Doc on handling worker with walltime#481
Merged
guillaumeeb merged 3 commits intoJan 24, 2021
Merged
Conversation
mivade
reviewed
Jan 23, 2021
mivade
left a comment
mivade
left a comment
There was a problem hiding this comment.
Thanks, this looks like great additional documentation! I've pointed out some typos and some suggestions to clarify the language a bit, but otherwise this looks good!
| - when you don't have a lot of room on you HPC platform and have only a few workers at a time (less than what you were hopping for when using scale or adapt). These workers will be killed (and others started) before you workload ends. | ||
| - when you really don't know how long your workload will take: all your workers could be killed before reaching the end. In this case, you'll want to use adaptive clusters so that Dask ensures some workers are always up. | ||
|
|
||
| If you don't set the proper parameters, you'll run into KilleWorker exceptions in those two cases. |
|
|
||
| If you don't set the proper parameters, you'll run into KilleWorker exceptions in those two cases. | ||
|
|
||
| The solution to this problem is to tell Dask up front that the workers have a finit life time: |
There was a problem hiding this comment.
Typo: finit -> finite. Similarly lifetime is usually spelled as a single word.
|
|
||
| The solution to this problem is to tell Dask up front that the workers have a finit life time: | ||
|
|
||
| - Use `--lifetime` worker option. This will enables infinite workloads using adaptive. Workers will be properly shut down before the scheduling system kills them, and all their states moved. |
| How to handle job queueing system walltime killing workers | ||
| ---------------------------------------------------------- | ||
|
|
||
| In dask-jobqueue, every worker processes run inside a job, and all jobs have a time limit in job queueing systems. |
There was a problem hiding this comment.
Should be "every worker process runs..."
| In dask-jobqueue, every worker processes run inside a job, and all jobs have a time limit in job queueing systems. | ||
| Reaching walltime can be troublesome in several cases: | ||
|
|
||
| - when you don't have a lot of room on you HPC platform and have only a few workers at a time (less than what you were hopping for when using scale or adapt). These workers will be killed (and others started) before you workload ends. |
There was a problem hiding this comment.
hopping -> hoping and "before you workload" -> "before your workload"
| The solution to this problem is to tell Dask up front that the workers have a finit life time: | ||
|
|
||
| - Use `--lifetime` worker option. This will enables infinite workloads using adaptive. Workers will be properly shut down before the scheduling system kills them, and all their states moved. | ||
| - Use `--lifetime-stagger` when dealing with many workers (say > 20): this will allow to avoid workers all terminating at the same time, and so to ease rebalancing tasks and scheduling burden. |
There was a problem hiding this comment.
"this will allow to avoid workers all" -> "this will prevent workers from"
"and so to ease" -> "and so ease" or (probably better) "thus"
| cluster.adapt(minimum=0, maximum=200) | ||
|
|
||
|
|
||
| Here is an example of a workflow taking advantage of this, if you wan't to give it a try or adapt it to your use case: |
Member
Author
|
Many thanks @mivade! I need to practice my english... |
andersy005
approved these changes
Jan 24, 2021
andersy005
left a comment
andersy005
left a comment
Member
There was a problem hiding this comment.
Thank you for putting this together, @guillaumeeb!
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
Finally, a little contribution from me, and a doc fix to a long standing issue.
Fixes #122.