[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14532418#comment-14532418
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-Yarn-trunk-Java8 #187 (See
[https://builds.apache.org/job/Hadoop-Yarn-trunk-Java8/187/])
YARN-3243. Moving CHANGES.txt entry to the right release. (vinodkv: rev
185e63a72638f01bb27e790c8dedf458923a2cae)
* hadoop-yarn-project/CHANGES.txt
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0, 2.7.1
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14532732#comment-14532732
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-Hdfs-trunk #2118 (See
[https://builds.apache.org/job/Hadoop-Hdfs-trunk/2118/])
YARN-3243. Moving CHANGES.txt entry to the right release. (vinodkv: rev
185e63a72638f01bb27e790c8dedf458923a2cae)
* hadoop-yarn-project/CHANGES.txt
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0, 2.7.1
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14532879#comment-14532879
]
Hudson commented on YARN-3243:
--
SUCCESS: Integrated in Hadoop-Mapreduce-trunk #2136 (See
[https://builds.apache.org/job/Hadoop-Mapreduce-trunk/2136/])
YARN-3243. Moving CHANGES.txt entry to the right release. (vinodkv: rev
185e63a72638f01bb27e790c8dedf458923a2cae)
* hadoop-yarn-project/CHANGES.txt
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0, 2.7.1
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14532764#comment-14532764
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-Hdfs-trunk-Java8 #177 (See
[https://builds.apache.org/job/Hadoop-Hdfs-trunk-Java8/177/])
YARN-3243. Moving CHANGES.txt entry to the right release. (vinodkv: rev
185e63a72638f01bb27e790c8dedf458923a2cae)
* hadoop-yarn-project/CHANGES.txt
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0, 2.7.1
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14532805#comment-14532805
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-Mapreduce-trunk-Java8 #187 (See
[https://builds.apache.org/job/Hadoop-Mapreduce-trunk-Java8/187/])
YARN-3243. Moving CHANGES.txt entry to the right release. (vinodkv: rev
185e63a72638f01bb27e790c8dedf458923a2cae)
* hadoop-yarn-project/CHANGES.txt
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0, 2.7.1
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14532431#comment-14532431
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-Yarn-trunk #920 (See
[https://builds.apache.org/job/Hadoop-Yarn-trunk/920/])
YARN-3243. Moving CHANGES.txt entry to the right release. (vinodkv: rev
185e63a72638f01bb27e790c8dedf458923a2cae)
* hadoop-yarn-project/CHANGES.txt
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0, 2.7.1
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14531036#comment-14531036
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-trunk-Commit #7748 (See
[https://builds.apache.org/job/Hadoop-trunk-Commit/7748/])
YARN-3243. Moving CHANGES.txt entry to the right release. (vinodkv: rev
185e63a72638f01bb27e790c8dedf458923a2cae)
* hadoop-yarn-project/CHANGES.txt
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0, 2.7.1
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14523187#comment-14523187
]
Thomas Graves commented on YARN-3243:
-
thanks [~leftnoteasy] I'll attempt to merge YARN-3434. If its not clean I'll
put up a patch for it.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14521580#comment-14521580
]
Thomas Graves commented on YARN-3243:
-
[~leftnoteasy] Can we pull this back into the branch-2.7?
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14522021#comment-14522021
]
Thomas Graves commented on YARN-3243:
-
I was wanting to pull YARN-3434 back into 2.7. It kind of depends on this one.
Atleast I think it would merge cleanly if this one was there.
This is also fixing a bug which I would like to see fixed in the 2.7 line if we
are going to use it. Its not a blocker since it exists in our 2.6 but it would
be nice to have. If we decide its to big then I'll just port YARN-3434 back
without it
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14522014#comment-14522014
]
Wangda Tan commented on YARN-3243:
--
Thanks comment from [~vinodkv].
I would +1 for back port this into branch-2.7, even if this patch is
potentially required to support non-exclusive node label, but this patch itself
is a bug fix instead of new feature.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14522008#comment-14522008
]
Vinod Kumar Vavilapalli commented on YARN-3243:
---
bq. Wangda Tan Can we pull this back into the branch-2.7?
I am not against it, but we need to rationalize why it needs to be pulled in.
Specifically given the fact that this is a big patch. Also, it wouldn't stop
here and you'll need more patches that depend on this?
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14522044#comment-14522044
]
Wangda Tan commented on YARN-3243:
--
[~tgraves], I think YARN-3434 needs YARN-3361, it cannot cleanly merge only
with YARN-3243. Could you check it?
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14522066#comment-14522066
]
Thomas Graves commented on YARN-3243:
-
It might to merge completely clean but it wouldn't require it for
functionality. It would be nice to have this in 2.7 either way though.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14522241#comment-14522241
]
Wangda Tan commented on YARN-3243:
--
[~tgraves], I just merged this to branch-2.7.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14366995#comment-14366995
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-Yarn-trunk-Java8 #136 (See
[https://builds.apache.org/job/Hadoop-Yarn-trunk-Java8/136/])
YARN-3243. CapacityScheduler should pass headroom from parent to children to
make sure ParentQueue obey its capacity limits. Contributed by Wangda Tan.
(jianhe: rev 487374b7fe0c92fc7eb1406c568952722b5d5b15)
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/common/fica/FiCaSchedulerApp.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/LeafQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestCapacityScheduler.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestLeafQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/ParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestApplicationLimits.java
* hadoop-yarn-project/CHANGES.txt
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestChildQueueOrder.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/AbstractCSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestReservations.java
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14367007#comment-14367007
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-Yarn-trunk #870 (See
[https://builds.apache.org/job/Hadoop-Yarn-trunk/870/])
YARN-3243. CapacityScheduler should pass headroom from parent to children to
make sure ParentQueue obey its capacity limits. Contributed by Wangda Tan.
(jianhe: rev 487374b7fe0c92fc7eb1406c568952722b5d5b15)
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/AbstractCSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestCapacityScheduler.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/ParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestLeafQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/LeafQueue.java
* hadoop-yarn-project/CHANGES.txt
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/common/fica/FiCaSchedulerApp.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestApplicationLimits.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestChildQueueOrder.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestReservations.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14367077#comment-14367077
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-Hdfs-trunk #2068 (See
[https://builds.apache.org/job/Hadoop-Hdfs-trunk/2068/])
YARN-3243. CapacityScheduler should pass headroom from parent to children to
make sure ParentQueue obey its capacity limits. Contributed by Wangda Tan.
(jianhe: rev 487374b7fe0c92fc7eb1406c568952722b5d5b15)
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/common/fica/FiCaSchedulerApp.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestChildQueueOrder.java
* hadoop-yarn-project/CHANGES.txt
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestCapacityScheduler.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestReservations.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/ParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/LeafQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/AbstractCSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestApplicationLimits.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestLeafQueue.java
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14367242#comment-14367242
]
Hudson commented on YARN-3243:
--
SUCCESS: Integrated in Hadoop-Hdfs-trunk-Java8 #127 (See
[https://builds.apache.org/job/Hadoop-Hdfs-trunk-Java8/127/])
YARN-3243. CapacityScheduler should pass headroom from parent to children to
make sure ParentQueue obey its capacity limits. Contributed by Wangda Tan.
(jianhe: rev 487374b7fe0c92fc7eb1406c568952722b5d5b15)
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestCapacityScheduler.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/LeafQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/AbstractCSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestReservations.java
* hadoop-yarn-project/CHANGES.txt
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestApplicationLimits.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestChildQueueOrder.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestLeafQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/ParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/common/fica/FiCaSchedulerApp.java
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14367301#comment-14367301
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-Mapreduce-trunk #2086 (See
[https://builds.apache.org/job/Hadoop-Mapreduce-trunk/2086/])
YARN-3243. CapacityScheduler should pass headroom from parent to children to
make sure ParentQueue obey its capacity limits. Contributed by Wangda Tan.
(jianhe: rev 487374b7fe0c92fc7eb1406c568952722b5d5b15)
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestApplicationLimits.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestReservations.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestCapacityScheduler.java
* hadoop-yarn-project/CHANGES.txt
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/AbstractCSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestLeafQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestChildQueueOrder.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/common/fica/FiCaSchedulerApp.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/ParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/LeafQueue.java
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14367317#comment-14367317
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-Mapreduce-trunk-Java8 #136 (See
[https://builds.apache.org/job/Hadoop-Mapreduce-trunk-Java8/136/])
YARN-3243. CapacityScheduler should pass headroom from parent to children to
make sure ParentQueue obey its capacity limits. Contributed by Wangda Tan.
(jianhe: rev 487374b7fe0c92fc7eb1406c568952722b5d5b15)
* hadoop-yarn-project/CHANGES.txt
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/LeafQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/AbstractCSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestApplicationLimits.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestCapacityScheduler.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestChildQueueOrder.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestReservations.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/ParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestLeafQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/common/fica/FiCaSchedulerApp.java
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14365601#comment-14365601
]
Hudson commented on YARN-3243:
--
FAILURE: Integrated in Hadoop-trunk-Commit #7349 (See
[https://builds.apache.org/job/Hadoop-trunk-Commit/7349/])
YARN-3243. CapacityScheduler should pass headroom from parent to children to
make sure ParentQueue obey its capacity limits. Contributed by Wangda Tan.
(jianhe: rev 487374b7fe0c92fc7eb1406c568952722b5d5b15)
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestCapacityScheduler.java
* hadoop-yarn-project/CHANGES.txt
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/ParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestLeafQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/AbstractCSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestApplicationLimits.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestReservations.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestParentQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CSQueue.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/common/fica/FiCaSchedulerApp.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/TestChildQueueOrder.java
*
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/LeafQueue.java
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Fix For: 2.8.0
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14360739#comment-14360739
]
Wangda Tan commented on YARN-3243:
--
YARN-3204 tracks findbugs warning, and test failure is not related to this
change.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14361274#comment-14361274
]
Jian He commented on YARN-3243:
---
one thing is that the approach of subtracting all reserved resources to pass
through various limits so as to dive down into the sub-queues may cause a lot
of dry loop, which can be fixed separately.
Patch looks good to me. +1
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14357431#comment-14357431
]
Hadoop QA commented on YARN-3243:
-
{color:red}-1 overall{color}. Here are the results of testing the latest
attachment
http://issues.apache.org/jira/secure/attachment/12703964/YARN-3243.4.patch
against trunk revision c3003eb.
{color:green}+1 @author{color}. The patch does not contain any @author
tags.
{color:green}+1 tests included{color}. The patch appears to include 6 new
or modified test files.
{color:green}+1 javac{color}. The applied patch does not increase the
total number of javac compiler warnings.
{color:green}+1 javadoc{color}. There were no new javadoc warning messages.
{color:green}+1 eclipse:eclipse{color}. The patch built with
eclipse:eclipse.
{color:red}-1 findbugs{color}. The patch appears to introduce 5 new
Findbugs (version 2.0.3) warnings.
{color:green}+1 release audit{color}. The applied patch does not increase
the total number of release audit warnings.
{color:red}-1 core tests{color}. The patch failed these unit tests in
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager:
org.apache.hadoop.yarn.server.resourcemanager.TestRM
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.TestContainerAllocation
org.apache.hadoop.yarn.server.resourcemanager.security.TestRMDelegationTokens
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.TestCapacitySchedulerNodeLabelUpdate
Test results:
https://builds.apache.org/job/PreCommit-YARN-Build/6919//testReport/
Findbugs warnings:
https://builds.apache.org/job/PreCommit-YARN-Build/6919//artifact/patchprocess/newPatchFindbugsWarningshadoop-yarn-server-resourcemanager.html
Console output: https://builds.apache.org/job/PreCommit-YARN-Build/6919//console
This message is automatically generated.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14359460#comment-14359460
]
Hadoop QA commented on YARN-3243:
-
{color:red}-1 overall{color}. Here are the results of testing the latest
attachment
http://issues.apache.org/jira/secure/attachment/12704207/YARN-3243.5.patch
against trunk revision b49c3a1.
{color:green}+1 @author{color}. The patch does not contain any @author
tags.
{color:green}+1 tests included{color}. The patch appears to include 6 new
or modified test files.
{color:green}+1 javac{color}. The applied patch does not increase the
total number of javac compiler warnings.
{color:green}+1 javadoc{color}. There were no new javadoc warning messages.
{color:green}+1 eclipse:eclipse{color}. The patch built with
eclipse:eclipse.
{color:red}-1 findbugs{color}. The patch appears to introduce 5 new
Findbugs (version 2.0.3) warnings.
{color:green}+1 release audit{color}. The applied patch does not increase
the total number of release audit warnings.
{color:red}-1 core tests{color}. The patch failed these unit tests in
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager:
org.apache.hadoop.yarn.server.resourcemanager.TestRM
Test results:
https://builds.apache.org/job/PreCommit-YARN-Build/6945//testReport/
Findbugs warnings:
https://builds.apache.org/job/PreCommit-YARN-Build/6945//artifact/patchprocess/newPatchFindbugsWarningshadoop-yarn-server-resourcemanager.html
Console output: https://builds.apache.org/job/PreCommit-YARN-Build/6945//console
This message is automatically generated.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14359230#comment-14359230
]
Hadoop QA commented on YARN-3243:
-
{color:red}-1 overall{color}. Here are the results of testing the latest
attachment
http://issues.apache.org/jira/secure/attachment/12704207/YARN-3243.5.patch
against trunk revision 06ce1d9.
{color:green}+1 @author{color}. The patch does not contain any @author
tags.
{color:red}-1 tests included{color}. The patch doesn't appear to include
any new or modified tests.
Please justify why no new tests are needed for this
patch.
Also please list what manual steps were performed to
verify this patch.
{color:green}+1 javac{color}. The applied patch does not increase the
total number of javac compiler warnings.
{color:green}+1 javadoc{color}. There were no new javadoc warning messages.
{color:green}+1 eclipse:eclipse{color}. The patch built with
eclipse:eclipse.
{color:red}-1 findbugs{color}. The patch appears to introduce 5 new
Findbugs (version 2.0.3) warnings.
{color:green}+1 release audit{color}. The applied patch does not increase
the total number of release audit warnings.
{color:red}-1 core tests{color}. The patch failed these unit tests in
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager:
org.apache.hadoop.yarn.server.resourcemanager.security.TestAMRMTokens
org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.TestAllocationFileLoaderService
org.apache.hadoop.yarn.server.resourcemanager.TestKillApplicationWithRMHA
org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.TestFairSchedulerQueueACLs
org.apache.hadoop.yarn.server.resourcemanager.recovery.TestZKRMStateStoreZKClientConnections
org.apache.hadoop.yarn.server.resourcemanager.recovery.TestZKRMStateStore
org.apache.hadoop.yarn.server.resourcemanager.recovery.TestLeveldbRMStateStore
org.apache.hadoop.yarn.server.resourcemanager.recovery.TestZKRMStateStorePerf
org.apache.hadoop.yarn.server.resourcemanager.scheduler.TestSchedulerUtils
org.apache.hadoop.yarn.server.resourcemanager.security.TestClientToAMTokens
org.apache.hadoop.yarn.server.resourcemanager.recovery.TestFSRMStateStore
org.apache.hadoop.yarn.server.resourcemanager.TestClientRMService
org.apache.hadoop.yarn.server.resourcemanager.TestAMAuthorization
Test results:
https://builds.apache.org/job/PreCommit-YARN-Build/6942//testReport/
Findbugs warnings:
https://builds.apache.org/job/PreCommit-YARN-Build/6942//artifact/patchprocess/newPatchFindbugsWarningshadoop-yarn-server-resourcemanager.html
Console output: https://builds.apache.org/job/PreCommit-YARN-Build/6942//console
This message is automatically generated.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA):
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14359275#comment-14359275
]
Wangda Tan commented on YARN-3243:
--
Seems like Jenkins issue, rekicked.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14359089#comment-14359089
]
Wangda Tan commented on YARN-3243:
--
Hi [~jianhe],
I've done all updates, thanks for your comments!
bq. CapacityScheduler, why following code is moved ?
This is because CS will call updateClusterResource after the a new node is
added, which will set resourceLimits of queues. To set queue's resourceLimits,
it needs to how much resource in each partition, so it need to call
labelManager.activeNode before setting queue's resourceLimits.
Attaching new patch (ver.5)
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch, YARN-3243.5.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14357684#comment-14357684
]
Hadoop QA commented on YARN-3243:
-
{color:red}-1 overall{color}. Here are the results of testing the latest
attachment
http://issues.apache.org/jira/secure/attachment/12703964/YARN-3243.4.patch
against trunk revision 30c428a.
{color:green}+1 @author{color}. The patch does not contain any @author
tags.
{color:green}+1 tests included{color}. The patch appears to include 6 new
or modified test files.
{color:green}+1 javac{color}. The applied patch does not increase the
total number of javac compiler warnings.
{color:red}-1 javadoc{color}. The javadoc tool appears to have generated
43 warning messages.
See
https://builds.apache.org/job/PreCommit-YARN-Build/6923//artifact/patchprocess/diffJavadocWarnings.txt
for details.
{color:green}+1 eclipse:eclipse{color}. The patch built with
eclipse:eclipse.
{color:red}-1 findbugs{color}. The patch appears to introduce 5 new
Findbugs (version 2.0.3) warnings.
{color:green}+1 release audit{color}. The applied patch does not increase
the total number of release audit warnings.
{color:red}-1 core tests{color}. The patch failed these unit tests in
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-api
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-common
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager:
org.apache.hadoop.yarn.server.resourcemanager.webapp.TestRMWebServicesDelegationTokenAuthentication
org.apache.hadoop.yarn.server.resourcemanager.webapp.TestRMWebappAuthentication
org.apache.hadoop.yarn.server.resourcemanager.webapp.TestRMWebAppFairScheduler
org.apache.hadoop.yarn.server.resourcemanager.webapp.TestRMWebServicesApps
org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.TestContainerAllocation
Test results:
https://builds.apache.org/job/PreCommit-YARN-Build/6923//testReport/
Findbugs warnings:
https://builds.apache.org/job/PreCommit-YARN-Build/6923//artifact/patchprocess/newPatchFindbugsWarningshadoop-yarn-server-resourcemanager.html
Console output: https://builds.apache.org/job/PreCommit-YARN-Build/6923//console
This message is automatically generated.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14357736#comment-14357736
]
Wangda Tan commented on YARN-3243:
--
Javadoc warning are mis-reported by Jenkins, findbugs warnings are tracked by
YARN-3204, test failures can pass locally.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14357846#comment-14357846
]
Jian He commented on YARN-3243:
---
- getResourceLimitsOfChild: code comments should match the variable name
- CapacityScheduler, why following code is moved ?
{code}
// update this node to node label manager
if (labelManager != null) {
labelManager.activateNode(nodeManager.getNodeID(),
nodeManager.getTotalCapability());
}
{code}
- needToUnreserve is removed, comments is invalid any more
{code}
// we got here by possibly ignoring parent queue capacity limits. If
// the
parameter needToUnreserve i
{code}
- the checkReservedContainers flag in canAssignToThisQueue is not needed,
instead check whether resourceCouldBeUnreserved is none or not
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch,
YARN-3243.4.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14355608#comment-14355608
]
Hadoop QA commented on YARN-3243:
-
{color:red}-1 overall{color}. Here are the results of testing the latest
attachment
http://issues.apache.org/jira/secure/attachment/12703715/YARN-3243.2.patch
against trunk revision aa92b76.
{color:green}+1 @author{color}. The patch does not contain any @author
tags.
{color:green}+1 tests included{color}. The patch appears to include 7 new
or modified test files.
{color:green}+1 javac{color}. The applied patch does not increase the
total number of javac compiler warnings.
{color:green}+1 javadoc{color}. There were no new javadoc warning messages.
{color:green}+1 eclipse:eclipse{color}. The patch built with
eclipse:eclipse.
{color:red}-1 findbugs{color}. The patch appears to introduce 5 new
Findbugs (version 2.0.3) warnings.
{color:green}+1 release audit{color}. The applied patch does not increase
the total number of release audit warnings.
{color:red}-1 core tests{color}. The patch failed these unit tests in
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager:
org.apache.hadoop.yarn.server.resourcemanager.TestWorkPreservingRMRestart
The following test timeouts occurred in
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager:
org.apache.hadoop.yarn.server.resourcemanager.TestRMRestart
Test results:
https://builds.apache.org/job/PreCommit-YARN-Build/6903//testReport/
Findbugs warnings:
https://builds.apache.org/job/PreCommit-YARN-Build/6903//artifact/patchprocess/newPatchFindbugsWarningshadoop-yarn-server-resourcemanager.html
Console output: https://builds.apache.org/job/PreCommit-YARN-Build/6903//console
This message is automatically generated.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14355857#comment-14355857
]
Hadoop QA commented on YARN-3243:
-
{color:red}-1 overall{color}. Here are the results of testing the latest
attachment
http://issues.apache.org/jira/secure/attachment/12703744/YARN-3243.3.patch
against trunk revision 64eb068.
{color:green}+1 @author{color}. The patch does not contain any @author
tags.
{color:green}+1 tests included{color}. The patch appears to include 7 new
or modified test files.
{color:green}+1 javac{color}. The applied patch does not increase the
total number of javac compiler warnings.
{color:green}+1 javadoc{color}. There were no new javadoc warning messages.
{color:green}+1 eclipse:eclipse{color}. The patch built with
eclipse:eclipse.
{color:red}-1 findbugs{color}. The patch appears to introduce 5 new
Findbugs (version 2.0.3) warnings.
{color:green}+1 release audit{color}. The applied patch does not increase
the total number of release audit warnings.
{color:green}+1 core tests{color}. The patch passed unit tests in
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager.
Test results:
https://builds.apache.org/job/PreCommit-YARN-Build/6906//testReport/
Findbugs warnings:
https://builds.apache.org/job/PreCommit-YARN-Build/6906//artifact/patchprocess/newPatchFindbugsWarningshadoop-yarn-server-resourcemanager.html
Console output: https://builds.apache.org/job/PreCommit-YARN-Build/6906//console
This message is automatically generated.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14355995#comment-14355995
]
Jian He commented on YARN-3243:
---
looks good overall, some comments:
- {{AbstractCSQueue#getCurrentLimitResource}}
-- add comments about how currentLimitResource is calculated
- getResourceLimitsOfChild
-- myLimits- parentLimits
-- myMaxAvailableResource - parentMaxAvailableResource
-- childMaxResource - childConfiguredMaxResource
- setHeadroomInfo - setQueueResourceLimitsInfo
- needExtraNewOrReservedContainer flag - better name ?
shouldAllocOrReserveNewContainer?
- similarly for the needExtraNewOrReservedContainer method
- revert TestContainerAllocation change
- {{ 1GB (am) + 5GB * 2 = 9GB }} 5GB should be 4GB
- Do you think passing down a QueueHeadRoom compared with QueueMaxLimit may
make the code simpler
- the checkLimitsToReserve may not need to be invoked if we are assigning a
reserved container
{code}
if (reservationsContinueLooking) {
// // we got here by possibly ignoring parent queue capacity limits. If
// // the parameter needToUnreserve is true it means we ignored one of
// // those limits in the chance we could unreserve. If we are here
// // we aren't trying to unreserve so we can't allocate
// // anymore due to that parent limit
// boolean res = checkLimitsToReserve(clusterResource,
{code}
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch, YARN-3243.2.patch, YARN-3243.3.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14353704#comment-14353704
]
Jian He commented on YARN-3243:
---
thanks Wangda !
- ParentQueue#canAssignToThisQueue,
{code}
if (totalUsedCapacityRatio = maxAvailCapacity) {
canAssign = false;
break;
}
{code}
instead of comparing with ratio, I think it might be simpler to compare
resource value
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14349238#comment-14349238
]
Wangda Tan commented on YARN-3243:
--
mvn eclipse:eclipse can get passed locally.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14348290#comment-14348290
]
Hadoop QA commented on YARN-3243:
-
{color:red}-1 overall{color}. Here are the results of testing the latest
attachment
http://issues.apache.org/jira/secure/attachment/12702730/YARN-3243.1.patch
against trunk revision 3482080.
{color:green}+1 @author{color}. The patch does not contain any @author
tags.
{color:green}+1 tests included{color}. The patch appears to include 7 new
or modified test files.
{color:green}+1 javac{color}. The applied patch does not increase the
total number of javac compiler warnings.
{color:green}+1 javadoc{color}. There were no new javadoc warning messages.
{color:red}-1 eclipse:eclipse{color}. The patch failed to build with
eclipse:eclipse.
{color:green}+1 findbugs{color}. The patch does not introduce any new
Findbugs (version 2.0.3) warnings.
{color:green}+1 release audit{color}. The applied patch does not increase
the total number of release audit warnings.
{color:green}+1 core tests{color}. The patch passed unit tests in .
Test results:
https://builds.apache.org/job/PreCommit-YARN-Build/6863//testReport/
Console output: https://builds.apache.org/job/PreCommit-YARN-Build/6863//console
This message is automatically generated.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Attachments: YARN-3243.1.patch
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14343995#comment-14343995
]
Wangda Tan commented on YARN-3243:
--
How reservation continous looking works now:
- For ParentQueue, if capacity after all reserved containers dropped = maximum
capacity of the queue, it will try to assign containers on children.
- For LeafQueue, if capacity after all reserved containers of an application
dropped + required maximum capacity of the LeafQueue, will continue.
- For application, if LeafQueue/ParentQueue has marked some containers needs to
unreserve, it will tries to unreserve a container with resource asked
resource.
But actually, we need make,
{{min(LeafQueue.limit - LeafQueue.usage, user.limit - user.usage) - required +
application-unreserved-resource = 0}}.
And as what we've done in YARN-3265, {{LeafQueue.limit = min(Parent.limit,
LeafQueue.max).}}
Otherwise, some capacity limits in the queue hierarchy will be violated.
Passing ResourceLimits in the hierarchy can enforce the limit described above
and also simplify code structure. Working on a patch now.
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (YARN-3243) CapacityScheduler should pass headroom from parent to children to make sure ParentQueue obey its capacity limits.
[
https://issues.apache.org/jira/browse/YARN-3243?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14334200#comment-14334200
]
Vinod Kumar Vavilapalli commented on YARN-3243:
---
+1, this will greatly simply the scheduling logic!
CapacityScheduler should pass headroom from parent to children to make sure
ParentQueue obey its capacity limits.
-
Key: YARN-3243
URL: https://issues.apache.org/jira/browse/YARN-3243
Project: Hadoop YARN
Issue Type: Bug
Components: capacityscheduler, resourcemanager
Reporter: Wangda Tan
Assignee: Wangda Tan
Now CapacityScheduler has some issues to make sure ParentQueue always obeys
its capacity limits, for example:
1) When allocating container of a parent queue, it will only check
parentQueue.usage parentQueue.max. If leaf queue allocated a container.size
(parentQueue.max - parentQueue.usage), parent queue can excess its max
resource limit, as following example:
{code}
A (usage=54, max=55)
/ \
A1 A2 (usage=1, max=55)
(usage=53, max=53)
{code}
Queue-A2 is able to allocate container since its usage max, but if we do
that, A's usage can excess A.max.
2) When doing continous reservation check, parent queue will only tell
children you need unreserve *some* resource, so that I will less than my
maximum resource, but it will not tell how many resource need to be
unreserved. This may lead to parent queue excesses configured maximum
capacity as well.
With YARN-3099/YARN-3124, now we have {{ResourceUsage}} class in each class,
*here is my proposal*:
- ParentQueue will set its children's ResourceUsage.headroom, which means,
*maximum resource its children can allocate*.
- ParentQueue will set its children's headroom to be (saying parent's name is
qA): min(qA.headroom, qA.max - qA.used). This will make sure qA's
ancestors' capacity will be enforced as well (qA.headroom is set by qA's
parent).
- {{needToUnReserve}} is not necessary, instead, children can get how much
resource need to be unreserved to keep its parent's resource limit.
- More over, with this, YARN-3026 will make a clear boundary between
LeafQueue and FiCaSchedulerApp, headroom will consider user-limit, etc.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
