[Data] Retry RaySystemError application errors
#45079
Merged
Conversation
This file contains 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
Signed-off-by: Balaji Veeramani <balaji@anyscale.com>
bveeramani
requested review from
ericl,
scv119,
c21,
amogkam,
scottjlee,
raulchen,
stephanie-wang and
omatthew98
as code owners
raulchen
approved these changes
May 1, 2024
| # Ray won't retry system errors if they're raised from task code. To ensure | ||
| # Ray Data is fault tolerant to those errors, we need to add RaySystemError | ||
| # to the retry_exceptions list. | ||
| "retry_exceptions": [ray.exceptions.RaySystemError], |
There was a problem hiding this comment.
what will happen if user already set this? If user already set it to a list, we should append this to it. If they already set it to True, we should do nothing.
python/ray/data/_internal/execution/operators/actor_pool_map_operator.py
Outdated
Show resolved
Hide resolved
Signed-off-by: Balaji Veeramani <balaji@anyscale.com>
harborn
pushed a commit
to harborn/ray
that referenced
this pull request
May 8, 2024
Ray typically automatically retries system errors. However, in some cases, tasks raise `RaySystemError`s and those errors aren't automatically retried. To ensure Ray Data is fault tolerant to those failures, this PR explicitly adds `RaySystemError` to the list of retried exceptions. Note: one reason why tasks raise `RaySystemError`s is that when we shut down a node, the raylet process can die before worker processes and `ray.get()` throws an exception since it can't access plasma in raylet. A longterm solution would be to fix the shutdown order: raylet should be the last process to exit. Signed-off-by: Balaji Veeramani <balaji@anyscale.com>
peytondmurray
pushed a commit
to peytondmurray/ray
that referenced
this pull request
May 13, 2024
Ray typically automatically retries system errors. However, in some cases, tasks raise `RaySystemError`s and those errors aren't automatically retried. To ensure Ray Data is fault tolerant to those failures, this PR explicitly adds `RaySystemError` to the list of retried exceptions. Note: one reason why tasks raise `RaySystemError`s is that when we shut down a node, the raylet process can die before worker processes and `ray.get()` throws an exception since it can't access plasma in raylet. A longterm solution would be to fix the shutdown order: raylet should be the last process to exit. Signed-off-by: Balaji Veeramani <balaji@anyscale.com>
ryanaoleary
pushed a commit
to ryanaoleary/ray
that referenced
this pull request
Jun 6, 2024
Ray typically automatically retries system errors. However, in some cases, tasks raise `RaySystemError`s and those errors aren't automatically retried. To ensure Ray Data is fault tolerant to those failures, this PR explicitly adds `RaySystemError` to the list of retried exceptions. Note: one reason why tasks raise `RaySystemError`s is that when we shut down a node, the raylet process can die before worker processes and `ray.get()` throws an exception since it can't access plasma in raylet. A longterm solution would be to fix the shutdown order: raylet should be the last process to exit. Signed-off-by: Balaji Veeramani <balaji@anyscale.com> Signed-off-by: Ryan O'Leary <ryanaoleary@google.com>
ryanaoleary
pushed a commit
to ryanaoleary/ray
that referenced
this pull request
Jun 7, 2024
Ray typically automatically retries system errors. However, in some cases, tasks raise `RaySystemError`s and those errors aren't automatically retried. To ensure Ray Data is fault tolerant to those failures, this PR explicitly adds `RaySystemError` to the list of retried exceptions. Note: one reason why tasks raise `RaySystemError`s is that when we shut down a node, the raylet process can die before worker processes and `ray.get()` throws an exception since it can't access plasma in raylet. A longterm solution would be to fix the shutdown order: raylet should be the last process to exit. Signed-off-by: Balaji Veeramani <balaji@anyscale.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Why are these changes needed?
Ray typically automatically retries system errors. However, in some cases, tasks raise
RaySystemErrors and those errors aren't automatically retried. To ensure Ray Data is fault tolerant to those failures, this PR explicitly addsRaySystemErrorto the list of retried exceptions.Note: one reason why tasks raise
RaySystemErrors is that when we shut down a node, the raylet process can die before worker processes andray.get()throws an exception since it can't access plasma in raylet. A longterm solution would be to fix the shutdown order: raylet should be the last process to exit.Related issue number
Deflakes #44892
Checks
git commit -s) in this PR.scripts/format.shto lint the changes in this PR.method in Tune, I've added it in
doc/source/tune/api/under thecorresponding
.rstfile.