Please reply if you use Mesos fine grained mode

classic Classic list List threaded Threaded
10 messages Options
Reply | Threaded
Open this post in threaded view
|

Please reply if you use Mesos fine grained mode

rxin
If you are using Spark with Mesos fine grained mode, can you please respond to this email explaining why you use it over the coarse grained mode?

Thanks.

Reply | Threaded
Open this post in threaded view
|

Re: Please reply if you use Mesos fine grained mode

Soren Macbeth
we use fine-grained mode. coarse-grained mode keeps JVMs around which often leads to OOMs, which in turn kill the entire executor, causing entire stages to be retried. In fine-grained mode, only the task fails and subsequently gets retried without taking out an entire stage or worse. 

On Tue, Nov 3, 2015 at 3:54 PM, Reynold Xin <[hidden email]> wrote:
If you are using Spark with Mesos fine grained mode, can you please respond to this email explaining why you use it over the coarse grained mode?

Thanks.


Reply | Threaded
Open this post in threaded view
|

Re: Please reply if you use Mesos fine grained mode

Jerry Lam
We "used" Spark on Mesos to build interactive data analysis platform because the interactive session could be long and might not use Spark for the entire session. It is very wasteful of resources if we used the coarse-grained mode because it keeps resource for the entire session. Therefore, fine-grained mode was used. 

Knowing that Spark now supports dynamic resource allocation with coarse grained mode, we were thinking about using it. However, we decided to switch to Yarn because in addition to dynamic allocation, it has better supports on security. 

On Tue, Nov 3, 2015 at 7:22 PM, Soren Macbeth <[hidden email]> wrote:
we use fine-grained mode. coarse-grained mode keeps JVMs around which often leads to OOMs, which in turn kill the entire executor, causing entire stages to be retried. In fine-grained mode, only the task fails and subsequently gets retried without taking out an entire stage or worse. 

On Tue, Nov 3, 2015 at 3:54 PM, Reynold Xin <[hidden email]> wrote:
If you are using Spark with Mesos fine grained mode, can you please respond to this email explaining why you use it over the coarse grained mode?

Thanks.



Reply | Threaded
Open this post in threaded view
|

Re: Please reply if you use Mesos fine grained mode

rxin
In reply to this post by Soren Macbeth
Soren,

If I understand how Mesos works correctly, even the fine grained mode keeps the JVMs around?


On Tue, Nov 3, 2015 at 4:22 PM, Soren Macbeth <[hidden email]> wrote:
we use fine-grained mode. coarse-grained mode keeps JVMs around which often leads to OOMs, which in turn kill the entire executor, causing entire stages to be retried. In fine-grained mode, only the task fails and subsequently gets retried without taking out an entire stage or worse. 

On Tue, Nov 3, 2015 at 3:54 PM, Reynold Xin <[hidden email]> wrote:
If you are using Spark with Mesos fine grained mode, can you please respond to this email explaining why you use it over the coarse grained mode?

Thanks.



Reply | Threaded
Open this post in threaded view
|

Re: Please reply if you use Mesos fine grained mode

MEETHU MATHEW
In reply to this post by rxin
Hi,

We are using Mesos fine grained mode because we can have multiple instances of spark to share machines and each application get resources dynamically allocated. 
 
Thanks & Regards, 
 Meethu M



On Wednesday, 4 November 2015 5:24 AM, Reynold Xin <[hidden email]> wrote:


If you are using Spark with Mesos fine grained mode, can you please respond to this email explaining why you use it over the coarse grained mode?

Thanks.



Reply | Threaded
Open this post in threaded view
|

Re: Please reply if you use Mesos fine grained mode

Timothy Chen
In reply to this post by rxin
Fine grain mode does reuse the same JVM but perhaps different placement or different allocated cores comparing to the same total memory allocation.

Tim

Sent from my iPhone

On Nov 3, 2015, at 6:00 PM, Reynold Xin <[hidden email]> wrote:

Soren,

If I understand how Mesos works correctly, even the fine grained mode keeps the JVMs around?


On Tue, Nov 3, 2015 at 4:22 PM, Soren Macbeth <[hidden email]> wrote:
we use fine-grained mode. coarse-grained mode keeps JVMs around which often leads to OOMs, which in turn kill the entire executor, causing entire stages to be retried. In fine-grained mode, only the task fails and subsequently gets retried without taking out an entire stage or worse. 

On Tue, Nov 3, 2015 at 3:54 PM, Reynold Xin <[hidden email]> wrote:
If you are using Spark with Mesos fine grained mode, can you please respond to this email explaining why you use it over the coarse grained mode?

Thanks.



Reply | Threaded
Open this post in threaded view
|

Re: Please reply if you use Mesos fine grained mode

Heller, Chris
In reply to this post by MEETHU MATHEW
We’ve been making use of both. Fine-grain mode makes sense for more ad-hoc work loads, and coarse-grained for more job like loads on a common data set. My preference is the fine-grain mode in all cases, but the overhead associated with its startup and the possibility that an overloaded cluster would be starved for resources makes coarse grain mode a reality at the moment. 

On Wednesday, 4 November 2015 5:24 AM, Reynold Xin <[hidden email]> wrote:


If you are using Spark with Mesos fine grained mode, can you please respond to this email explaining why you use it over the coarse grained mode?

Thanks.



Reply | Threaded
Open this post in threaded view
|

Re: Please reply if you use Mesos fine grained mode

Timothy Chen
Hi Chris,

How does coarse grain mode gives you less starvation in your overloaded cluster? Is it just because it allocates all resources at once (which I think in a overloaded cluster allows less things to run at once).

Tim


On Nov 4, 2015, at 4:21 AM, Heller, Chris <[hidden email]> wrote:

We’ve been making use of both. Fine-grain mode makes sense for more ad-hoc work loads, and coarse-grained for more job like loads on a common data set. My preference is the fine-grain mode in all cases, but the overhead associated with its startup and the possibility that an overloaded cluster would be starved for resources makes coarse grain mode a reality at the moment. 

On Wednesday, 4 November 2015 5:24 AM, Reynold Xin <[hidden email]> wrote:


If you are using Spark with Mesos fine grained mode, can you please respond to this email explaining why you use it over the coarse grained mode?

Thanks.



Reply | Threaded
Open this post in threaded view
|

Re: Please reply if you use Mesos fine grained mode

Iulian Dragoș
Probably because only coarse-grained mode respects `spark.cores.max` right now. See (and maybe review ;-)) #9027 (sorry for the shameless plug).

iulian

On Wed, Nov 4, 2015 at 5:05 PM, Timothy Chen <[hidden email]> wrote:
Hi Chris,

How does coarse grain mode gives you less starvation in your overloaded cluster? Is it just because it allocates all resources at once (which I think in a overloaded cluster allows less things to run at once).

Tim


On Nov 4, 2015, at 4:21 AM, Heller, Chris <[hidden email]> wrote:

We’ve been making use of both. Fine-grain mode makes sense for more ad-hoc work loads, and coarse-grained for more job like loads on a common data set. My preference is the fine-grain mode in all cases, but the overhead associated with its startup and the possibility that an overloaded cluster would be starved for resources makes coarse grain mode a reality at the moment. 

On Wednesday, 4 November 2015 5:24 AM, Reynold Xin <[hidden email]> wrote:


If you are using Spark with Mesos fine grained mode, can you please respond to this email explaining why you use it over the coarse grained mode?

Thanks.






--

--
Iulian Dragos

------
Reactive Apps on the JVM
www.typesafe.com

Reply | Threaded
Open this post in threaded view
|

Re: Please reply if you use Mesos fine grained mode

Heller, Chris
In reply to this post by Timothy Chen
Correct. Its just that with coarse mode we grab the resources up front, so its either available or not. But using resources on demand, as with a fine grained mode, just means the potential to starve out an individual job. There is also the sharing of RDDs that coarse gives you which would need something like Tachyon to achieve in fine grain mode.


From: Timothy Chen <[hidden email]>
Date: Wednesday, November 4, 2015 at 11:05 AM
To: "Heller, Chris" <[hidden email]>
Cc: Reynold Xin <[hidden email]>, "[hidden email]" <[hidden email]>
Subject: Re: Please reply if you use Mesos fine grained mode

Hi Chris,

How does coarse grain mode gives you less starvation in your overloaded cluster? Is it just because it allocates all resources at once (which I think in a overloaded cluster allows less things to run at once).

Tim


On Nov 4, 2015, at 4:21 AM, Heller, Chris <[hidden email]> wrote:

We’ve been making use of both. Fine-grain mode makes sense for more ad-hoc work loads, and coarse-grained for more job like loads on a common data set. My preference is the fine-grain mode in all cases, but the overhead associated with its startup and the possibility that an overloaded cluster would be starved for resources makes coarse grain mode a reality at the moment. 

On Wednesday, 4 November 2015 5:24 AM, Reynold Xin <[hidden email]> wrote:


If you are using Spark with Mesos fine grained mode, can you please respond to this email explaining why you use it over the coarse grained mode?

Thanks.