, Dec 30, 2013 at 8:32 AM, Steven Núñez steven.nu...@illation.commailto:steven.nu...@illation.com wrote: Seasons Greetings All, I’m doing a bit of a write-up on the various Hadoop distributions and would like to understand exactly what packages are installed by the Apache version of Ambari. It’s an exciting place to be working (big data Hadoop) but the lines are blurred in many ways. The way I see the open source landscape now is something like this (from a management/installation/configuration perspective): BigTop - RPM like packaging for Hadoop Ambari - GUI management/monitoring/provisioning Looking at it from a vendor perspective, we’ve got (I know there are others, this is just for discussion): BigTop (packaging) CDH HDP Apache Bigtop Cloudera Cloudera Manager (closed source, commercial) Hortonworks / Apache Ambari(open source) The CDH, BigTop and HDP (I assume) base distributions require a lot of manual configuration, so the best way to spin up a cluster with a reasonable set of applications (say HDFS, YARN, Hive, HCatalog, HBase, ZooKeeper, Oozie, Pig, Sqoop) is to use CDH + CM or Ambari + HDP. Is there an equivalent for Apache? If I use the kit found at ambari.apache.orghttp://ambari.apache.org to spin up a cluster, do I get Apache components, or the HDP distribution? I’m trying to define the ‘Apache distribution’ in my mind, if there is one, and understand exactly what its capabilities are, and cluster management is rather fundamental, since not many folks have the luxury of spending time climbing the long, steep learning curve of Hadoop ecosystem configuration. Cheers, - SteveN signature.asc Description: Digital signature
Re: Cluster Management: OpenSource Vendor Options
On 12/30/2013 08:32 AM, Steven Núñez wrote: The CDH, BigTop and HDP (I assume) base distributions require a lot of manual configuration, so the best way to spin up a cluster with a reasonable set of applications (say HDFS, YARN, Hive, HCatalog, HBase, ZooKeeper, Oozie, Pig, Sqoop) is to use CDH + CM or Ambari + HDP. Some people have also automated this through tools such as Puppet, Chef or Ansible. Thanks, Bruno
