we are looking to build a cluster of Compute Nodes for Deep Learning model training jobs, some of them on the cloud and others locally, that have NVIDIA GPUs in them. We felt that using Mesos and the framework Marathon (M&M) would be our best options to schedule the cluster. However the documentations for (M&M) seem to be very ambiguous (or at least to me, sorry I'm an intern) and I'm running into a lot of issues concerning Zookeeper and the connections between the nodes.
Plus, it seems like Mesosphere are giving much more importance to DC/OS when it comes to tutorials and docs, and I guess it will also be patched more regularly and its interfaces (GUI and CLI) look much more user-friendly.
So I was wondering if by dropping the exploration of (M&M) and moving to DC/OS, would we lose a lot of control over the cluster? In M&M do we have perks that cannot be given in the Open Source Edition of DC/OS? like monitoring the machines, logging results etc.. If I ask my manager we might also get the Enterprise edition so that's not really a problem, but does DC/OS apply an abstraction layer that isn't really preferable to advanced users?
DC/OS is build around Apache Mesos and Marathon and gives a good default setup for zookeeper, networking, .... So IMO it is a good place to start as you can still use all M&M and Mesos features + the DC/OS features and ease of setup.
Disclaimer: I am working for Mesosphere.
Related
I was wondering how compatible Dask is with cluster orchestrators other than Kubernetes. In particular, I am interested in Marathon and Nomad. From my research, Nomad doesn't seem to address Dask at all, although it claims to be agnostic, being able to handle "anything." Marathon does have a GitHub project named dask-marathon, however it hasn't been updated in 4 years, and it specifically states that the project is a proof of concept onlym with no guarantee of quality or future maintenance. Any words of wisdom here would be greatly appreciated, since I do not want to use Kubernetes at this point.
For options on how to deploy Dask I recommend looking at the setup documentation: https://docs.dask.org/en/latest/setup.html
You're correct that the marathon project hasn't been updated in a long while. There have been some other dask+mesos projects in the past. I don't know how active they are though.
I'm trying to figure out what would be the reasons for using Mesos. Can you come up with other ones?
Running all of your services in the same cluster instead of dedicated clusters (your end-applications + DevOps such as Jenkins)
Running different maturity applications in same cluster (dev, test, production), or is this viable? Kubernetes has a similar approach with Labels
Mesos simplifies the use of traditional distributed applications such as Hadoop by easing deployment, unified API, bin-packing of resources
Full-disclosure: I currently work at Twitter and I'm involved in both Apache Mesos and Aurora.
Mesos uses cases can vary based upon a few dimensions: scale (10 servers vs 10s of thousands), available hardware (dedicated/static or in the public cloud/scalable), and workloads (primarily services, batch, or both).
Your list is a great start. Here are a few additional use cases / features to add.
Container Orchestration
As container runtimes like Docker have become popular, lots of potential users are looking at Mesos + a scheduler to manage orchestration once container images are created. Mesos is already quite mature and has been proven at scale, which I think has given it a leg up over some emergent solutions.
Increased Resource Utilization
For companies running >50 servers, a common motivation for adopting Mesos is to increase resource utilization to reduce CapEx. There are a number of examples of this in both the public and private cloud. In the case of Ebay they have been running Jenkins on Mesos and were able to reduce their VM footprint. Mesosphere has also published a case study of HubSpot (runnning on AWS), and how they've been able to replace hundreds of smaller servers with dozens of larger ones by more-efficiently using their available hardware.
Preemption
At Twitter we're running Mesos via one scheduler: Apache Aurora. One of the ways we can improve utilization relates to your use case: running different maturity applications in the same cluster. Aurora has a concept of environments, so you can run applications that are production, development, or test. Additionally, Aurora has a built-in preemption feature which allows it to prioritize production over non-production tasks, killing non-production tasks when those resources are needed to run production ones as well as a priority system within each environment.
Long-term, functionality related to preemption will also be located in the Mesos core itself -- it's a killer feature related to both increased resource utilization and running different maturity applications (dev, test, prod). There are a few Mesos tickets to follow if you're interested in keeping up to date, including MESOS-155 for preemption, and MESOS-1474 for inverse offers.
Colocating Batch and Services
Running batch and services in a shared Mesos cluster will be key to driving up utilization even further as js84 points out. Check out Project Myriad, an effort to colocate Mesos and YARN workloads in the same cluster. At this time I'm not aware of any large deployments running both batch and services, but it's certainly the direction the community is moving in as it becomes easier for multiple frameworks to run in a shared cluster.
At least one additional use case comes to mind: Development SDK for developing distributed applications. If you have a look at Mesos Frameworks you will find a number of frameworks which have been developed on top of Mesos. Also interesting Apple's Siri framework powering Siri.
Regarding your 1): One additional angle you should keep in mind here is scaling your applications in the same cluster. I.e. at peak load of your website, shift resources easily towards the webservers while scaling down the Hadoop analytical processing.
Both Marathon and Aurora are built on Mesos and supposedly are engineered for running long running services. My questions are:
What are their differences? I have struggled in finding any good explanations regarding their key differences
Do these frameworks run anything that runs on Linux? For Marathon they state that it can run anything that "is executable in a shell" but this is sort of vague :)
Thanks!
Disclaimer: I am the VP of Apache Aurora, and have been the tech lead of the Aurora team at Twitter for ~5 years. My likely-biased opinions are my own and do not necessarily represent those of Twitter or the ASF.
Do these frameworks run anything that runs on Linux? For Marathon they
state that it can run anything that "is executable in a shell" but
this is sort of vague :)
Essentially, yes. Ultimately these systems are sophisticated machinery to execute shell code somewhere in a cluster :-)
What are their differences? I have struggled in finding any good
explanations regarding their key differences
Aurora and Marathon do indeed offer similar feature sets, both being classified as "service schedulers". In other words, you hand us instructions for how to run your application servers, and we do our best to keep them up.
I'll offer some differences in broad strokes. When it comes to shortcomings mentioned in each, I think it's safe to say that the communities are aware and intend to fix them.
Ease of use
Aurora is not easy to install. It will likely feel like you are trailblazing while setting it up. It exposes a thrift API, which means you'll need a thrift client to interact with it programmatically (a REST-like API is coming, but is vaporware at the moment), or use our command line client. Aurora has a DSL for configuration which can be daunting, but allows you to easily share templates and common patterns as you use the system more.
Marathon, on the other hand, helps you to run 'Hello World' as quickly as possible. It has great docs to do this in many environments and there's little overhead to get going. It has a REST API, making it easier to adapt to custom tools. It uses JSON for configuration, which is easy to start with but more prone to cargo culting.
Targeted use cases
Aurora has always been designed to handle a large engineering organization. The clusters at Twitter have tens of thousands of machines and hundreds of engineers using them. It is critical to Twitter's business. As a result, we take our requirements of scale, stability, and security very seriously. We make sure to only condone features that we believe are trustworthy at scale in production (for example, we have our Docker support labeled as beta because of known issues with Docker itself and the Mesos-Docker integration). We also have features like preemption that make our clusters suitable for mixing business-critical services with prototypes and experiments.
I can't make any claim for or against Marathon's scalability. On the feature front, Marathon has build out features quickly, but this can feel bleeding edge in practice (Docker support is a good example). This is not always due to Marathon itself, but also layers down the stack. Marathon does not provide preemption.
Ownership
To some, ownership and governance of a project is important. It feel that in practice it does not define the openness of a project, but for some people/companies the legal fine print can be a deal-breaker.
Marathon is owned by a company (Mesosphere)
To some, this is beneficial, to others is is not. It means that you can pay for support and features. It also means that there is something to be sold, and the project direction is ultimately decided by Mesosphere's interests.
Aurora is owned by the Apache Software Foundation
This means it is subject to the governance model of the ASF, driven by the community. Aurora does not have paying customers, and there is not currently a software shop that you can pay for development.
tl;dr If you are just getting your feet wet with running services on Mesos, I would suggest Marathon as your first port of call. It will be easier for you to get running and poke around the ecosystem. If you are forming the 'private cloud strategy' for a company, I suggest seriously considering Aurora, as it is proven and specifically designed for that.
So I've been evaluating both and this is my summary.
Aurora
[+] also handles recurring jobs
[+] finer grained, extensive file-based configuration
[+] has namespaces so multiple environments can co-exist
[-] read-only UI, no official API
[~] file based configuration and cli based execution brings overhead (which can be justified with more extensive feature set)
Marathon
[+] very easy to setup and use
[+] UI that provides control and extensive API (even with features missing from UI at the moment)
[+] event bus to listen in on api calls
[-] handles only long-running jobs
[-] does not have separate deployment-run-cleanup steps, these if necessary need to be combined in a script of one-liner
Even though Aurora has better capabilities, I prefer Marathon due to Auroras complexity/overhead and lack of UI (for control) & API
I have more experience with Marathon.
Ideological:
Marathon is a relatively tested product that is used in production at AirBnB. Aurora is an early Apache project (so YMMV).
Both are open source and active. Feel free to contribute pull requests or file issues!
Technical:
Marathon doesn't schedule batch tasks or cron jobs
Marathon has a friendly UI and better health indicators (in 0.8.x)
In regards to your second question, you can run any command or docker container, and Mesos will do the resource isolation for you. If you have 50% CentOS nodes and 50% Ubuntu nodes and you run a task that executes apt-get, the task will have a 50% chance of failure. Mesos and Marathon have no awareness of the actual machines.
Disclaimer: I don't have hands-on experience with Aurora, only with Marathon.
ad Q1: In a nutshell Apache Aurora is capable of doing what Marathon + Chronos can provide, that is, schedule both long-running services and recurring (batch) jobs; see also Aurora user guide.
ad Q2: Yes, anything. Currently based on cgroups and Docker but hey, you can roll your own.
I am relatively new to all these, but I'm having troubles getting a clear picture among the listed technologies.
Though, all of these try to solve different problems, but do have things in common too. I would like to understand what are the things that are common and what is different. It is likely that the combination of few would be great fit, if so what are they?
I am listing a few of them along with questions, but it would be great if someone lists all of them in detail and answers the questions.
Kubernetes vs Mesos:
This link
What's the difference between Apache's Mesos and Google's Kubernetes
provides a good insight into the differences, but I'm unable to understand as to why Kubernetes should run on top of Mesos. Is it more to do with coming together of two opensource solutions?
Kubernetes vs Core-OS Fleet:
If I use kubernetes, is fleet required?
How does Docker-Swarm fit into all the above?
Disclosure: I'm a lead engineer on Kubernetes
I think that Mesos and Kubernetes are largely aimed at solving similar problems of running clustered applications, they have different histories and different approaches to solving the problem.
Mesos focuses its energy on very generic scheduling, and plugging in multiple different schedulers. This means that it enables systems like Hadoop and Marathon to co-exist in the same scheduling environment. Mesos is less focused on running containers. Mesos existed prior to widespread interest in containers and has been re-factored in parts to support containers.
In contrast, Kubernetes was designed from the ground up to be an environment for building distributed applications from containers. It includes primitives for replication and service discovery as core primitives, where-as such things are added via frameworks in Mesos. The primary goal of Kubernetes is a system for building, running and managing distributed systems.
Fleet is a lower-level task distributor. It is useful for bootstrapping a cluster system, for example CoreOS uses it to distribute the kubernetes agents and binaries out to the machines in a cluster in order to turn-up a kubernetes cluster. It is not really intended to solve the same distributed application development problems, think of it more like systemd/init.d/upstart for your cluster. It's not required if you run kubernetes, you can use other tools (e.g. Salt, Puppet, Ansible, Chef, ...) to accomplish the same binary distribution.
Swarm is an effort by Docker to extend the existing Docker API to make a cluster of machines look like a single Docker API. Fundamentally, our experience at Google and elsewhere indicates that the node API is insufficient for a cluster API. You can see a bunch of discussion on this here: https://github.com/docker/docker/pull/8859 and here: https://github.com/docker/docker/issues/8781
Join us on IRC # #google-containers if you want to talk more.
I think the simplest answer is that there is no simple answer. The swift rise to power of containers, and Docker in particular has left a power vacuum for "container scheduling and orchestration", whatever that might mean. In reality, that means you have a number of technologies that can work in harmony on some levels, but with certain aspects in competition. For example, Kubernetes can be used as a one stop shop for deploying and managing containers on a compute cluster (as Google originally designed it), but could also sit atop Fleet, making use of the resilience tier that Fleet provides on CoreOS.
As this Google vid states Kubernetes is not a complete out the box container scaling solution, but is a good statement to start from. In the same way, you would at some stage expect Apache Mesos to be able to work with Kubernetes, but not with Marathon, in as much as Marathon appears to fulfil the same role as Kubernetes. Somewhere I think I've read these could become part of the same effort, but I could be wrong about that - it's really about the strategic direction of Mesosphere and the corresponding adoption of Kubernetes principles.
In the DockerCon keynote, Solomon Hykes suggested Swarm would be a tier that could provide a common interface onto the many orchestration and scheduling frameworks. From what I can see, Swarm is designed to provide a smooth Docker deployment workflow, working with some existing container workflow frameworks such as Deis, but flexible enough to yield to "heavyweight" deployment and resource management such as Mesos.
Hope this helps - this could be an enormous post. I think the key is that these are young, evolving services that will likely merge and become interoperable, but we need to ride out the next 12 months to see how it plays out. There's some very clever people on the problem, so the future looks very bright.
As far as I understand it:
Mesos, Kubernetes and Fleet are all trying to solve a very similar problem. The idea is that you abstract away all your hardware from developers and the 'cluster management tool' sorts it all out for you. Then all you need to do is give a container to the cluster, give it some info (keep it running permanently, scale up if X happens etc) and the cluster manager will make it happen.
With Mesos, it does all the cluster management for you, but it doesn't include the scheduler. The scheduler is the bit that says, ok this process needs 2 procs and 512MB RAM, and I have a machine over there with that free, so I'll run it on that machine. There are some plugin schedulers available for Mesos: Marathon and Chronos and you can write your own. This gives you a lot of power of resource distribution and cluster scaling etc.
Fleet and Kubernetes seem to abstract away those sorts of details (so you don't have to write your own scheduler basically). This means you have to define your tasks and submit them in the format/manner defined by Fleet or Kubernetes and then they take over and schedule the tasks (containers) for you.
So I guess: Using Mesos may mean a bit more work in writing your own scheduler, but potentially provides more flexibility if required.
I think the idea of running Kubernetes on top of Mesos is that Kubernetes acts as the scheduler for Mesos. Personally I'm not sure what benefits this brings over running one or the other on its own though (hopefully someone will jump in and explain!)
As MikeB said.. it's early days, and it's all up for grabs (keep an eye on Amazon's ECS as well) so there are many competing standards and a lot of overlap!
-edit- I didn't mention Docker swarm as I don't really have much experience with it.
For anyone coming to this after 2017 fleet is deprecated. Do not use it anymore.
Fleet docs say "fleet is no longer actively developed or maintained by CoreOS" and link to Container orchestration: Moving from fleet to Kubernetes. Fleet was removed from Container Linux (formerly known as CoreOS Linux) and replaced with Kubernetes kubelet (agent). This coincided with a corporate pivot to offer Tectonic (a Kubernetes distro) as their primary product.
I am learning Mapreduce and Hadoop now. I know I can do some tests and run some samples on a singe node. But I really want to do some practice on a real distributed environment. So I want to ask :
Is there a website which can offer a distributed environment for me to do some experiments?
Somebody told me that I can use Amazon web service to build a distributed environment. Is it real? Does someone have such an experience?
And I want to know how you guys learn hadoop before you use it in your work?
Thank you!
There are a few options:
If you just want to learn about the Map/Reduce paradigm, I would recommend you take a look at JSMapReduce. This is embedded directly in the browser, you have nothing to install, and you can create real Map/Reduce programs.
If you want to learn about Hadoop specifically, Amazon has this thing called Elastic Map Reduce which is essentially Hadoop running on AWS, so this enables you to write your Hadoop job, decide how many machines you want in your cluster, which type of machines you want, and then run it, and EMR will do everything, bootstrap the machines for you, run your job and store the results on S3. I would recommend looking at this tutorial to get an idea how to setup a job on EMR. Just remember, EMR is not free, so you'll have to pay for your computing resources.
Alternatively if you're not looking to pay the cost of EMR, you could always setup Hadoop on your local machine in non-distributed mode, and experiment with it, as described here. Even if it's a single node setup, the abstractions will be the same as if you were using a big cluster, so it's a good way to get up to speed and then go on EMR or a real cluster when you want to get serious.
Amazon offers a free tier, so you can spin up some vms and try experimenting that way. The micro instances they have aren't very powerful, but are fine for small scale tests.
You can also spin up VMs on your desktop if it is powerful enough. I have done this myself using VMPlayer. You can install any flavor of Linux you like for free. Ubuntu is pretty easy to start with. When you setup the networking for your VMs, be sure to use bridged networking. That way each VM will get its own IP address on your network so they can communicate with each other.
Well, it's maybe not about '100% online' but should give really good alternative with some details.
If you are not ready to pay for online cluster resources (such as EMR solution mentioned here) and you don't like to build your own cluster but you are not satisfied with single node setup, you can try to build virtual cluster on powerful enough desktop.
You need minimun 3 VM, I prefer Ubuntu. 4 is better. To see real Hadoop you need minimal replication factor 3. So you need 3 dataNode, 3 taskTrackers. Well, you also need nameNode / JobTracker - it could be one of nodes used for dataNode but I'd recommend to have separate VM. If you need HBase, for example, you again need one Master and minimum 3 RegionServer. So, again, you need 3 but better 4 VM,
There is pretty good free product, Cloudera CDH which is 'somewhat commercial' Hadoop distribution. They also have manager with GUI and simplified installation. BTW they have even prepared demo VMs but I never have used them. You can download everything here. They also host lot of materials about Hadoop and their environment.
Alternative between completely free solution with VMs on desktop and paid service like EMR is your virtual cluster built on top of one dedicated server if you have spare. This is what I personally did. One physical server powered by VmWare free solution, 4 virtual machine, 1 SSD for OS and 3 'general' HDD for storages. Every VM runs Ubuntu 11.04 (again free). Cloudera manager free edition, CDH. So everything is free but you need some hardware that is often available as spare. And you have playground. OK, you need to invest time but by my mind you will get greatest experience from this approach.
Although I do not know much about it, another option may be Greenplum's analytic workbench (1000 node cluster w/ Hadoop for testing): http://www.greenplum.com/solutions/analytics-workbench