Re: Review Request 30710: add mesos role feature
On Feb. 17, 2015, 11:09 p.m., Maxim Khutornenko wrote:
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java,
line 114
https://reviews.apache.org/r/30710/diff/3/?file=861559#file861559line114
This is breaking Resources/Offer encapsulation and is not addressing
the long term possiblility of making role-biased scheduling decisions.
A possible solution could be de-anonymize `Resources` by adding a
`StringOptional role` to it. The `role` would be left
`Optional.Stringabsent()` any time an anonymous `Resources` instance is
required or populated to the respective framework role when a role-specific
resources are initilized. This would obviously require refactoring
`Resources` to support internal mapping of resource vectors by role.
The above would also eliminate the need for wrapper classes and will
keep the door open for future role-based scheduling optimizations.
I am curious what others think here?
May I have your skype No.? let's discuss it
- lozh...@ebay.com
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/#review72793
---
On Feb. 13, 2015, 10 a.m., lozh...@ebay.com zhang wrote:
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/
---
(Updated Feb. 13, 2015, 10 a.m.)
Review request for Aurora, Joshua Cohen, Maxim Khutornenko, and Bill Farner.
Bugs: AURORA-1109
https://issues.apache.org/jira/browse/AURORA-1109
Repository: aurora
Description
---
## Problems
We are from eBay platform team. Previously, we used marathon to generate
Jenkins master instance in dedicated vms and recieve resource offer from same
dedicated vms. For the details, please refer to
http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU
Now, we found Aurora is more stable and powerful. We are moving from Marathon
to Aurora. During the move, we found there is no mesos role in Aurora now.
But we need use mesos role way to solve the problem in section Frameworks
stopped receiving offers after a while of the given url.
Here is a snippet of the problem description:
*We noticed occurred after we used Marathon to create the initial set of CI
masters. As those CI masters started registering themselves as frameworks,
Marathon stopped receiving any offers from Mesos; essentially, no new CI
masters could be launched. Let’s start with Marathon. In the DRF model, it
was unfair to treat Marathon in the same bucket/role alongside hundreds of
connected Jenkins frameworks. After launching all these Jenkins frameworks,
Marathon had a large resource share and Mesos would aggressively offer
resources to frameworks that were using little or no resources. Marathon was
placed last in priority and got starved out.*
*We decided to define a dedicated Mesos role for Marathon and to have all of
the Mesos slaves that were reserved for Jenkins master instances support that
Mesos role. Jenkins frameworks were left with the default role “*”.*This
solved the problem – Mesos offered resources per role and hence Marathon
never got starved out. A framework with a special role will get resource
offers from both slaves supporting that special role and also from the
default role “*”.* However, since we were using placement constraints,
Marathon accepted resource offers only from slaves that supported both the
role and the placement constraints.*
## Solution
So we add role feature is the source code to solve the problem in same way:
When accept a resource offer, Aurora will send back the needed resources to
Mesos with the mesos role in resource offer.
How to configure the Mesos role:
1.Add cmd option --mesos_role=${Mesos role name} when start Aurora scheduler.
We change the test cases according code change. Each changed test case is
green
Merge https://github.com/zhanglong2015/incubator-aurora
Diffs
-
src/main/java/org/apache/aurora/scheduler/ResourceSlot.java
1a158b4e0be94762ad0480e8ce74b19bacc90c97
src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java
31aa2bbaab3d97875493ad75c4d2c7c82ac7fa58
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java
b5a3140e3560f790d1db496dca3c2ee0dc96a195
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java
d0994203b5650f44ca2eb32e1e2aa61875163854
src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java
5340d651b298ec8aa079e73d6d2f652fdf876293
src/main/java/org/apache/aurora/scheduler/mesos/OfferWrapper.java
PRE-CREATION
Re: Review Request 30710: add mesos role feature
On Feb. 17, 2015, 11:09 p.m., Maxim Khutornenko wrote: src/main/java/org/apache/aurora/scheduler/configuration/Resources.java, line 114 https://reviews.apache.org/r/30710/diff/3/?file=861559#file861559line114 This is breaking Resources/Offer encapsulation and is not addressing the long term possiblility of making role-biased scheduling decisions. A possible solution could be de-anonymize `Resources` by adding a `StringOptional role` to it. The `role` would be left `Optional.Stringabsent()` any time an anonymous `Resources` instance is required or populated to the respective framework role when a role-specific resources are initilized. This would obviously require refactoring `Resources` to support internal mapping of resource vectors by role. The above would also eliminate the need for wrapper classes and will keep the door open for future role-based scheduling optimizations. I am curious what others think here? lozh...@ebay.com zhang wrote: May I have your skype No.? let's discuss it The wrapper classes is used to track which resources have bean allocated. Do you mean tracking the resouce allocation in Resources instead? if so, there are some issues: 1.Noramally each task will has two Resources instance: one for task itself, one for executor. So is unable to track the resource allcation in Resources instance, unless we use thread local variable or something like this. I think another solution is: 1.keep Resources/Offer unchanged, so the Resources/Offer encapsulation is not broken. 2.add a new class ResourceAllocationTracker which is used to track the resource allocation, this class's constructor should accept a Offer as argument. 3.pass the instance of ResourceAllocationTracker to Resources.toResourceList as new parameter 4.in Resources.toResourceList, evertytime a resource is allocated, will be saved in ResourceAllocationTracker instance. - lozh...@ebay.com --- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/30710/#review72793 --- On Feb. 13, 2015, 10 a.m., lozh...@ebay.com zhang wrote: --- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/30710/ --- (Updated Feb. 13, 2015, 10 a.m.) Review request for Aurora, Joshua Cohen, Maxim Khutornenko, and Bill Farner. Bugs: AURORA-1109 https://issues.apache.org/jira/browse/AURORA-1109 Repository: aurora Description --- ## Problems We are from eBay platform team. Previously, we used marathon to generate Jenkins master instance in dedicated vms and recieve resource offer from same dedicated vms. For the details, please refer to http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU Now, we found Aurora is more stable and powerful. We are moving from Marathon to Aurora. During the move, we found there is no mesos role in Aurora now. But we need use mesos role way to solve the problem in section Frameworks stopped receiving offers after a while of the given url. Here is a snippet of the problem description: *We noticed occurred after we used Marathon to create the initial set of CI masters. As those CI masters started registering themselves as frameworks, Marathon stopped receiving any offers from Mesos; essentially, no new CI masters could be launched. Let’s start with Marathon. In the DRF model, it was unfair to treat Marathon in the same bucket/role alongside hundreds of connected Jenkins frameworks. After launching all these Jenkins frameworks, Marathon had a large resource share and Mesos would aggressively offer resources to frameworks that were using little or no resources. Marathon was placed last in priority and got starved out.* *We decided to define a dedicated Mesos role for Marathon and to have all of the Mesos slaves that were reserved for Jenkins master instances support that Mesos role. Jenkins frameworks were left with the default role “*”.*This solved the problem – Mesos offered resources per role and hence Marathon never got starved out. A framework with a special role will get resource offers from both slaves supporting that special role and also from the default role “*”.* However, since we were using placement constraints, Marathon accepted resource offers only from slaves that supported both the role and the placement constraints.* ## Solution So we add role feature is the source code to solve the problem in same way: When accept a resource offer, Aurora will send back the needed resources to Mesos with the mesos role in resource offer. How to configure the Mesos role: 1.Add cmd
Re: Review Request 30710: add mesos role feature
On Feb. 17, 2015, 11:09 p.m., Maxim Khutornenko wrote:
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java,
line 174
https://reviews.apache.org/r/30710/diff/3/?file=861559#file861559line174
Prefer using guava `Ordering` [1] instead. It will avoid excessive
collection copies.
[1] - https://code.google.com/p/guava-libraries/wiki/OrderingExplained
Will change
On Feb. 17, 2015, 11:09 p.m., Maxim Khutornenko wrote:
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java,
line 139
https://reviews.apache.org/r/30710/diff/3/?file=861559#file861559line139
PORTS should be treated the same as other Mesos resources as far as
framework role goes. We need to set the framework role on a matching port
`Resource`.
Will Change
- lozh...@ebay.com
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/#review72793
---
On Feb. 13, 2015, 10 a.m., lozh...@ebay.com zhang wrote:
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/
---
(Updated Feb. 13, 2015, 10 a.m.)
Review request for Aurora, Joshua Cohen, Maxim Khutornenko, and Bill Farner.
Bugs: AURORA-1109
https://issues.apache.org/jira/browse/AURORA-1109
Repository: aurora
Description
---
## Problems
We are from eBay platform team. Previously, we used marathon to generate
Jenkins master instance in dedicated vms and recieve resource offer from same
dedicated vms. For the details, please refer to
http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU
Now, we found Aurora is more stable and powerful. We are moving from Marathon
to Aurora. During the move, we found there is no mesos role in Aurora now.
But we need use mesos role way to solve the problem in section Frameworks
stopped receiving offers after a while of the given url.
Here is a snippet of the problem description:
*We noticed occurred after we used Marathon to create the initial set of CI
masters. As those CI masters started registering themselves as frameworks,
Marathon stopped receiving any offers from Mesos; essentially, no new CI
masters could be launched. Let’s start with Marathon. In the DRF model, it
was unfair to treat Marathon in the same bucket/role alongside hundreds of
connected Jenkins frameworks. After launching all these Jenkins frameworks,
Marathon had a large resource share and Mesos would aggressively offer
resources to frameworks that were using little or no resources. Marathon was
placed last in priority and got starved out.*
*We decided to define a dedicated Mesos role for Marathon and to have all of
the Mesos slaves that were reserved for Jenkins master instances support that
Mesos role. Jenkins frameworks were left with the default role “*”.*This
solved the problem – Mesos offered resources per role and hence Marathon
never got starved out. A framework with a special role will get resource
offers from both slaves supporting that special role and also from the
default role “*”.* However, since we were using placement constraints,
Marathon accepted resource offers only from slaves that supported both the
role and the placement constraints.*
## Solution
So we add role feature is the source code to solve the problem in same way:
When accept a resource offer, Aurora will send back the needed resources to
Mesos with the mesos role in resource offer.
How to configure the Mesos role:
1.Add cmd option --mesos_role=${Mesos role name} when start Aurora scheduler.
We change the test cases according code change. Each changed test case is
green
Merge https://github.com/zhanglong2015/incubator-aurora
Diffs
-
src/main/java/org/apache/aurora/scheduler/ResourceSlot.java
1a158b4e0be94762ad0480e8ce74b19bacc90c97
src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java
31aa2bbaab3d97875493ad75c4d2c7c82ac7fa58
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java
b5a3140e3560f790d1db496dca3c2ee0dc96a195
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java
d0994203b5650f44ca2eb32e1e2aa61875163854
src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java
5340d651b298ec8aa079e73d6d2f652fdf876293
src/main/java/org/apache/aurora/scheduler/mesos/OfferWrapper.java
PRE-CREATION
src/main/java/org/apache/aurora/scheduler/mesos/ResourceWrapper.java
PRE-CREATION
src/main/java/org/apache/aurora/scheduler/state/TaskAssigner.java
e1c29747c9854cf75bf63f6f085cf40ca68989af
Re: Review Request 30710: add mesos role feature
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/#review72113
---
Ship it!
Master (61e6c35) is green with this patch.
./build-support/jenkins/build.sh
I will refresh this build result if you post a review containing @ReviewBot
retry
- Aurora ReviewBot
On Feb. 12, 2015, 6:12 a.m., lozh...@ebay.com zhang wrote:
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/
---
(Updated Feb. 12, 2015, 6:12 a.m.)
Review request for Aurora, Joshua Cohen and Bill Farner.
Bugs: AURORA-1109
https://issues.apache.org/jira/browse/AURORA-1109
Repository: aurora
Description
---
## Problems
We are from eBay platform team. Previously, we used marathon to generate
Jenkins master instance in dedicated vms and recieve resource offer from same
dedicated vms. For the details, please refer to
http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU
Now, we found Aurora is more stable and powerful. We are moving from Marathon
to Aurora. During the move, we found there is no mesos role in Aurora now.
But we need use mesos role way to solve the problem in section Frameworks
stopped receiving offers after a while of the given url.
Here is a snippet of the problem description:
*We noticed occurred after we used Marathon to create the initial set of CI
masters. As those CI masters started registering themselves as frameworks,
Marathon stopped receiving any offers from Mesos; essentially, no new CI
masters could be launched. Let’s start with Marathon. In the DRF model, it
was unfair to treat Marathon in the same bucket/role alongside hundreds of
connected Jenkins frameworks. After launching all these Jenkins frameworks,
Marathon had a large resource share and Mesos would aggressively offer
resources to frameworks that were using little or no resources. Marathon was
placed last in priority and got starved out.*
*We decided to define a dedicated Mesos role for Marathon and to have all of
the Mesos slaves that were reserved for Jenkins master instances support that
Mesos role. Jenkins frameworks were left with the default role “*”.*This
solved the problem – Mesos offered resources per role and hence Marathon
never got starved out. A framework with a special role will get resource
offers from both slaves supporting that special role and also from the
default role “*”.* However, since we were using placement constraints,
Marathon accepted resource offers only from slaves that supported both the
role and the placement constraints.*
## Solution
So we add role feature is the source code to solve the problem in same way:
When accept a resource offer, Aurora will send back the needed resources to
Mesos with the mesos role in resource offer.
How to configure the Mesos role:
1.Add cmd option --mesos_role=${Mesos role name} when start Aurora scheduler.
We change the test cases according code change. Each changed test case is
green
Merge https://github.com/zhanglong2015/incubator-aurora
Diffs
-
src/main/java/org/apache/aurora/scheduler/ResourceSlot.java
1a158b4e0be94762ad0480e8ce74b19bacc90c97
src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java
31aa2bbaab3d97875493ad75c4d2c7c82ac7fa58
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java
b5a3140e3560f790d1db496dca3c2ee0dc96a195
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java
d0994203b5650f44ca2eb32e1e2aa61875163854
src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java
5340d651b298ec8aa079e73d6d2f652fdf876293
src/main/java/org/apache/aurora/scheduler/mesos/OfferWrapper.java
PRE-CREATION
src/main/java/org/apache/aurora/scheduler/mesos/ResourceWrapper.java
PRE-CREATION
src/main/java/org/apache/aurora/scheduler/state/TaskAssigner.java
e1c29747c9854cf75bf63f6f085cf40ca68989af
src/test/java/org/apache/aurora/scheduler/async/GcExecutorLauncherTest.java
422d5a9a42310979752eb7282658316c2b772419
src/test/java/org/apache/aurora/scheduler/configuration/ResourcesTest.java
d6febb8998e05257cabe8d193cefa0b6c79f197e
src/test/java/org/apache/aurora/scheduler/mesos/MesosTaskFactoryImplTest.java
5f08d00d39f016af9bc296e517ad49b66ab5a8de
src/test/java/org/apache/aurora/scheduler/state/TaskAssignerImplTest.java
411a55a8d85f60bb2703468f2d69b64b2736eee4
Diff: https://reviews.apache.org/r/30710/diff/
Testing
---
:buildSrc:compileJava UP-TO-DATE
:buildSrc:compileGroovy UP-TO-DATE
:buildSrc:processResources UP-TO-DATE
:buildSrc:classes UP-TO-DATE
Re: Review Request 30710: add mesos role feature
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/
---
(Updated Feb. 12, 2015, 6:12 a.m.)
Review request for Aurora, Joshua Cohen and Bill Farner.
Changes
---
1.Left the role unset in register framework message if no role specified in
command line
2.For below case, will use the resource with role aurora first. If the resource
with role aurora is not enough for task, then use resource with role *
together. For example, a task need 5 CPU, then willl use cpus(aurora):4 and
cpus(*):1.
I am afraid the current approach is not going to work in a cluster with
mesos slaves configured to offer multi-role resources. Consider an example
where a slave is configured to offer both role-specific and general pool CPU:
--resources=cpus(aurora):4;cpus(*):2
Bugs: AURORA-1109
https://issues.apache.org/jira/browse/AURORA-1109
Repository: aurora
Description
---
## Problems
We are from eBay platform team. Previously, we used marathon to generate
Jenkins master instance in dedicated vms and recieve resource offer from same
dedicated vms. For the details, please refer to
http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU
Now, we found Aurora is more stable and powerful. We are moving from Marathon
to Aurora. During the move, we found there is no mesos role in Aurora now. But
we need use mesos role way to solve the problem in section Frameworks stopped
receiving offers after a while of the given url.
Here is a snippet of the problem description:
*We noticed occurred after we used Marathon to create the initial set of CI
masters. As those CI masters started registering themselves as frameworks,
Marathon stopped receiving any offers from Mesos; essentially, no new CI
masters could be launched. Let’s start with Marathon. In the DRF model, it was
unfair to treat Marathon in the same bucket/role alongside hundreds of
connected Jenkins frameworks. After launching all these Jenkins frameworks,
Marathon had a large resource share and Mesos would aggressively offer
resources to frameworks that were using little or no resources. Marathon was
placed last in priority and got starved out.*
*We decided to define a dedicated Mesos role for Marathon and to have all of
the Mesos slaves that were reserved for Jenkins master instances support that
Mesos role. Jenkins frameworks were left with the default role “*”.*This solved
the problem – Mesos offered resources per role and hence Marathon never got
starved out. A framework with a special role will get resource offers from both
slaves supporting that special role and also from the default role “*”.*
However, since we were using placement constraints, Marathon accepted resource
offers only from slaves that supported both the role and the placement
constraints.*
## Solution
So we add role feature is the source code to solve the problem in same way:
When accept a resource offer, Aurora will send back the needed resources to
Mesos with the mesos role in resource offer.
How to configure the Mesos role:
1.Add cmd option --mesos_role=${Mesos role name} when start Aurora scheduler.
We change the test cases according code change. Each changed test case is green
Merge https://github.com/zhanglong2015/incubator-aurora
Diffs (updated)
-
src/main/java/org/apache/aurora/scheduler/ResourceSlot.java
1a158b4e0be94762ad0480e8ce74b19bacc90c97
src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java
31aa2bbaab3d97875493ad75c4d2c7c82ac7fa58
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java
b5a3140e3560f790d1db496dca3c2ee0dc96a195
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java
d0994203b5650f44ca2eb32e1e2aa61875163854
src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java
5340d651b298ec8aa079e73d6d2f652fdf876293
src/main/java/org/apache/aurora/scheduler/mesos/OfferWrapper.java
PRE-CREATION
src/main/java/org/apache/aurora/scheduler/mesos/ResourceWrapper.java
PRE-CREATION
src/main/java/org/apache/aurora/scheduler/state/TaskAssigner.java
e1c29747c9854cf75bf63f6f085cf40ca68989af
src/test/java/org/apache/aurora/scheduler/async/GcExecutorLauncherTest.java
422d5a9a42310979752eb7282658316c2b772419
src/test/java/org/apache/aurora/scheduler/configuration/ResourcesTest.java
d6febb8998e05257cabe8d193cefa0b6c79f197e
src/test/java/org/apache/aurora/scheduler/mesos/MesosTaskFactoryImplTest.java
5f08d00d39f016af9bc296e517ad49b66ab5a8de
src/test/java/org/apache/aurora/scheduler/state/TaskAssignerImplTest.java
411a55a8d85f60bb2703468f2d69b64b2736eee4
Diff: https://reviews.apache.org/r/30710/diff/
Testing
---
:buildSrc:compileJava UP-TO-DATE
:buildSrc:compileGroovy UP-TO-DATE
:buildSrc:processResources
Re: Review Request 30710: add mesos role feature
On Feb. 9, 2015, 4:56 p.m., Zameer Manji wrote:
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java,
line 117
https://reviews.apache.org/r/30710/diff/2/?file=852094#file852094line117
I don't think we can safely do this right now. I know the default value
is '*'
(https://github.com/apache/mesos/blob/master/include/mesos/mesos.proto#L128),
but we have hit bugs in the past where Mesos will reject protobufs as
changed when we start setting a previously unset field to a default value.
See https://issues.apache.org/jira/browse/MESOS-2309 for details.
Currently as this stands, I think this will cause existing Aurora
instances to be rejected by the Master.
lozh...@ebay.com zhang wrote:
How about if we left the role unset in register framework message if no
role specified in command line?
I think that's a good step to take. When you make this change can you reference
MESOS-2309 as the reason why in a comment?
- Zameer
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/#review71727
---
On Feb. 9, 2015, 4:49 p.m., lozh...@ebay.com zhang wrote:
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/
---
(Updated Feb. 9, 2015, 4:49 p.m.)
Review request for Aurora, Joshua Cohen and Bill Farner.
Bugs: AURORA-1109
https://issues.apache.org/jira/browse/AURORA-1109
Repository: aurora
Description
---
## Problems
We are from eBay platform team. Previously, we used marathon to generate
Jenkins master instance in dedicated vms and recieve resource offer from same
dedicated vms. For the details, please refer to
http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU
Now, we found Aurora is more stable and powerful. We are moving from Marathon
to Aurora. During the move, we found there is no mesos role in Aurora now.
But we need use mesos role way to solve the problem in section Frameworks
stopped receiving offers after a while of the given url.
Here is a snippet of the problem description:
*We noticed occurred after we used Marathon to create the initial set of CI
masters. As those CI masters started registering themselves as frameworks,
Marathon stopped receiving any offers from Mesos; essentially, no new CI
masters could be launched. Let’s start with Marathon. In the DRF model, it
was unfair to treat Marathon in the same bucket/role alongside hundreds of
connected Jenkins frameworks. After launching all these Jenkins frameworks,
Marathon had a large resource share and Mesos would aggressively offer
resources to frameworks that were using little or no resources. Marathon was
placed last in priority and got starved out.*
*We decided to define a dedicated Mesos role for Marathon and to have all of
the Mesos slaves that were reserved for Jenkins master instances support that
Mesos role. Jenkins frameworks were left with the default role “*”.*This
solved the problem – Mesos offered resources per role and hence Marathon
never got starved out. A framework with a special role will get resource
offers from both slaves supporting that special role and also from the
default role “*”.* However, since we were using placement constraints,
Marathon accepted resource offers only from slaves that supported both the
role and the placement constraints.*
## Solution
So we add role feature is the source code to solve the problem in same way:
When accept a resource offer, Aurora will send back the needed resources to
Mesos with the mesos role in resource offer.
How to configure the Mesos role:
1.Add cmd option --mesos_role=${Mesos role name} when start Aurora scheduler.
We change the test cases according code change. Each changed test case is
green
Merge https://github.com/zhanglong2015/incubator-aurora
Diffs
-
src/main/java/org/apache/aurora/scheduler/ResourceSlot.java
1a158b4e0be94762ad0480e8ce74b19bacc90c97
src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java
31aa2bbaab3d97875493ad75c4d2c7c82ac7fa58
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java
b5a3140e3560f790d1db496dca3c2ee0dc96a195
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java
d0994203b5650f44ca2eb32e1e2aa61875163854
src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java
5340d651b298ec8aa079e73d6d2f652fdf876293
src/main/java/org/apache/aurora/scheduler/state/TaskAssigner.java
e1c29747c9854cf75bf63f6f085cf40ca68989af
Re: Review Request 30710: add mesos role feature
On Feb. 10, 2015, 12:56 a.m., Zameer Manji wrote:
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java,
line 117
https://reviews.apache.org/r/30710/diff/2/?file=852094#file852094line117
I don't think we can safely do this right now. I know the default value
is '*'
(https://github.com/apache/mesos/blob/master/include/mesos/mesos.proto#L128),
but we have hit bugs in the past where Mesos will reject protobufs as
changed when we start setting a previously unset field to a default value.
See https://issues.apache.org/jira/browse/MESOS-2309 for details.
Currently as this stands, I think this will cause existing Aurora
instances to be rejected by the Master.
How about if we left the role unset in register framework message if no role
specified in command line?
- lozh...@ebay.com
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/#review71727
---
On Feb. 10, 2015, 12:49 a.m., lozh...@ebay.com zhang wrote:
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/
---
(Updated Feb. 10, 2015, 12:49 a.m.)
Review request for Aurora, Joshua Cohen and Bill Farner.
Bugs: AURORA-1109
https://issues.apache.org/jira/browse/AURORA-1109
Repository: aurora
Description
---
## Problems
We are from eBay platform team. Previously, we used marathon to generate
Jenkins master instance in dedicated vms and recieve resource offer from same
dedicated vms. For the details, please refer to
http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU
Now, we found Aurora is more stable and powerful. We are moving from Marathon
to Aurora. During the move, we found there is no mesos role in Aurora now.
But we need use mesos role way to solve the problem in section Frameworks
stopped receiving offers after a while of the given url.
Here is a snippet of the problem description:
*We noticed occurred after we used Marathon to create the initial set of CI
masters. As those CI masters started registering themselves as frameworks,
Marathon stopped receiving any offers from Mesos; essentially, no new CI
masters could be launched. Let’s start with Marathon. In the DRF model, it
was unfair to treat Marathon in the same bucket/role alongside hundreds of
connected Jenkins frameworks. After launching all these Jenkins frameworks,
Marathon had a large resource share and Mesos would aggressively offer
resources to frameworks that were using little or no resources. Marathon was
placed last in priority and got starved out.*
*We decided to define a dedicated Mesos role for Marathon and to have all of
the Mesos slaves that were reserved for Jenkins master instances support that
Mesos role. Jenkins frameworks were left with the default role “*”.*This
solved the problem – Mesos offered resources per role and hence Marathon
never got starved out. A framework with a special role will get resource
offers from both slaves supporting that special role and also from the
default role “*”.* However, since we were using placement constraints,
Marathon accepted resource offers only from slaves that supported both the
role and the placement constraints.*
## Solution
So we add role feature is the source code to solve the problem in same way:
When accept a resource offer, Aurora will send back the needed resources to
Mesos with the mesos role in resource offer.
How to configure the Mesos role:
1.Add cmd option --mesos_role=${Mesos role name} when start Aurora scheduler.
We change the test cases according code change. Each changed test case is
green
Merge https://github.com/zhanglong2015/incubator-aurora
Diffs
-
src/main/java/org/apache/aurora/scheduler/ResourceSlot.java
1a158b4e0be94762ad0480e8ce74b19bacc90c97
src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java
31aa2bbaab3d97875493ad75c4d2c7c82ac7fa58
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java
b5a3140e3560f790d1db496dca3c2ee0dc96a195
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java
d0994203b5650f44ca2eb32e1e2aa61875163854
src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java
5340d651b298ec8aa079e73d6d2f652fdf876293
src/main/java/org/apache/aurora/scheduler/state/TaskAssigner.java
e1c29747c9854cf75bf63f6f085cf40ca68989af
src/test/java/org/apache/aurora/scheduler/async/GcExecutorLauncherTest.java
422d5a9a42310979752eb7282658316c2b772419
Re: Review Request 30710: add mesos role feature
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/#review71754
---
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java
https://reviews.apache.org/r/30710/#comment117619
I am afraid the current approach is not going to work in a cluster with
mesos slaves configured to offer multi-role resources. Consider an example
where a slave is configured to offer both role-specific and general pool CPU:
```
--resources=cpus(aurora):4;cpus(*):2
```
A correspondent offer resources section could be:
```
...
resources :
[
{
name: cpus,
type: SCALAR,
scalar : { value : 4.0 },
role: aurora,
},
{
name: cpus,
type: SCALAR,
scalar : { value : 2.0 },
role: *,
}
]
...
```
Given the above, a matching task with 6.0 CPU would require a TaskInfo with
the same resources section (4.0-aurora, 2.0-*) or mesos would reject the task
launch.
The current state of Aurora where a framework role is not configurable
gives us the luxury of not supporting multi-role allocations. However, if we
are going to support role-based allocations I think we should do it the proper
way and fully support single and multi-role assignments. This is going to be
especially critical when the mesos dynamic reservation [1] lands.
[1] -
https://docs.google.com/document/d/1e3j69pfBgtc8xM00DhcuiMl6ImkEB5na0TzOMyzrg8A/edit#
- Maxim Khutornenko
On Feb. 10, 2015, 12:49 a.m., lozh...@ebay.com zhang wrote:
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/
---
(Updated Feb. 10, 2015, 12:49 a.m.)
Review request for Aurora, Joshua Cohen and Bill Farner.
Bugs: AURORA-1109
https://issues.apache.org/jira/browse/AURORA-1109
Repository: aurora
Description
---
## Problems
We are from eBay platform team. Previously, we used marathon to generate
Jenkins master instance in dedicated vms and recieve resource offer from same
dedicated vms. For the details, please refer to
http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU
Now, we found Aurora is more stable and powerful. We are moving from Marathon
to Aurora. During the move, we found there is no mesos role in Aurora now.
But we need use mesos role way to solve the problem in section Frameworks
stopped receiving offers after a while of the given url.
Here is a snippet of the problem description:
*We noticed occurred after we used Marathon to create the initial set of CI
masters. As those CI masters started registering themselves as frameworks,
Marathon stopped receiving any offers from Mesos; essentially, no new CI
masters could be launched. Let’s start with Marathon. In the DRF model, it
was unfair to treat Marathon in the same bucket/role alongside hundreds of
connected Jenkins frameworks. After launching all these Jenkins frameworks,
Marathon had a large resource share and Mesos would aggressively offer
resources to frameworks that were using little or no resources. Marathon was
placed last in priority and got starved out.*
*We decided to define a dedicated Mesos role for Marathon and to have all of
the Mesos slaves that were reserved for Jenkins master instances support that
Mesos role. Jenkins frameworks were left with the default role “*”.*This
solved the problem – Mesos offered resources per role and hence Marathon
never got starved out. A framework with a special role will get resource
offers from both slaves supporting that special role and also from the
default role “*”.* However, since we were using placement constraints,
Marathon accepted resource offers only from slaves that supported both the
role and the placement constraints.*
## Solution
So we add role feature is the source code to solve the problem in same way:
When accept a resource offer, Aurora will send back the needed resources to
Mesos with the mesos role in resource offer.
How to configure the Mesos role:
1.Add cmd option --mesos_role=${Mesos role name} when start Aurora scheduler.
We change the test cases according code change. Each changed test case is
green
Merge https://github.com/zhanglong2015/incubator-aurora
Diffs
-
src/main/java/org/apache/aurora/scheduler/ResourceSlot.java
1a158b4e0be94762ad0480e8ce74b19bacc90c97
src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java
31aa2bbaab3d97875493ad75c4d2c7c82ac7fa58
Re: Review Request 30710: add mesos role feature
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/#review71727
---
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java
https://reviews.apache.org/r/30710/#comment117596
I don't think we can safely do this right now. I know the default value is
'*'
(https://github.com/apache/mesos/blob/master/include/mesos/mesos.proto#L128),
but we have hit bugs in the past where Mesos will reject protobufs as changed
when we start setting a previously unset field to a default value.
See https://issues.apache.org/jira/browse/MESOS-2309 for details.
Currently as this stands, I think this will cause existing Aurora instances
to be rejected by the Master.
- Zameer Manji
On Feb. 9, 2015, 4:49 p.m., lozh...@ebay.com zhang wrote:
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/
---
(Updated Feb. 9, 2015, 4:49 p.m.)
Review request for Aurora, Joshua Cohen and Bill Farner.
Bugs: AURORA-1109
https://issues.apache.org/jira/browse/AURORA-1109
Repository: aurora
Description
---
## Problems
We are from eBay platform team. Previously, we used marathon to generate
Jenkins master instance in dedicated vms and recieve resource offer from same
dedicated vms. For the details, please refer to
http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU
Now, we found Aurora is more stable and powerful. We are moving from Marathon
to Aurora. During the move, we found there is no mesos role in Aurora now.
But we need use mesos role way to solve the problem in section Frameworks
stopped receiving offers after a while of the given url.
Here is a snippet of the problem description:
*We noticed occurred after we used Marathon to create the initial set of CI
masters. As those CI masters started registering themselves as frameworks,
Marathon stopped receiving any offers from Mesos; essentially, no new CI
masters could be launched. Let’s start with Marathon. In the DRF model, it
was unfair to treat Marathon in the same bucket/role alongside hundreds of
connected Jenkins frameworks. After launching all these Jenkins frameworks,
Marathon had a large resource share and Mesos would aggressively offer
resources to frameworks that were using little or no resources. Marathon was
placed last in priority and got starved out.*
*We decided to define a dedicated Mesos role for Marathon and to have all of
the Mesos slaves that were reserved for Jenkins master instances support that
Mesos role. Jenkins frameworks were left with the default role “*”.*This
solved the problem – Mesos offered resources per role and hence Marathon
never got starved out. A framework with a special role will get resource
offers from both slaves supporting that special role and also from the
default role “*”.* However, since we were using placement constraints,
Marathon accepted resource offers only from slaves that supported both the
role and the placement constraints.*
## Solution
So we add role feature is the source code to solve the problem in same way:
When accept a resource offer, Aurora will send back the needed resources to
Mesos with the mesos role in resource offer.
How to configure the Mesos role:
1.Add cmd option --mesos_role=${Mesos role name} when start Aurora scheduler.
We change the test cases according code change. Each changed test case is
green
Merge https://github.com/zhanglong2015/incubator-aurora
Diffs
-
src/main/java/org/apache/aurora/scheduler/ResourceSlot.java
1a158b4e0be94762ad0480e8ce74b19bacc90c97
src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java
31aa2bbaab3d97875493ad75c4d2c7c82ac7fa58
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java
b5a3140e3560f790d1db496dca3c2ee0dc96a195
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java
d0994203b5650f44ca2eb32e1e2aa61875163854
src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java
5340d651b298ec8aa079e73d6d2f652fdf876293
src/main/java/org/apache/aurora/scheduler/state/TaskAssigner.java
e1c29747c9854cf75bf63f6f085cf40ca68989af
src/test/java/org/apache/aurora/scheduler/async/GcExecutorLauncherTest.java
422d5a9a42310979752eb7282658316c2b772419
src/test/java/org/apache/aurora/scheduler/configuration/ResourcesTest.java
d6febb8998e05257cabe8d193cefa0b6c79f197e
src/test/java/org/apache/aurora/scheduler/mesos/MesosTaskFactoryImplTest.java
5f08d00d39f016af9bc296e517ad49b66ab5a8de
Re: Review Request 30710: add mesos role feature
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/
---
(Updated Feb. 9, 2015, 4:01 a.m.)
Review request for Aurora and Bill Farner.
Changes
---
add link to jira
Bugs: AURORA-1109
https://issues.apache.org/jira/browse/AURORA-1109
Repository: aurora
Description
---
## Problems
We are from eBay platform team. Previously, we used marathon to generate
Jenkins master instance in dedicated vms and recieve resource offer from same
dedicated vms. For the details, please refer to
http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU
Now, we found Aurora is more stable and powerful. We are moving from Marathon
to Aurora. During the move, we found there is no mesos role in Aurora now. But
we need use mesos role way to solve the problem in section Frameworks stopped
receiving offers after a while of the given url.
Here is a snippet of the problem description:
*We noticed occurred after we used Marathon to create the initial set of CI
masters. As those CI masters started registering themselves as frameworks,
Marathon stopped receiving any offers from Mesos; essentially, no new CI
masters could be launched. Let’s start with Marathon. In the DRF model, it was
unfair to treat Marathon in the same bucket/role alongside hundreds of
connected Jenkins frameworks. After launching all these Jenkins frameworks,
Marathon had a large resource share and Mesos would aggressively offer
resources to frameworks that were using little or no resources. Marathon was
placed last in priority and got starved out.*
*We decided to define a dedicated Mesos role for Marathon and to have all of
the Mesos slaves that were reserved for Jenkins master instances support that
Mesos role. Jenkins frameworks were left with the default role “*”.*This solved
the problem – Mesos offered resources per role and hence Marathon never got
starved out. A framework with a special role will get resource offers from both
slaves supporting that special role and also from the default role “*”.*
However, since we were using placement constraints, Marathon accepted resource
offers only from slaves that supported both the role and the placement
constraints.*
## Solution
So we add role feature is the source code to solve the problem in same way:
When accept a resource offer, Aurora will send back the needed resources to
Mesos with the mesos role in resource offer.
How to configure the Mesos role:
1.Add cmd option --mesos_role=${Mesos role name} when start Aurora scheduler.
We change the test cases according code change. Each changed test case is green
Merge https://github.com/zhanglong2015/incubator-aurora
Diffs
-
src/main/java/org/apache/aurora/scheduler/ResourceSlot.java
1a158b4e0be94762ad0480e8ce74b19bacc90c97
src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java
31aa2bbaab3d97875493ad75c4d2c7c82ac7fa58
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java
b5a3140e3560f790d1db496dca3c2ee0dc96a195
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java
d0994203b5650f44ca2eb32e1e2aa61875163854
src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java
5340d651b298ec8aa079e73d6d2f652fdf876293
src/main/java/org/apache/aurora/scheduler/state/TaskAssigner.java
e1c29747c9854cf75bf63f6f085cf40ca68989af
src/test/java/org/apache/aurora/scheduler/async/GcExecutorLauncherTest.java
422d5a9a42310979752eb7282658316c2b772419
src/test/java/org/apache/aurora/scheduler/configuration/ResourcesTest.java
d6febb8998e05257cabe8d193cefa0b6c79f197e
src/test/java/org/apache/aurora/scheduler/mesos/MesosTaskFactoryImplTest.java
5f08d00d39f016af9bc296e517ad49b66ab5a8de
src/test/java/org/apache/aurora/scheduler/state/TaskAssignerImplTest.java
411a55a8d85f60bb2703468f2d69b64b2736eee4
Diff: https://reviews.apache.org/r/30710/diff/
Testing
---
:buildSrc:compileJava UP-TO-DATE
:buildSrc:compileGroovy UP-TO-DATE
:buildSrc:processResources UP-TO-DATE
:buildSrc:classes UP-TO-DATE
:buildSrc:jar UP-TO-DATE
:buildSrc:assemble UP-TO-DATE
:buildSrc:compileTestJava UP-TO-DATE
:buildSrc:compileTestGroovy UP-TO-DATE
:buildSrc:processTestResources UP-TO-DATE
:buildSrc:testClasses UP-TO-DATE
:buildSrc:test UP-TO-DATE
:buildSrc:check UP-TO-DATE
:buildSrc:build UP-TO-DATE
:api:generateThriftJava
:api:classesThrift
Note: Some input files use unchecked or unsafe operations.
Note: Recompile with -Xlint:unchecked for details.
:api:checkPython
:api:generateThriftEntitiesJava
:api:classesThriftEntities
:api:compileJava UP-TO-DATE
:api:generateThriftResources
:api:processResources UP-TO-DATE
:api:classes
:api:jar
:compileJava
Note: Writing
Re: Review Request 30710: add mesos role feature
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/#review71407
---
Ship it!
Master (5059040) is green with this patch.
./build-support/jenkins/build.sh
I will refresh this build result if you post a review containing @ReviewBot
retry
- Aurora ReviewBot
On Feb. 6, 2015, 9:23 a.m., lozh...@ebay.com zhang wrote:
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/
---
(Updated Feb. 6, 2015, 9:23 a.m.)
Review request for Aurora and Bill Farner.
Repository: aurora
Description
---
## Problems
We are from eBay platform team. Previously, we used marathon to generate
Jenkins master instance in dedicated vms and recieve resource offer from same
dedicated vms. For the details, please refer to
http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU
Now, we found Aurora is more stable and powerful. We are moving from Marathon
to Aurora. During the move, we found there is no mesos role in Aurora now.
But we need use mesos role way to solve the problem in section Frameworks
stopped receiving offers after a while of the given url.
Here is a snippet of the problem description:
*We noticed occurred after we used Marathon to create the initial set of CI
masters. As those CI masters started registering themselves as frameworks,
Marathon stopped receiving any offers from Mesos; essentially, no new CI
masters could be launched. Let’s start with Marathon. In the DRF model, it
was unfair to treat Marathon in the same bucket/role alongside hundreds of
connected Jenkins frameworks. After launching all these Jenkins frameworks,
Marathon had a large resource share and Mesos would aggressively offer
resources to frameworks that were using little or no resources. Marathon was
placed last in priority and got starved out.*
*We decided to define a dedicated Mesos role for Marathon and to have all of
the Mesos slaves that were reserved for Jenkins master instances support that
Mesos role. Jenkins frameworks were left with the default role “*”.*This
solved the problem – Mesos offered resources per role and hence Marathon
never got starved out. A framework with a special role will get resource
offers from both slaves supporting that special role and also from the
default role “*”.* However, since we were using placement constraints,
Marathon accepted resource offers only from slaves that supported both the
role and the placement constraints.*
## Solution
So we add role feature is the source code to solve the problem in same way:
When accept a resource offer, Aurora will send back the needed resources to
Mesos with the mesos role in resource offer.
How to configure the Mesos role:
1.Add cmd option --mesos_role=${Mesos role name} when start Aurora scheduler.
We change the test cases according code change. Each changed test case is
green
Merge https://github.com/zhanglong2015/incubator-aurora
Diffs
-
src/main/java/org/apache/aurora/scheduler/ResourceSlot.java
1a158b4e0be94762ad0480e8ce74b19bacc90c97
src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java
31aa2bbaab3d97875493ad75c4d2c7c82ac7fa58
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java
b5a3140e3560f790d1db496dca3c2ee0dc96a195
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java
d0994203b5650f44ca2eb32e1e2aa61875163854
src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java
5340d651b298ec8aa079e73d6d2f652fdf876293
src/main/java/org/apache/aurora/scheduler/state/TaskAssigner.java
e1c29747c9854cf75bf63f6f085cf40ca68989af
src/test/java/org/apache/aurora/scheduler/async/GcExecutorLauncherTest.java
422d5a9a42310979752eb7282658316c2b772419
src/test/java/org/apache/aurora/scheduler/configuration/ResourcesTest.java
d6febb8998e05257cabe8d193cefa0b6c79f197e
src/test/java/org/apache/aurora/scheduler/mesos/MesosTaskFactoryImplTest.java
5f08d00d39f016af9bc296e517ad49b66ab5a8de
src/test/java/org/apache/aurora/scheduler/state/TaskAssignerImplTest.java
411a55a8d85f60bb2703468f2d69b64b2736eee4
Diff: https://reviews.apache.org/r/30710/diff/
Testing
---
:buildSrc:compileJava UP-TO-DATE
:buildSrc:compileGroovy UP-TO-DATE
:buildSrc:processResources UP-TO-DATE
:buildSrc:classes UP-TO-DATE
:buildSrc:jar UP-TO-DATE
:buildSrc:assemble UP-TO-DATE
:buildSrc:compileTestJava UP-TO-DATE
:buildSrc:compileTestGroovy UP-TO-DATE
:buildSrc:processTestResources UP-TO-DATE
:buildSrc:testClasses UP-TO-DATE
:buildSrc:test UP-TO-DATE
:buildSrc:check UP-TO-DATE
Re: Review Request 30710: add mesos role feature
---
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/30710/
---
(Updated Feb. 6, 2015, 9:23 a.m.)
Review request for Aurora and Bill Farner.
Changes
---
align code style with apache
Repository: aurora
Description
---
## Problems
We are from eBay platform team. Previously, we used marathon to generate
Jenkins master instance in dedicated vms and recieve resource offer from same
dedicated vms. For the details, please refer to
http://www.ebaytechblog.com/2014/04/04/delivering-ebays-ci-solution-with-apache-mesos-part-i/#.VNQUuC6_SPU
Now, we found Aurora is more stable and powerful. We are moving from Marathon
to Aurora. During the move, we found there is no mesos role in Aurora now. But
we need use mesos role way to solve the problem in section Frameworks stopped
receiving offers after a while of the given url.
Here is a snippet of the problem description:
*We noticed occurred after we used Marathon to create the initial set of CI
masters. As those CI masters started registering themselves as frameworks,
Marathon stopped receiving any offers from Mesos; essentially, no new CI
masters could be launched. Let’s start with Marathon. In the DRF model, it was
unfair to treat Marathon in the same bucket/role alongside hundreds of
connected Jenkins frameworks. After launching all these Jenkins frameworks,
Marathon had a large resource share and Mesos would aggressively offer
resources to frameworks that were using little or no resources. Marathon was
placed last in priority and got starved out.*
*We decided to define a dedicated Mesos role for Marathon and to have all of
the Mesos slaves that were reserved for Jenkins master instances support that
Mesos role. Jenkins frameworks were left with the default role “*”.*This solved
the problem – Mesos offered resources per role and hence Marathon never got
starved out. A framework with a special role will get resource offers from both
slaves supporting that special role and also from the default role “*”.*
However, since we were using placement constraints, Marathon accepted resource
offers only from slaves that supported both the role and the placement
constraints.*
## Solution
So we add role feature is the source code to solve the problem in same way:
When accept a resource offer, Aurora will send back the needed resources to
Mesos with the mesos role in resource offer.
How to configure the Mesos role:
1.Add cmd option --mesos_role=${Mesos role name} when start Aurora scheduler.
We change the test cases according code change. Each changed test case is green
Merge https://github.com/zhanglong2015/incubator-aurora
Diffs (updated)
-
src/main/java/org/apache/aurora/scheduler/ResourceSlot.java
1a158b4e0be94762ad0480e8ce74b19bacc90c97
src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java
31aa2bbaab3d97875493ad75c4d2c7c82ac7fa58
src/main/java/org/apache/aurora/scheduler/configuration/Resources.java
b5a3140e3560f790d1db496dca3c2ee0dc96a195
src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java
d0994203b5650f44ca2eb32e1e2aa61875163854
src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java
5340d651b298ec8aa079e73d6d2f652fdf876293
src/main/java/org/apache/aurora/scheduler/state/TaskAssigner.java
e1c29747c9854cf75bf63f6f085cf40ca68989af
src/test/java/org/apache/aurora/scheduler/async/GcExecutorLauncherTest.java
422d5a9a42310979752eb7282658316c2b772419
src/test/java/org/apache/aurora/scheduler/configuration/ResourcesTest.java
d6febb8998e05257cabe8d193cefa0b6c79f197e
src/test/java/org/apache/aurora/scheduler/mesos/MesosTaskFactoryImplTest.java
5f08d00d39f016af9bc296e517ad49b66ab5a8de
src/test/java/org/apache/aurora/scheduler/state/TaskAssignerImplTest.java
411a55a8d85f60bb2703468f2d69b64b2736eee4
Diff: https://reviews.apache.org/r/30710/diff/
Testing
---
:buildSrc:compileJava UP-TO-DATE
:buildSrc:compileGroovy UP-TO-DATE
:buildSrc:processResources UP-TO-DATE
:buildSrc:classes UP-TO-DATE
:buildSrc:jar UP-TO-DATE
:buildSrc:assemble UP-TO-DATE
:buildSrc:compileTestJava UP-TO-DATE
:buildSrc:compileTestGroovy UP-TO-DATE
:buildSrc:processTestResources UP-TO-DATE
:buildSrc:testClasses UP-TO-DATE
:buildSrc:test UP-TO-DATE
:buildSrc:check UP-TO-DATE
:buildSrc:build UP-TO-DATE
:api:generateThriftJava
:api:classesThrift
Note: Some input files use unchecked or unsafe operations.
Note: Recompile with -Xlint:unchecked for details.
:api:checkPython
:api:generateThriftEntitiesJava
:api:classesThriftEntities
:api:compileJava UP-TO-DATE
:api:generateThriftResources
:api:processResources UP-TO-DATE
:api:classes
:api:jar
:compileJava
Note: Writing
file:/root/incubator-aurora/dist/classes/main/com/twitter/common/args/apt/cmdline.arg.info.txt.2
:processResources
:classes
:jar
:assemble
Re: Review Request 30710: add mesos role feature
On Feb. 6, 2015, 7:02 a.m., Aurora ReviewBot wrote: Master (5059040) is red with this patch. ./build-support/jenkins/build.sh 1 warning :processJmhResources UP-TO-DATE :jmhClasses :checkstyleJmh :jsHint :checkstyleMain[ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java:18: Wrong order for 'java.util.Collections' import. Order should be: java, javax, scala, com, net, org. Each group should be separated by a single blank line. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java:51: Wrong order for 'com.google.common.annotations.VisibleForTesting' import. Order should be: java, javax, scala, com, net, org. Each group should be separated by a single blank line. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java:185: Line is longer than 100 characters (found 111). [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java:188:5: File contains tab characters (this is the first instance). [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:103: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:105:1: File contains tab characters (this is the first instance). [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:105:5: 'if' is not followed by whitespace. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:128:83: ',' is not followed by whitespace. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:145: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:317: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:319: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:331: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java:56: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java:115:27: '+' is not preceded with whitespace. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java:118:1: File contains tab characters (this is the first instance). [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java:18: Wrong order for 'java.util.List' import. Order should be: java, javax, scala, com, net, org. Each group should be separated by a single blank line. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java:47: Wrong order for 'com.google.common.annotations.VisibleForTesting' import. Order should be: java, javax, scala, com, net, org. Each group should be separated by a single blank line. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java:115: File contains a sequence of empty lines. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java:120:1: File contains tab characters (this is the first instance). [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java:177: Line is longer than 100 characters (found 104). FAILED FAILURE: Build failed with an exception. * What went wrong: Execution failed for task ':checkstyleMain'. Checkstyle rule violations were found. See the report at: file:///home/jenkins/jenkins-slave/workspace/AuroraBot/dist/reports/checkstyle/main.xml * Try: Run with --stacktrace option to get the stack trace. Run with --info or
Re: Review Request 30710: add mesos role feature
--- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/30710/#review71386 --- Master (5059040) is red with this patch. ./build-support/jenkins/build.sh 1 warning :processJmhResources UP-TO-DATE :jmhClasses :checkstyleJmh :jsHint :checkstyleMain[ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java:18: Wrong order for 'java.util.Collections' import. Order should be: java, javax, scala, com, net, org. Each group should be separated by a single blank line. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java:51: Wrong order for 'com.google.common.annotations.VisibleForTesting' import. Order should be: java, javax, scala, com, net, org. Each group should be separated by a single blank line. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java:185: Line is longer than 100 characters (found 111). [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/async/GcExecutorLauncher.java:188:5: File contains tab characters (this is the first instance). [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:103: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:105:1: File contains tab characters (this is the first instance). [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:105:5: 'if' is not followed by whitespace. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:128:83: ',' is not followed by whitespace. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:145: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:317: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:319: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/configuration/Resources.java:331: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java:56: Line has trailing spaces. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java:115:27: '+' is not preceded with whitespace. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/CommandLineDriverSettingsModule.java:118:1: File contains tab characters (this is the first instance). [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java:18: Wrong order for 'java.util.List' import. Order should be: java, javax, scala, com, net, org. Each group should be separated by a single blank line. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java:47: Wrong order for 'com.google.common.annotations.VisibleForTesting' import. Order should be: java, javax, scala, com, net, org. Each group should be separated by a single blank line. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java:115: File contains a sequence of empty lines. [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java:120:1: File contains tab characters (this is the first instance). [ant:checkstyle] /home/jenkins/jenkins-slave/workspace/AuroraBot/src/main/java/org/apache/aurora/scheduler/mesos/MesosTaskFactory.java:177: Line is longer than 100 characters (found 104). FAILED FAILURE: Build failed with an exception. * What went wrong: Execution failed for task ':checkstyleMain'. Checkstyle rule violations were found. See the report at: file:///home/jenkins/jenkins-slave/workspace/AuroraBot/dist/reports/checkstyle/main.xml * Try: Run with --stacktrace option to get the stack trace. Run with --info or
