YARN-11005. Implement the core QUEUE_LENGTH_THEN_RESOURCES OContainer allocation policy #3717
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
… allocation policy * Implement the `QUEUE_LENGTH_THEN_RESOURCES` `LoadComparator` * Add `Resource` requested to select node methods when allocating `OPPORTUNISTIC` containers * Add node capacity and node allocated resources to `ClusterNode` * Extend `DominantResourceCalculator` to be able to compute min-share * Add tests and extend existing ones to consider 3 resource dimensions
|
💔 -1 overall
This message was automatically generated. |
|
💔 -1 overall
This message was automatically generated. |
|
🎊 +1 overall
This message was automatically generated. |
goiri
approved these changes
Nov 29, 2021
|
🎊 +1 overall
This message was automatically generated. |
bibinchundatt
requested changes
Dec 2, 2021
| } | ||
|
|
||
| @Test | ||
| public void testQueueLengthThenResourcesSort() { |
Contributor
There was a problem hiding this comment.
Could you add a test case when the resource size of node changes. How its getting handled in O containers
Contributor
Author
|
@goiri and @bibinchundatt thanks for your reviews! |
|
🎊 +1 overall
This message was automatically generated. |
| capability, currAllocated); | ||
| if (resourceCalculator.fitsIn(requested, currAvailable)) { | ||
| final Resource newAllocated = Resources.add(currAllocated, requested); | ||
| c.setAllocatedResource(newAllocated); |
Contributor
There was a problem hiding this comment.
with multiple application running in parallel the allocated resource update could be wrong here. I think we take clusterNode level lock when we are updating / reading and during checks
Contributor
Author
There was a problem hiding this comment.
Yeah, I saw that since the original code also wasn't handling parallelism correctly I thought it might've been fine to not use locks, but ensuring correctness is probably better here.
Contributor
Author
|
@bibinchundatt thanks for the review! Added a new commit addressing your comments. |
goiri
reviewed
Dec 3, 2021
| readLock.lock(); | ||
| try { | ||
| return this.capability; | ||
| } |
| readLock.lock(); | ||
| try { | ||
| return this.allocatedResource; | ||
| } |
| } | ||
| } | ||
|
|
||
| public boolean compareAndIncrementAllocation( |
Contributor
Author
There was a problem hiding this comment.
How so? We need to set queueLength if it fits, and ignore it if it does not.
Member
There was a problem hiding this comment.
I mean in terms of style, I think:
public boolean compareAndIncrementAllocation(final int incrementQLen) {
Fits in less than 100 characters (it may even be 120 now.
| int diff = 0; | ||
| switch (this) { | ||
| case QUEUE_LENGTH_THEN_RESOURCES: | ||
| diff = o1.getQueueLength() - o2.getQueueLength(); |
Member
There was a problem hiding this comment.
This math is a little intense.
It might be better to extract it to a function just for this so you can return things right away and have comments throughout.
| break; | ||
| } | ||
|
|
||
| // NOTE: cluster resource should be |
Member
There was a problem hiding this comment.
Do we just go ahead with leaving a note?
It doesn't seem very maintanable.
...rg/apache/hadoop/yarn/server/resourcemanager/scheduler/distributed/NodeQueueLoadMonitor.java
Show resolved
Hide resolved
|
🎊 +1 overall
This message was automatically generated. |
…on, add javadocs, formatting issues
Contributor
Author
|
@goiri thanks for the additional round of review! Latest commit should address your comments, please let me know if any other concerns. |
|
🎊 +1 overall
This message was automatically generated. |
goiri
approved these changes
Dec 6, 2021
|
🎊 +1 overall
This message was automatically generated. |
bibinchundatt
approved these changes
Dec 7, 2021
| comparator == LoadComparator.QUEUE_LENGTH || | ||
| comparator == LoadComparator.QUEUE_LENGTH_THEN_RESOURCES)) { | ||
| clusterNode | ||
| .setQueueWaitTime(estimatedQueueWaitTime) |
Contributor
There was a problem hiding this comment.
Every second we are taking too many write locks of ClusterNode. IIUC that not the best optimized option. We should probably take the write lock once and update all the values like waitTime,SetQueueLength,labels etc .
Contributor
Author
There was a problem hiding this comment.
Good catch, will address in a new iteration.
|
🎊 +1 overall
This message was automatically generated. |
|
🎊 +1 overall
This message was automatically generated. |
goiri
approved these changes
Dec 8, 2021
HarshitGupta11
pushed a commit
to HarshitGupta11/hadoop
that referenced
this pull request
Nov 28, 2022
… allocation policy (apache#3717)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description of PR
QUEUE_LENGTH_THEN_RESOURCESLoadComparatorResourcerequested to select node methods when allocatingOPPORTUNISTICcontainersClusterNodeDominantResourceCalculatorto be able to compute min-shareHow was this patch tested?