ODIn/spark-instrumented-optimizer
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity

[SPARK-30480][PYTHON][TESTS] Increases the memory limit being tested in 'WorkerMemoryTest.test_memory_limit'

Browse source 
### What changes were proposed in this pull request?

This PR proposes to increase the memory in `WorkerMemoryTest.test_memory_limit` in order to make the test pass with PyPy.

The test is currently failed only in PyPy as below in some PRs unexpectedly:

```
Current mem limits: 18446744073709551615 of max 18446744073709551615

Setting mem limits to 1048576 of max 1048576

RPython traceback:
  File "pypy_module_pypyjit_interp_jit.c", line 289, in portal_5
  File "pypy_interpreter_pyopcode.c", line 3468, in handle_bytecode__AccessDirect_None
  File "pypy_interpreter_pyopcode.c", line 5558, in dispatch_bytecode__AccessDirect_None
out of memory: couldn't allocate the next arena
ERROR
```

It seems related to how PyPy allocates the memory and GC works PyPy-specifically. There seems nothing wrong in this configuration implementation itself in PySpark side.

I roughly tested in higher PyPy versions on Ubuntu (PyPy v7.3.0) and this test seems passing fine so I suspect this might be an issue in old PyPy behaviours.

The change only increases the limit so it would not affect actual memory allocations. It just needs to test if the limit is properly set in worker sides. For clarification, the memory is unlimited in the machine if not set.

### Why are the changes needed?

To make the tests pass and unblock other PRs.

### Does this PR introduce any user-facing change?

No.

### How was this patch tested?

Manually and Jenkins should test it out.

Closes #27186 from HyukjinKwon/SPARK-30480.

Authored-by: HyukjinKwon <gurwls223@apache.org>
Signed-off-by: HyukjinKwon <gurwls223@apache.org>
This commit is contained in:
HyukjinKwon 2020-01-13 18:47:15 +09:00
parent eefcc7d762
commit 0823aec463
2 changed files with 5 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
python/pyspark/tests/test_worker.py
View file

@ -183,7 +183,7 @@ class WorkerReuseTest(PySparkTestCase):
class WorkerMemoryTest(PySparkTestCase):
def test_memory_limit(self):
self.sc._conf.set("spark.executor.pyspark.memory", "1m")
self.sc._conf.set("spark.executor.pyspark.memory", "2g")
rdd = self.sc.parallelize(xrange(1), 1)
def getrlimit():
@ -194,8 +194,8 @@ class WorkerMemoryTest(PySparkTestCase):
self.assertTrue(len(actual) == 1)
self.assertTrue(len(actual[0]) == 2)
[(soft_limit, hard_limit)] = actual
self.assertEqual(soft_limit, 1024 * 1024)
self.assertEqual(hard_limit, 1024 * 1024)
self.assertEqual(soft_limit, 2 * 1024 * 1024 * 1024)
self.assertEqual(hard_limit, 2 * 1024 * 1024 * 1024)
if __name__ == "__main__":

4
python/run-tests.py
View file

@ -267,8 +267,8 @@ def main():
python_implementation = subprocess_check_output(
[python_exec, "-c", "import platform; print(platform.python_implementation())"],
universal_newlines=True).strip()
LOGGER.debug("%s python_implementation is %s", python_exec, python_implementation)
LOGGER.debug("%s version is: %s", python_exec, subprocess_check_output(
LOGGER.info("%s python_implementation is %s", python_exec, python_implementation)
LOGGER.info("%s version is: %s", python_exec, subprocess_check_output(
[python_exec, "--version"], stderr=subprocess.STDOUT, universal_newlines=True).strip())
if should_test_modules:
for module in modules_to_test: