gaogaotiantian opened a new pull request, #53783:
URL: https://github.com/apache/spark/pull/53783
<!--
Thanks for sending a pull request! Here are some tips for you:
1. If this is your first time, please read our contributor guidelines:
https://spark.apache.org/contributing.html
2. Ensure you have added or run the appropriate tests for your PR:
https://spark.apache.org/developer-tools.html
3. If the PR is unfinished, add '[WIP]' in your PR title, e.g.,
'[WIP][SPARK-XXXX] Your PR title ...'.
4. Be sure to keep the PR description updated to reflect all changes.
5. Please write your PR title to summarize what this PR proposes.
6. If possible, provide a concise example to reproduce the issue for a
faster review.
7. If you want to add a new configuration, please read the guideline first
for naming configurations in
'core/src/main/scala/org/apache/spark/internal/config/ConfigEntry.scala'.
8. If you want to add or modify an error type or message, please read the
guideline first in
'common/utils/src/main/resources/error/README.md'.
-->
### What changes were proposed in this pull request?
<!--
Please clarify what changes you are proposing. The purpose of this section
is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster
reviews in your PR. See the examples below.
1. If you refactor some codes with changing classes, showing the class
hierarchy will help reviewers.
2. If you fix some SQL features, you can provide some references of other
DBMSes.
3. If there is design documentation, please add the link.
4. If there is a discussion in the mailing list, please add the link.
-->
Commented out `del_remote_cache` in `RemoteModelRef`. This is a temporary
solution for the deadlock issue of `test_distributed_lda`. It's acceptable but
there might be better options.
### Why are the changes needed?
<!--
Please clarify why the changes are needed. For instance,
1. If you propose a new API, clarify the use case for a new API.
2. If you fix a bug, you can clarify why it is a bug.
-->
We have a very flaky test `test_distributed_lda`
[Failure1](https://github.com/gaogaotiantian/spark/actions/runs/20940523104/job/60173040026),
[Failure2](https://github.com/apache/spark/actions/runs/20938424034/job/60166745846),
[Failure3](https://github.com/apache/spark/actions/runs/20908277105/job/60066145491).
This is a deadlock issue with the traceback:
```
(Python) File
"/__w/spark/spark/python/pyspark/ml/tests/connect/test_parity_clustering.py",
line 30, in <module>
main()
(Python) File
"/__w/spark/spark/python/pyspark/testing/unittestutils.py", line 43, in main
unittest.main(module=module, testRunner=testRunner, verbosity=2)
(Python) File "/usr/lib/python3.11/unittest/main.py", line 102, in
__init__
self.runTests()
(Python) File "/usr/lib/python3.11/unittest/main.py", line 274, in
runTests
self.result = testRunner.run(self.test)
(Python) File
"/usr/local/lib/python3.11/dist-packages/xmlrunner/runner.py", line 67, in run
test(result)
(Python) File "/usr/lib/python3.11/unittest/suite.py", line 84, in
__call__
return self.run(*args, **kwds)
(Python) File "/usr/lib/python3.11/unittest/suite.py", line 122, in run
test(result)
(Python) File "/usr/lib/python3.11/unittest/suite.py", line 84, in
__call__
return self.run(*args, **kwds)
(Python) File "/usr/lib/python3.11/unittest/suite.py", line 122, in run
test(result)
(Python) File "/usr/lib/python3.11/unittest/case.py", line 678, in
__call__
return self.run(*args, **kwds)
(Python) File "/usr/lib/python3.11/unittest/case.py", line 623, in run
self._callTestMethod(testMethod)
(Python) File "/usr/lib/python3.11/unittest/case.py", line 579, in
_callTestMethod
if method() is not None:
(Python) File
"/__w/spark/spark/python/pyspark/ml/tests/test_clustering.py", line 466, in
test_distributed_lda
self.assertEqual(str(model), str(model2))
(Python) File "/__w/spark/spark/python/pyspark/ml/wrapper.py", line 474,
in __repr__
return self._call_java("toString")
(Python) File "/__w/spark/spark/python/pyspark/ml/util.py", line 322, in
wrapped
return remote_call()
(Python) File "/__w/spark/spark/python/pyspark/ml/util.py", line 308, in
remote_call
(_, properties, _) = session.client.execute_command(command)
(Python) File
"/__w/spark/spark/python/pyspark/sql/connect/client/core.py", line 1162, in
execute_command
data, _, metrics, observed_metrics, properties =
self._execute_and_fetch(
(Python) File
"/__w/spark/spark/python/pyspark/sql/connect/client/core.py", line 1664, in
_execute_and_fetch
for response in self._execute_and_fetch_as_iterator(
(Python) File
"/__w/spark/spark/python/pyspark/sql/connect/client/core.py", line 1621, in
_execute_and_fetch_as_iterator
generator = ExecutePlanResponseReattachableIterator(
(Python) File
"/__w/spark/spark/python/pyspark/sql/connect/client/reattach.py", line 127, in
__init__
self._stub.ExecutePlan(self._initial_request, metadata=metadata)
(Python) File
"/usr/local/lib/python3.11/dist-packages/grpc/_channel.py", line 1396, in
__call__
call = self._managed_call(
(Python) File
"/usr/local/lib/python3.11/dist-packages/grpc/_channel.py", line 1785, in create
call = state.channel.integrated_call(
(Python) File "/usr/lib/python3.11/threading.py", line 905, in __init__
self._started = Event()
(Python) File "/usr/lib/python3.11/threading.py", line 563, in __init__
self._cond = Condition(Lock())
(Python) File "/usr/lib/python3.11/threading.py", line 254, in __init__
self._release_save = lock._release_save
(Python) File "/__w/spark/spark/python/pyspark/ml/util.py", line 379, in
wrapped
self._remote_model_obj.release_ref()
(Python) File "/__w/spark/spark/python/pyspark/ml/util.py", line 162, in
release_ref
del_remote_cache(self.ref_id)
(Python) File "/__w/spark/spark/python/pyspark/ml/util.py", line 358, in
del_remote_cache
session.client._delete_ml_cache([ref_id])
(Python) File
"/__w/spark/spark/python/pyspark/sql/connect/client/core.py", line 2137, in
_delete_ml_cache
(_, properties, _) = self.execute_command(command)
(Python) File
"/__w/spark/spark/python/pyspark/sql/connect/client/core.py", line 1162, in
execute_command
data, _, metrics, observed_metrics, properties =
self._execute_and_fetch(
(Python) File
"/__w/spark/spark/python/pyspark/sql/connect/client/core.py", line 1664, in
_execute_and_fetch
for response in self._execute_and_fetch_as_iterator(
(Python) File
"/__w/spark/spark/python/pyspark/sql/connect/client/core.py", line 1621, in
_execute_and_fetch_as_iterator
generator = ExecutePlanResponseReattachableIterator(
(Python) File
"/__w/spark/spark/python/pyspark/sql/connect/client/reattach.py", line 127, in
__init__
self._stub.ExecutePlan(self._initial_request, metadata=metadata)
(Python) File
"/usr/local/lib/python3.11/dist-packages/grpc/_channel.py", line 1396, in
__call__
call = self._managed_call(
(Python) File
"/usr/local/lib/python3.11/dist-packages/grpc/_channel.py", line 1784, in create
with state.lock:
```
The deadlock happened in `create` function, the issue is that it gets the
exclusive lock to create a gRPC command, then another gRPC command is being
created during the process and trying to get the same exclusive lock (which is
not re-entriable).
This is triggered by a garbage collection, then `__del__` method of
`JavaWrapper` tried to release cache by sending a gRPC command to remote.
The design is fundamentally wrong and there is no easy fix. You simply can't
send gRPC command in `__del__` because that method can be triggered at any
arbitrary point (as gc can happen anytime).
I can think of three options:
1. Like this, we just do not clear cache. Simplest way to solve the issue.
Connect client will send a command to clear cache when it's being deleted.
2. We stop gc while we make command. Not the best solution but it might just
work. This kind of plays with Python internal mechanism but the code would be
clean and we can have the desired behavior.
3. Do something like `execute_command_later`, and queue the commands to run
in the next synchronous call to `execute_command`. A bit refactoring to the
existing framework, command will be delayed, but code wise it might be the most
accurate.
Notice that this problem is not only affecting this specific test, or even
our general test suite. This is a real deadlock problem that could happen to
our users. We should backport the final fix.
For now, we need to make a quick decision about how to mitigate the flaky
test - it's interrupting our workflow. One flaky test will make the whole
workflow fragile.
### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such as
new features, bug fixes, or other behavior changes. Documentation-only updates
are not considered user-facing changes.
If yes, please clarify the previous behavior and the change this PR proposes
- provide the console output, description and/or an example to show the
behavior difference if possible.
If possible, please also clarify if this is a user-facing change compared to
the released Spark versions or within the unreleased branches such as master.
If no, write 'No'.
-->
Not really, user should not know that cache is not being cleared eagerly.
### How was this patch tested?
<!--
If tests were added, say they were added here. Please make sure to add some
test cases that check the changes thoroughly including negative and positive
cases if possible.
If it was tested in a way different from regular unit tests, please clarify
how you tested step by step, ideally copy and paste-able, so that other
reviewers can test and check, and descendants can verify in the future.
If tests were not added, please describe why they were not added and/or why
it was difficult to add.
If benchmark tests were added, please run the benchmarks in GitHub Actions
for the consistent environment, and the instructions could accord to:
https://spark.apache.org/developer-tools.html#github-workflow-benchmarks.
-->
It's not super reproducible so we need to rely on CI.
### Was this patch authored or co-authored using generative AI tooling?
<!--
If generative AI tooling has been used in the process of authoring this
patch, please include the
phrase: 'Generated-by: ' followed by the name of the tool and its version.
If no, write 'No'.
Please refer to the [ASF Generative Tooling
Guidance](https://www.apache.org/legal/generative-tooling.html) for details.
-->
No
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
