spark on mesos with docker issue

[Stavros Kontopoulos]

Hi,

Im using spark on mesos on docker. I have linked my slaves to the master
and a
spark repl works fine inside the master container.

If i try to crate the same spark repl form the host i get stuck at the
point when the framework tries to register to the mesos master (here the
framework is the spark repl itself).
I can ping the container from my host and vice versa. So networking its not
the problem.
What i noticed form the logs is that mesos does not resolve the correct ip:

Framework failover timeout, removing framework
b3605c33-f573-4d40-806f-b9b0abee2e32-0012 (Spark shell) at
scheduler-b6dbe4f8-4760-4cbe-bb5e-a991e35aba1b@127.0.1.1:40186

docker0 interface is on 172.17.x.x and my host is one such ip so i didnt
expect there to see
127.0.1.1. I have tried several things like spark.driver.host,
SPARK_LOCAL_IP to be set correctly but with no result...
I suspect this is a mesos problem on docker...

Thnx,

S.

Re: mesos-ui

[Benjamin Mahler]

This is great Alberto! Seems like we should have a document page on the
website that tracks alternative third party webui choices. If you could
send a patch that would be much appreciated.

Ben

On Thu, Oct 22, 2015 at 5:11 AM, Alberto García Lamela <
alberto.garc...@hotmail.com> wrote:

> Hi there,
> We've done some work here so the mesos-ui does not require a nodejs server 
> anymore
> and makes it easy to replace the default UI.
>
> It provides deeper and easier integration with mesos and zookeeper via the 
> Mesos core "MESOS_WEBUI_DIR" flag for replacing the default UI.
> Any dependency with external static files it's been removed too.
>
> It can be tested by builiding the static files via "gulp build" and running 
> mesos with '--webui_dir=/path-to/mesos-ui/build'
> see Readme -> https://github.com/Capgemini/mesos-ui
>
> or just by running the docker image which is just a mesos-master pointing to 
> this UI by default.
>
> docker run --net=host -e MESOS_LOG_DIR=/logs capgemini/mesos-ui:latest
>
> see -> https://hub.docker.com/r/capgemini/mesos-ui/
>
> We hope this makes easy to integrate and run it and looking forward to get 
> some feedback!
>
> Thanks,
> Alberto.
>
>
>
> From: graham.x.tay...@capgemini.com
> To: user@mesos.apache.org
> Subject: Re: mesos-ui
> Date: Fri, 9 Oct 2015 12:16:38 +0200
>
>
> 
>
> Capgemini is a trading name used by the Capgemini Group of companies which 
> includes Capgemini UK plc, a company registered in England and Wales (number 
> 943935) whose registered office is at No. 1, Forge End, Woking, Surrey, GU21 
> 6DB.
> This message contains information that may be privileged or confidential and 
> is the property of the Capgemini Group. It is intended only for the person to 
> whom it is addressed. If you are not the intended recipient, you are not 
> authorized to read, print, retain, copy, disseminate, distribute, or use this 
> message or any part thereof. If you receive this message in error, please 
> notify the sender immediately and delete all copies of this message.
>
>
>
> --Forwarded Message Attachment--
> From: graham.x.tay...@capgemini.com
> To: user@mesos.apache.org
> Subject: Re: mesos-ui
> Date: Fri, 9 Oct 2015 10:16:38 +
>
> Hey Ben,
> Thanks for the response. At the moment we run a nodejs server, but it
> doesn’t do too much - it serves some static files (for the UI) and has some
> middleware for proxying requests to the backend Mesos API endpoints, so it
> might not be too much work to shift the majority of that logic right into
> the front-end. We’ve raised
> https://github.com/Capgemini/mesos-ui/issues/44 to take a look at what
> might be involved.
>
> Thanks,
> Graham.
>
>
> On 8 Oct 2015, at 20:47, Benjamin Mahler 
> wrote:
>
> Thanks for sharing Graham.
>
> Does this webui require a server to be running? Or does it run entirely in
> the browser and reach out to existing master/agent endpoints?
>
> In the latter case, this is also how the built-in webui works, and so it
> would be really easy for us to add a flag on the master that allows one to
> change which webui assets the master serves up. This would make it easy to
> drop in your webui as a replacement.
>
> Ben
>
> On Thu, Oct 8, 2015 at 9:55 AM, haosdent  wrote:
>
> very beautiful!
>
> On Fri, Oct 9, 2015 at 12:41 AM, Brandon Gulla 
> wrote:
>
> Look's great! Can't wait to test it out. Thanks.
>
>
> On Thu, Oct 8, 2015 at 12:03 PM, Taylor, Graham <
> graham.x.tay...@capgemini.com> wrote:
>
>
> 
>
> Capgemini is a trading name used by the Capgemini Group of companies which
> includes Capgemini UK plc, a company registered in England and Wales
> (number 943935) whose registered office is at No. 1, Forge End, Woking,
> Surrey, GU21 6DB.
> This message contains information that may be privileged or confidential
> and is the property of the Capgemini Group. It is intended only for the
> person to whom it is addressed. If you are not the intended recipient, you
> are not authorized to read, print, retain, copy, disseminate, distribute,
> or use this message or any part thereof. If you receive this message in
> error, please notify the sender immediately and delete all copies of this
> message.
>
>
> -- Forwarded message --
> From: "Taylor, Graham" 
> To: "user@mesos.apache.org" 
> Cc:
> Date: Thu, 8 Oct 2015 16:03:00 +
> Subject: mesos-ui
> Hey folks,
> We open-sourced https://github.com/Capgemini/mesos-ui yesterday which
> might be of interest to some peeps on this mailing list.
> If you’re at Mesoscon EU and want to talk about it you can get in touch
> with one of our engineers (Cam Parry https://twitter.com/Wallies9) who is
> also at the con and will be attending the Hackathon on Friday.
>
> Thanks!
> Graham.
>
>
>
>
>
> --
> Brandon
>
>
>
>
> --
> Best Regards,
> Haosdent Huang
>
>
>

Re: Recover docker containers when mesos-slave is contained in docker as well

[haosdent]

And if your slave id change after restart, the slave would be treated as a
new slave and could not recover successfully. It would recover the docker
containers which start with "mesos-${slave_id}".

On Thu, Oct 22, 2015 at 7:08 PM, haosdent  wrote:

> As far as I know, when mesos-slave run in docker, the recover process is
> same as the slave run outside. But you need make sure you add "-v
> /var/run/docker.sock:/var/run/docker.sock" when you launch mesos-slave.
>
> On Thu, Oct 22, 2015 at 6:59 PM, Grzegorz Graczyk 
> wrote:
>
>> Docker is running when slave exits - and so are docker containers started
>> by mesos slave. The problem starts when slave is online again and cannot
>> see already started containers and recover them...
>> Isn't this supposed to fix that problem?
>> https://issues.apache.org/jira/browse/MESOS-2115
>>
>> On 22 October 2015 at 12:54, Klaus Ma  wrote:
>>
>>> It seems we can NOT keep docker running but slave exit.
>>>
>>> On Thu, Oct 22, 2015 at 6:18 AM, Grzegorz Graczyk 
>>> wrote:
>>>
 Hi everyone,
 I was wondering if it's possible to recover running docker containers
 after restart of mesos-slave?
 If it is possible - what are the requirements to do so?

 Regards,
 Grzegorz Graczyk

>>>
>>>
>>>
>>> --
>>> Da (Klaus), Ma (马达) | PMP® | Advisory Software Engineer
>>> Platform Symphony/DCOS Development & Support, STG, IBM GCG
>>> +86-10-8245 4084 | mad...@cn.ibm.com | http://www.cguru.net
>>>
>>
>>
>
>
> --
> Best Regards,
> Haosdent Huang
>



-- 
Best Regards,
Haosdent Huang

Re: Recover docker containers when mesos-slave is contained in docker as well

[Timothy Chen]

Hi Grzegorz,

Yes it's possible, but do require some configuration for the slave recover the 
running containers. This is needed to run Mesos on CoreOS as well, so it's made 
possible I believe around 0.24.1 or later.

Basically to have the slave to recover the task containers, the executors that 
watches the tasks need to be launched in containers as well. This is made 
possible with the docker_mesos_image slave flag, where the docker containerizer 
will use this image to launch executors. This should be the same image used to 
launch the slave itself.

Also when launching the slave in a docker container it must have the following 
docker flags:

--pid=host (so all processes can be visible to the slave)
-v /var/run/docker.sock:/var/run/docker.sock (slave can then launch containers 
as peers)
-v /tmp/mesos:/tmp/mesos (we need slave work directory information to persist 
when slave recovers. You can also create a separate dir on the host if you want 
to run multiple slaves)

Tim

> On Oct 22, 2015, at 6:59 PM, Grzegorz Graczyk  wrote:
> 
> Docker is running when slave exits - and so are docker containers started by 
> mesos slave. The problem starts when slave is online again and cannot see 
> already started containers and recover them...
> Isn't this supposed to fix that problem? 
> https://issues.apache.org/jira/browse/MESOS-2115
> 
>> On 22 October 2015 at 12:54, Klaus Ma  wrote:
>> It seems we can NOT keep docker running but slave exit. 
>> 
>>> On Thu, Oct 22, 2015 at 6:18 AM, Grzegorz Graczyk  
>>> wrote:
>>> Hi everyone,
>>> I was wondering if it's possible to recover running docker containers after 
>>> restart of mesos-slave?
>>> If it is possible - what are the requirements to do so?
>>> 
>>> Regards,
>>> Grzegorz Graczyk
>> 
>> 
>> 
>> -- 
>> Da (Klaus), Ma (马达) | PMP® | Advisory Software Engineer 
>> Platform Symphony/DCOS Development & Support, STG, IBM GCG 
>> +86-10-8245 4084 | mad...@cn.ibm.com | http://www.cguru.net
> 

RE: mesos-ui

[Alberto García Lamela]

Hi there,
We've done some work here so the mesos-ui does not require a nodejs server 
anymore
and makes it easy to replace the default UI.

It provides deeper and easier integration with mesos and zookeeper via the 
Mesos core "MESOS_WEBUI_DIR" flag for replacing the default UI.
Any dependency with external static files it's been removed too.

It can be tested by builiding the static files via "gulp build" and running 
mesos with '--webui_dir=/path-to/mesos-ui/build'
see Readme -> https://github.com/Capgemini/mesos-ui

or just by running the docker image which is just a mesos-master pointing to 
this UI by default.
docker run --net=host -e MESOS_LOG_DIR=/logs capgemini/mesos-ui:latest
see -> https://hub.docker.com/r/capgemini/mesos-ui/

We hope this makes easy to integrate and run it and looking forward to get some 
feedback!

Thanks,
Alberto.

From: graham.x.tay...@capgemini.com
To: user@mesos.apache.org
Subject: Re: mesos-ui
Date: Fri, 9 Oct 2015 12:16:38 +0200

 

 
Capgemini is a trading name used by the Capgemini Group of companies which 
includes Capgemini UK plc, a company registered in England and Wales (number 
943935) whose registered office is at No. 1, Forge End, Woking, Surrey, GU21 
6DB.
This message contains information that may be privileged or confidential and is 
the property of the Capgemini Group. It is intended only for the person to whom 
it is addressed. If you are not the intended recipient, you are not authorized 
to read, print, retain, copy, disseminate, distribute, or use this message or 
any part thereof. If you receive this message in error, please notify the 
sender immediately and delete all copies of this message.


--Forwarded Message Attachment--
From: graham.x.tay...@capgemini.com
To: user@mesos.apache.org
Subject: Re: mesos-ui
Date: Fri, 9 Oct 2015 10:16:38 +

Hey Ben,Thanks for the response. At the moment we run a nodejs server, but it 
doesn’t do too much - it serves some static files (for the UI) and has some 
middleware for proxying requests to the backend Mesos API endpoints, so it 
might not be too much work to shift the majority of that logic right into the 
front-end. We’ve raised https://github.com/Capgemini/mesos-ui/issues/44 to take 
a look at what might be involved.
Thanks,Graham.

On 8 Oct 2015, at 20:47, Benjamin Mahler  
wrote:Thanks for sharing Graham.
Does this webui require a server to be running? Or does it run entirely in the 
browser and reach out to existing master/agent endpoints?
In the latter case, this is also how the built-in webui works, and so it would 
be really easy for us to add a flag on the master that allows one to change 
which webui assets the master serves up. This would make it easy to drop in 
your webui as a replacement.
Ben
On Thu, Oct 8, 2015 at 9:55 AM, haosdent  wrote:
very beautiful!
On Fri, Oct 9, 2015 at 12:41 AM, Brandon Gulla  wrote:
Look's great! Can't wait to test it out. Thanks.

On Thu, Oct 8, 2015 at 12:03 PM, Taylor, Graham  
wrote:






Capgemini is a trading name used by the Capgemini Group of companies which 
includes Capgemini UK plc, a company registered in England and Wales (number 
943935) whose registered office is at No. 1, Forge End, Woking, Surrey, GU21 
6DB.

This message contains information that may be privileged or confidential and is 
the property of the Capgemini Group. It is intended only for the person to whom 
it is addressed. If you are not the intended recipient, you are not authorized 
to read, print, retain, copy, disseminate, distribute, or use this message or 
any part thereof. If you receive this message in error, please notify the 
sender immediately and delete all copies of this message.



-- Forwarded message --
From: "Taylor, Graham" 
To: "user@mesos.apache.org" 
Cc: 
Date: Thu, 8 Oct 2015 16:03:00 +
Subject: mesos-ui
Hey folks,We open-sourced https://github.com/Capgemini/mesos-ui yesterday which 
might be of interest to some peeps on this mailing list.If you’re at Mesoscon 
EU and want to talk about it you can get in touch with one of our engineers 
(Cam Parry https://twitter.com/Wallies9) who is also at the con and will be 
attending the Hackathon on Friday.
Thanks!Graham.



-- 
Brandon



-- 
Best Regards,
Haosdent Huang




  

Re: spark on mesos with docker issue

[Stavros Kontopoulos]

Thnx Rad, sounds pretty cool :). Elizabeth one note for the jira ticket i
do not run the cluster with zookeeper, i use mesos master in stand alone
mode... i guess it makes no difference right?

On Thu, Oct 22, 2015 at 10:38 PM, Rad Gruchalski 
wrote:

> There are 2 things:
>
>  - Akka remote in 2.3.x does not support advertising hostname / port
> different to what it binds to
>  - All other services: file server, broadcast server, repl class server do
> not support advertising hostnames / ports different than what they bind to
>
> Just to expand on the previous one, we are in the process of contributing
> the following bits:
>
>  - akka-remote bind-hostname and bind-port backport to akka 2.3.x (not the
> typesafe closed support implementation, our own implementation)
>  - spark patches for spark 1.4.0+ which enable running Spark on Mesos in
> Docker Bridge networking
>
> Unfortunately, due to the nature of how my employer operates I can’t share
> the code yet. We are working with our legal team to make these available
> asap.
>
> We do run this stuff in production.
>
> Kind regards,
> Radek Gruchalski
> ra...@gruchalski.com 
> de.linkedin.com/in/radgruchalski/
>
>
> *Confidentiality:*This communication is intended for the above-named
> person and may be confidential and/or legally privileged.
> If it has come to you in error you must take no action based on it, nor
> must you copy or show it to anyone; please delete/destroy and inform the
> sender immediately.
>
> On Thursday, 22 October 2015 at 22:28, Iulian Dragoș wrote:
>
>
>
> On Thu, Oct 22, 2015 at 9:10 PM, Rad Gruchalski 
> wrote:
>
> Stavros,
>
> Spark does not support this. I am currently in the process of submitting
> patches for it however it first has to pass through the legal team at the
> company I work for.
>
>
> What exactly is missing in Spark?
>
>
> Kind regards,
> Radek Gruchalski
> ra...@gruchalski.com 
> de.linkedin.com/in/radgruchalski/
>
>
> *Confidentiality:*This communication is intended for the above-named
> person and may be confidential and/or legally privileged.
> If it has come to you in error you must take no action based on it, nor
> must you copy or show it to anyone; please delete/destroy and inform the
> sender immediately.
>
> On Thursday, 22 October 2015 at 21:08, Stavros Kontopoulos wrote:
>
> Bridge... with the latest mesos library vesion 0.25...
>
> On Thu, Oct 22, 2015 at 9:07 PM, Elizabeth Lingg 
> wrote:
>
> Are you using Bridge or Host Networking?
>
> -Elizabeth
>
>
>
> On Thu, Oct 22, 2015 at 12:02 PM, Stavros Kontopoulos <
> st.kontopou...@gmail.com> wrote:
>
> Hi,
>
> Im using spark on mesos on docker. I have linked my slaves to the master
> and a
> spark repl works fine inside the master container.
>
> If i try to crate the same spark repl form the host i get stuck at the
> point when the framework tries to register to the mesos master (here the
> framework is the spark repl itself).
> I can ping the container from my host and vice versa. So networking its
> not the problem.
> What i noticed form the logs is that mesos does not resolve the correct ip:
>
> Framework failover timeout, removing framework
> b3605c33-f573-4d40-806f-b9b0abee2e32-0012 (Spark shell) at
> scheduler-b6dbe4f8-4760-4cbe-bb5e-a991e35aba1b@127.0.1.1:40186
>
> docker0 interface is on 172.17.x.x and my host is one such ip so i didnt
> expect there to see
> 127.0.1.1. I have tried several things like spark.driver.host,
> SPARK_LOCAL_IP to be set correctly but with no result...
> I suspect this is a mesos problem on docker...
>
> Thnx,
>
> S.
>
>
>
>
>
>
>
> --
>
> --
> Iulian Dragos
>
> --
> Reactive Apps on the JVM
> www.typesafe.com
>
>
>

Re: spark on mesos with docker issue

[Iulian Dragoș]

On Thu, Oct 22, 2015 at 9:10 PM, Rad Gruchalski 
wrote:

> Stavros,
>
> Spark does not support this. I am currently in the process of submitting
> patches for it however it first has to pass through the legal team at the
> company I work for.
>

What exactly is missing in Spark?


> Kind regards,
> Radek Gruchalski
> ra...@gruchalski.com 
> de.linkedin.com/in/radgruchalski/
>
>
> *Confidentiality:*This communication is intended for the above-named
> person and may be confidential and/or legally privileged.
> If it has come to you in error you must take no action based on it, nor
> must you copy or show it to anyone; please delete/destroy and inform the
> sender immediately.
>
> On Thursday, 22 October 2015 at 21:08, Stavros Kontopoulos wrote:
>
> Bridge... with the latest mesos library vesion 0.25...
>
> On Thu, Oct 22, 2015 at 9:07 PM, Elizabeth Lingg 
> wrote:
>
> Are you using Bridge or Host Networking?
>
> -Elizabeth
>
>
>
> On Thu, Oct 22, 2015 at 12:02 PM, Stavros Kontopoulos <
> st.kontopou...@gmail.com> wrote:
>
> Hi,
>
> Im using spark on mesos on docker. I have linked my slaves to the master
> and a
> spark repl works fine inside the master container.
>
> If i try to crate the same spark repl form the host i get stuck at the
> point when the framework tries to register to the mesos master (here the
> framework is the spark repl itself).
> I can ping the container from my host and vice versa. So networking its
> not the problem.
> What i noticed form the logs is that mesos does not resolve the correct ip:
>
> Framework failover timeout, removing framework
> b3605c33-f573-4d40-806f-b9b0abee2e32-0012 (Spark shell) at
> scheduler-b6dbe4f8-4760-4cbe-bb5e-a991e35aba1b@127.0.1.1:40186
>
> docker0 interface is on 172.17.x.x and my host is one such ip so i didnt
> expect there to see
> 127.0.1.1. I have tried several things like spark.driver.host,
> SPARK_LOCAL_IP to be set correctly but with no result...
> I suspect this is a mesos problem on docker...
>
> Thnx,
>
> S.
>
>
>
>
>


-- 

--
Iulian Dragos

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

Re: spark on mesos with docker issue

[Elizabeth Lingg]

Right, this issue is related to the advertising the hostname / port in
libprocess that is reachable by the Mesos Master. It should not be related
to the Mesos Master running in Standalone (non HA) mode and not using
Zookeeper.

-Elizabeth

On Thu, Oct 22, 2015 at 1:45 PM, Stavros Kontopoulos <
st.kontopou...@gmail.com> wrote:

> Thnx Rad, sounds pretty cool :). Elizabeth one note for the jira ticket i
> do not run the cluster with zookeeper, i use mesos master in stand alone
> mode... i guess it makes no difference right?
>
> On Thu, Oct 22, 2015 at 10:38 PM, Rad Gruchalski 
> wrote:
>
>> There are 2 things:
>>
>>  - Akka remote in 2.3.x does not support advertising hostname / port
>> different to what it binds to
>>  - All other services: file server, broadcast server, repl class server
>> do not support advertising hostnames / ports different than what they bind
>> to
>>
>> Just to expand on the previous one, we are in the process of contributing
>> the following bits:
>>
>>  - akka-remote bind-hostname and bind-port backport to akka 2.3.x (not
>> the typesafe closed support implementation, our own implementation)
>>  - spark patches for spark 1.4.0+ which enable running Spark on Mesos in
>> Docker Bridge networking
>>
>> Unfortunately, due to the nature of how my employer operates I can’t
>> share the code yet. We are working with our legal team to make these
>> available asap.
>>
>> We do run this stuff in production.
>>
>> Kind regards,
>> Radek Gruchalski
>> ra...@gruchalski.com 
>> de.linkedin.com/in/radgruchalski/
>>
>>
>> *Confidentiality:*This communication is intended for the above-named
>> person and may be confidential and/or legally privileged.
>> If it has come to you in error you must take no action based on it, nor
>> must you copy or show it to anyone; please delete/destroy and inform the
>> sender immediately.
>>
>> On Thursday, 22 October 2015 at 22:28, Iulian Dragoș wrote:
>>
>>
>>
>> On Thu, Oct 22, 2015 at 9:10 PM, Rad Gruchalski 
>> wrote:
>>
>> Stavros,
>>
>> Spark does not support this. I am currently in the process of submitting
>> patches for it however it first has to pass through the legal team at the
>> company I work for.
>>
>>
>> What exactly is missing in Spark?
>>
>>
>> Kind regards,
>> Radek Gruchalski
>> ra...@gruchalski.com 
>> de.linkedin.com/in/radgruchalski/
>>
>>
>> *Confidentiality:*This communication is intended for the above-named
>> person and may be confidential and/or legally privileged.
>> If it has come to you in error you must take no action based on it, nor
>> must you copy or show it to anyone; please delete/destroy and inform the
>> sender immediately.
>>
>> On Thursday, 22 October 2015 at 21:08, Stavros Kontopoulos wrote:
>>
>> Bridge... with the latest mesos library vesion 0.25...
>>
>> On Thu, Oct 22, 2015 at 9:07 PM, Elizabeth Lingg > > wrote:
>>
>> Are you using Bridge or Host Networking?
>>
>> -Elizabeth
>>
>>
>>
>> On Thu, Oct 22, 2015 at 12:02 PM, Stavros Kontopoulos <
>> st.kontopou...@gmail.com> wrote:
>>
>> Hi,
>>
>> Im using spark on mesos on docker. I have linked my slaves to the master
>> and a
>> spark repl works fine inside the master container.
>>
>> If i try to crate the same spark repl form the host i get stuck at the
>> point when the framework tries to register to the mesos master (here the
>> framework is the spark repl itself).
>> I can ping the container from my host and vice versa. So networking its
>> not the problem.
>> What i noticed form the logs is that mesos does not resolve the correct
>> ip:
>>
>> Framework failover timeout, removing framework
>> b3605c33-f573-4d40-806f-b9b0abee2e32-0012 (Spark shell) at
>> scheduler-b6dbe4f8-4760-4cbe-bb5e-a991e35aba1b@127.0.1.1:40186
>>
>> docker0 interface is on 172.17.x.x and my host is one such ip so i didnt
>> expect there to see
>> 127.0.1.1. I have tried several things like spark.driver.host,
>> SPARK_LOCAL_IP to be set correctly but with no result...
>> I suspect this is a mesos problem on docker...
>>
>> Thnx,
>>
>> S.
>>
>>
>>
>>
>>
>>
>>
>> --
>>
>> --
>> Iulian Dragos
>>
>> --
>> Reactive Apps on the JVM
>> www.typesafe.com
>>
>>
>>
>

Re: spark on mesos with docker issue

[Rad Gruchalski]

There are 2 things:  

 - Akka remote in 2.3.x does not support advertising hostname / port different 
to what it binds to
 - All other services: file server, broadcast server, repl class server do not 
support advertising hostnames / ports different than what they bind to

Just to expand on the previous one, we are in the process of contributing the 
following bits:  

 - akka-remote bind-hostname and bind-port backport to akka 2.3.x (not the 
typesafe closed support implementation, our own implementation)
 - spark patches for spark 1.4.0+ which enable running Spark on Mesos in Docker 
Bridge networking

Unfortunately, due to the nature of how my employer operates I can’t share the 
code yet. We are working with our legal team to make these available asap.

We do run this stuff in production.  










Kind regards,

Radek Gruchalski

ra...@gruchalski.com (mailto:ra...@gruchalski.com)
 
(mailto:ra...@gruchalski.com)
de.linkedin.com/in/radgruchalski/ (http://de.linkedin.com/in/radgruchalski/)

Confidentiality:
This communication is intended for the above-named person and may be 
confidential and/or legally privileged.
If it has come to you in error you must take no action based on it, nor must 
you copy or show it to anyone; please delete/destroy and inform the sender 
immediately.



On Thursday, 22 October 2015 at 22:28, Iulian Dragoș wrote:

>  
>  
> On Thu, Oct 22, 2015 at 9:10 PM, Rad Gruchalski  (mailto:ra...@gruchalski.com)> wrote:
> > Stavros,  
> >  
> > Spark does not support this. I am currently in the process of submitting 
> > patches for it however it first has to pass through the legal team at the 
> > company I work for.
>  
> What exactly is missing in Spark?
>   
> >  
> >  
> >  
> >  
> >  
> >  
> >  
> >  
> >  
> >  
> > Kind regards,

> > Radek Gruchalski
> > 
ra...@gruchalski.com (mailto:ra...@gruchalski.com)
 
> > (mailto:ra...@gruchalski.com)
> > de.linkedin.com/in/radgruchalski/ (http://de.linkedin.com/in/radgruchalski/)
> >  
> > Confidentiality:
> > This communication is intended for the above-named person and may be 
> > confidential and/or legally privileged.
> > If it has come to you in error you must take no action based on it, nor 
> > must you copy or show it to anyone; please delete/destroy and inform the 
> > sender immediately.
> >  
> >  
> >  
> > On Thursday, 22 October 2015 at 21:08, Stavros Kontopoulos wrote:
> >  
> > > Bridge... with the latest mesos library vesion 0.25...
> > >  
> > > On Thu, Oct 22, 2015 at 9:07 PM, Elizabeth Lingg  > > (mailto:elizab...@mesosphere.io)> wrote:
> > > > Are you using Bridge or Host Networking?
> > > >  
> > > > -Elizabeth
> > > >  
> > > >  
> > > >  
> > > > On Thu, Oct 22, 2015 at 12:02 PM, Stavros Kontopoulos 
> > > >  wrote:
> > > > > Hi,
> > > > >  
> > > > > Im using spark on mesos on docker. I have linked my slaves to the 
> > > > > master and a
> > > > > spark repl works fine inside the master container.
> > > > >  
> > > > > If i try to crate the same spark repl form the host i get stuck at 
> > > > > the point when the framework tries to register to the mesos master 
> > > > > (here the framework is the spark repl itself).
> > > > > I can ping the container from my host and vice versa. So networking 
> > > > > its not the problem.
> > > > > What i noticed form the logs is that mesos does not resolve the 
> > > > > correct ip:
> > > > >  
> > > > > Framework failover timeout, removing framework 
> > > > > b3605c33-f573-4d40-806f-b9b0abee2e32-0012 (Spark shell) at 
> > > > > scheduler-b6dbe4f8-4760-4cbe-bb5e-a991e35aba1b@127.0.1.1:40186 
> > > > > (http://scheduler-b6dbe4f8-4760-4cbe-bb5e-a991e35aba1b@127.0.1.1:40186)
> > > > >  
> > > > > docker0 interface is on 172.17.x.x and my host is one such ip so i 
> > > > > didnt expect there to see  
> > > > > 127.0.1.1. I have tried several things like spark.driver.host, 
> > > > > SPARK_LOCAL_IP to be set correctly but with no result...
> > > > > I suspect this is a mesos problem on docker...
> > > > >  
> > > > > Thnx,
> > > > >  
> > > > > S.
> > > >  
> > >  
> >  
>  
>  
>  
> --  
>  
> --
> Iulian Dragos
>  
> --
> Reactive Apps on the JVM
> www.typesafe.com (http://www.typesafe.com)
>  

Notes from off-line sync on processes for Mesos road maps

[Niklas Nielsen]

Hi everyone,

Yesterday, a few of us met up to discuss processes for Mesos road maps: how
to capture core values of the project, how to prioritize features from
users running different workloads and how to delegate specifics of road
mapping to the working groups.

The meeting notes are available here:
https://docs.google.com/document/d/11LLDKQBZObduNUG7_bayBIU6VMKq2E8CvmaI0tt1U4M/edit#

Ben H will follow up with a write up with a proposal for a vision statement
and I will create working groups for project strategy for those of you, who
are interested in process and the high-level view and interactions between
working groups (similar to yesterdays meeting).

Let us know if you have any questions.

Cheers,
Niklas

Recover docker containers when mesos-slave is contained in docker as well

[Grzegorz Graczyk]

Hi everyone,
I was wondering if it's possible to recover running docker containers after
restart of mesos-slave?
If it is possible - what are the requirements to do so?

Regards,
Grzegorz Graczyk

Re: mesos-slave service does not work

[Brandon Gulla]

if that doesn't seem to work, an old favorite of mine is:

start mesos-slave && ps -ef | grep mesos-slave

assuming the ps can fire off before the mesos-slave exits, everyone loves a
good race condition.



On Thu, Oct 22, 2015 at 9:44 PM, Klaus Ma  wrote:

> For the bash, I think you can use "bash -x" to see what command are
> executing, and check the environment variables of the shell.
>
> On Thu, Oct 22, 2015 at 12:07 PM, Ahmet Emre Aladağ 
> wrote:
>
>> Hi,
>>
>> I performed the mesosphere 0.25 installation with 3 nodes
>> (master/zookeeper/slaves).
>>
>> "sudo start mesos-master" works but when I type: "sudo start mesos-slave",
>>
>> "/usr/bin/mesos-init-wrapper slave" is called but it does not spawn a
>> slave process. It just finishes. I had to change the
>> /etc/init/mesos-slave.conf to
>>
>> exec mesos-slave --containerizers=docker,mesos --master="zk://.../mesos"
>>  --hostname="" --ip="" --log_dir="/var/log/mesos"
>>
>> to make it work. This way it works but it can't read the conf files under
>> /etc/mesos-slave. I think the wrapper's main function is to read those
>> confs.
>>
>> What could be the reason for this? How can I debug it? Bash code seems
>> complicated.
>>
>>
>>
>
>
> --
> Da (Klaus), Ma (马达) | PMP® | Advisory Software Engineer
> Platform Symphony/DCOS Development & Support, STG, IBM GCG
> +86-10-8245 4084 | mad...@cn.ibm.com | http://www.cguru.net
>



-- 
Brandon

mesos-slave service does not work

[Ahmet Emre Aladağ]

Hi,

I performed the mesosphere 0.25 installation with 3 nodes
(master/zookeeper/slaves).

"sudo start mesos-master" works but when I type: "sudo start mesos-slave",

"/usr/bin/mesos-init-wrapper slave" is called but it does not spawn a slave
process. It just finishes. I had to change the /etc/init/mesos-slave.conf to

exec mesos-slave --containerizers=docker,mesos --master="zk://.../mesos"
 --hostname="" --ip="" --log_dir="/var/log/mesos"

to make it work. This way it works but it can't read the conf files under
/etc/mesos-slave. I think the wrapper's main function is to read those
confs.

What could be the reason for this? How can I debug it? Bash code seems
complicated.