[MRG] Nested backend context manager#538
Merged
ogrisel merged 5 commits intojoblib:masterfrom Jan 23, 2018
Merged
Conversation
7b48b85 to
5fa776e
Compare
5fa776e to
f4a0965
Compare
Codecov Report
@@ Coverage Diff @@
## master #538 +/- ##
==========================================
+ Coverage 94.78% 95.04% +0.25%
==========================================
Files 38 38
Lines 4969 5005 +36
==========================================
+ Hits 4710 4757 +47
+ Misses 259 248 -11
Continue to review full report at Codecov.
|
e934b17 to
05bae7b
Compare
7d45b07 to
ec61f0a
Compare
…d backend by default
ec61f0a to
66b74b6
Compare
…d backend by default
66b74b6 to
fca1e83
Compare
…grisel/joblib into ogrisel-nested-backend-context-manager
| delayed(nested_call, check_pickle=False)(i) | ||
| for i in range(5) | ||
| ) | ||
| assert ba.i == 1 |
There was a problem hiding this comment.
Does this look like the right number of calls to retrieve, given my Parallel calls?
Contributor
Author
There was a problem hiding this comment.
Yes, there is one only call to retrieve per call to Parallel.__call__.
TomAugspurger
added a commit
to TomAugspurger/distributed
that referenced
this pull request
Jan 23, 2018
This implements the new API introduced in joblib/joblib#538 for to ensure nested parallelism works correctly. This breaks API as previously `DaskDistributedBackend` had a `client` attribute with the `Client` that was used. Nested parallelism will be workers creating a `DaskDistributedBackend`, with no 'scheduler_host'. To avoid deadlocks, worker threads will secede and rejoin.
TomAugspurger
added a commit
to TomAugspurger/distributed
that referenced
this pull request
Jan 23, 2018
This implements the new API introduced in joblib/joblib#538 for to ensure nested parallelism works correctly. This breaks API as previously `DaskDistributedBackend` had a `client` attribute with the `Client` that was used. Nested parallelism will be workers creating a `DaskDistributedBackend`, with no 'scheduler_host'. To avoid deadlocks, worker threads will secede and rejoin.
TomAugspurger
added a commit
to TomAugspurger/distributed
that referenced
this pull request
Jan 23, 2018
This implements the new API introduced in joblib/joblib#538 for to ensure nested parallelism works correctly. This breaks API as previously `DaskDistributedBackend` had a `client` attribute with the `Client` that was used. Nested parallelism will be workers creating a `DaskDistributedBackend`, with no 'scheduler_host'. To avoid deadlocks, worker threads will secede and rejoin.
TomAugspurger
added a commit
to TomAugspurger/distributed
that referenced
this pull request
Jan 23, 2018
This implements the new API introduced in joblib/joblib#538 for to ensure nested parallelism works correctly. This breaks API as previously `DaskDistributedBackend` had a `client` attribute with the `Client` that was used. Nested parallelism will be workers creating a `DaskDistributedBackend`, with no 'scheduler_host'. To avoid deadlocks, worker threads will secede and rejoin.
TomAugspurger
added a commit
to TomAugspurger/distributed
that referenced
this pull request
Jan 23, 2018
This implements the new API introduced in joblib/joblib#538 for to ensure nested parallelism works correctly. This breaks API as previously `DaskDistributedBackend` had a `client` attribute with the `Client` that was used. Nested parallelism will be workers creating a `DaskDistributedBackend`, with no 'scheduler_host'. To avoid deadlocks, worker threads will secede and rejoin.
ogrisel
commented
Jan 23, 2018
| 'retrieval_context' this could lead to deadlock, as all the workers | ||
| managed by the backend may be "busy" waiting for the nested parallel | ||
| calls to finish, but the backend has no free workers to execute those | ||
| tasks. |
Contributor
Author
There was a problem hiding this comment.
Very nice explanation :)
| delayed(nested_call, check_pickle=False)(i) | ||
| for i in range(5) | ||
| ) | ||
| assert ba.i == 1 |
Contributor
Author
There was a problem hiding this comment.
Yes, there is one only call to retrieve per call to Parallel.__call__.
|
|
||
| with parallel_backend("retrieval") as (ba, _): | ||
| Parallel(n_jobs=2)( | ||
| delayed(nested_call, check_pickle=False)(i) |
Contributor
Author
There was a problem hiding this comment.
We can probably remove that check_pickle=False now that loky is the default backend and that cloudpickle is used to pickle function definitions.
Contributor
Author
|
Merged! Thanks @TomAugspurger. |
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
2 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.
While reviewing #516 (loky integration) and #537 (backend negotiation based on code hints / semantic constraints), I realized that the
with parallel_backend(backend_name):idiom should support nesting when possible.Here is a (currently failing) test to demonstrate what I mean. I will work on an implementation. If we ensure that backend instances should be picklable, that should be quite easy to implement although I am worried about the performance overhead it could induce.