Skip to content

SlurmClusterExecutor - create a separate working directory for each job #698

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jan-janssen merged 5 commits into from
Jul 3, 2025
Merged

SlurmClusterExecutor - create a separate working directory for each job #698

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
981b77e
SlurmClusterExecutor - create a separate working directory for each job
jan-janssen Jul 2, 2025
cab77e3
The cache directory is a mandatory input
jan-janssen Jul 2, 2025
9535958
extend tests
jan-janssen Jul 2, 2025
da6f689
test length after successful execution
jan-janssen Jul 2, 2025
0d9ce2b
fix
jan-janssen Jul 2, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 5 additions & 3 deletions executorlib/task_scheduler/file/queue_spawner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,18 +10,19 @@

def execute_with_pysqa(
command: list,
cache_directory: str,
task_dependent_lst: Optional[list[int]] = None,
file_name: Optional[str] = None,
resource_dict: Optional[dict] = None,
config_directory: Optional[str] = None,
backend: Optional[str] = None,
cache_directory: Optional[str] = None,
) -> Optional[int]:
"""
Execute a command by submitting it to the queuing system

Args:
command (list): The command to be executed.
cache_directory (str): The directory to store the HDF5 files.
task_dependent_lst (list): A list of subprocesses that the current subprocess depends on. Defaults to [].
file_name (str): Name of the HDF5 file which contains the Python function
resource_dict (dict): resource dictionary, which defines the resources used for the execution of the function.
Expand All @@ -30,7 +31,6 @@ def execute_with_pysqa(
}
config_directory (str, optional): path to the config directory.
backend (str, optional): name of the backend used to spawn tasks.
cache_directory (str): The directory to store the HDF5 files.

Returns:
int: queuing system ID
Expand All @@ -50,7 +50,9 @@ def execute_with_pysqa(
if "cwd" in resource_dict and resource_dict["cwd"] is not None:
cwd = resource_dict["cwd"]
else:
cwd = cache_directory
folder = command[-1].split("_i.h5")[0]
cwd = os.path.join(cache_directory, folder)
os.makedirs(cwd, exist_ok=True)
Comment on lines +53 to +55

@coderabbitai coderabbitai Bot Jul 2, 2025

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Potential issue

Add input validation and improve robustness of folder name extraction.

The current implementation has several potential issues:

  1. Missing input validation: No check if command is empty, which would cause IndexError on command[-1]
  2. Brittle string parsing: The code assumes command[-1] contains "_i.h5", but if it doesn't, the entire string becomes the folder name
  3. Path security: The folder name comes from user input and could contain path traversal characters like "../"

Consider this more robust implementation:

-            folder = command[-1].split("_i.h5")[0]
-            cwd = os.path.join(cache_directory, folder)
-            os.makedirs(cwd, exist_ok=True)
+            if not command:
+                raise ValueError("Command list cannot be empty")
+            
+            # Extract folder name safely
+            last_arg = command[-1]
+            if "_i.h5" in last_arg:
+                folder = last_arg.split("_i.h5")[0]
+            else:
+                # Fallback: use basename without extension
+                folder = os.path.splitext(os.path.basename(last_arg))[0]
+            
+            # Sanitize folder name to prevent path traversal
+            folder = os.path.basename(folder)  # Remove any path components
+            if not folder or folder in (".", ".."):
+                folder = "default_job"
+            
+            cwd = os.path.join(cache_directory, folder)
+            os.makedirs(cwd, exist_ok=True)
📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change
folder = command[-1].split("_i.h5")[0]
cwd = os.path.join(cache_directory, folder)
os.makedirs(cwd, exist_ok=True)
if not command:
raise ValueError("Command list cannot be empty")
# Extract folder name safely
last_arg = command[-1]
if "_i.h5" in last_arg:
folder = last_arg.split("_i.h5")[0]
else:
# Fallback: use basename without extension
folder = os.path.splitext(os.path.basename(last_arg))[0]
# Sanitize folder name to prevent path traversal
folder = os.path.basename(folder) # Remove any path components
if not folder or folder in (".", ".."):
folder = "default_job"
cwd = os.path.join(cache_directory, folder)
os.makedirs(cwd, exist_ok=True)
🤖 Prompt for AI Agents 
In executorlib/task_scheduler/file/queue_spawner.py around lines 53 to 55, add
validation to ensure the command list is not empty before accessing command[-1]
to prevent IndexError. Check if the last element contains the expected "_i.h5"
substring before splitting; if not, handle it gracefully or raise an error.
Sanitize the extracted folder name to remove or neutralize any path traversal
characters like "../" to avoid security risks. Use a safe method to construct
the folder path and then create the directory with os.makedirs as before.

submit_kwargs = {
"command": " ".join(command),
"dependency_list": [str(qid) for qid in task_dependent_lst],
Expand Down
14 changes: 14 additions & 0 deletions tests/test_fluxclusterexecutor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,20 @@ def test_executor(self):
cloudpickle_register(ind=1)
fs1 = exe.submit(mpi_funct, 1)
self.assertFalse(fs1.done())
self.assertEqual(len(os.listdir("executorlib_cache")), 2)
self.assertEqual(fs1.result(), [(1, 2, 0), (1, 2, 1)])
self.assertTrue(fs1.done())

def test_executor_no_cwd(self):
with FluxClusterExecutor(
resource_dict={"cores": 2},
block_allocation=False,
cache_directory="executorlib_cache",
) as exe:
cloudpickle_register(ind=1)
fs1 = exe.submit(mpi_funct, 1)
self.assertFalse(fs1.done())
self.assertEqual(len(os.listdir("executorlib_cache")), 2)
self.assertEqual(fs1.result(), [(1, 2, 0), (1, 2, 1)])
self.assertTrue(fs1.done())
Comment on lines +47 to 58

@coderabbitai coderabbitai Bot Jul 2, 2025

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Verification agent

🧩 Analysis chain

Test may not correctly validate the separate working directory behavior.

The new test method is good but has a few concerns:

  1. Directory structure validation: Both tests check len(os.listdir("executorlib_cache")) == 2, but if the new logic creates subdirectories when cwd is not specified, shouldn't the directory structure be different between the two tests?

  2. Single job limitation: The test only submits one job, but the PR objective mentions creating "separate working directories for each job". Consider adding a test that submits multiple jobs to verify they get separate directories.

Let me verify the expected behavior by checking the implementation:

Consider adding a test that submits multiple jobs to better validate the separate directory functionality:

+    def test_executor_multiple_jobs_no_cwd(self):
+        with FluxClusterExecutor(
+            resource_dict={"cores": 2},
+            block_allocation=False,
+            cache_directory="executorlib_cache",
+        ) as exe:
+            cloudpickle_register(ind=1)
+            fs1 = exe.submit(mpi_funct, 1)
+            fs2 = exe.submit(mpi_funct, 2)
+            
+            # Verify both jobs complete successfully
+            self.assertEqual(fs1.result(), [(1, 2, 0), (1, 2, 1)])
+            self.assertEqual(fs2.result(), [(2, 2, 0), (2, 2, 1)])
+            
+            # Verify separate directories were created
+            cache_contents = os.listdir("executorlib_cache")
+            self.assertGreater(len(cache_contents), 2, "Expected separate directories for each job")

🏁 Script executed:

#!/bin/bash
# Description: Check how the working directory logic works when cwd is not specified
# Expected: Find the logic that creates subdirectories based on command arguments

ast-grep --pattern 'def execute_with_pysqa($$$) {
  $$$
}'

# Also check for any directory creation logic
rg -A 10 -B 5 "cache_directory.*cwd|cwd.*cache_directory"

Length of output: 2974

Enhance directory‐structure assertions and add multi‐job coverage

The queue_spawner now creates a separate subdirectory under executorlib_cache for each task when no cwd is specified, whereas with an explicit cwd all output lands directly in that directory. The existing test_executor_no_cwd only checks that there are two entries, but doesn’t verify they are directories—and the “with cwd” test likewise only checks the count, not that they’re files. Also, we should submit more than one job to ensure each job still gets its own workspace.

Please update tests/test_fluxclusterexecutor.py around lines 47–58 as follows:

• In test_executor_no_cwd, after listing executorlib_cache, assert that each entry is a directory:

entries = os.listdir("executorlib_cache")
self.assertEqual(len(entries), 2)
for e in entries:
    self.assertTrue(
        os.path.isdir(os.path.join("executorlib_cache", e)),
        f"Expected '{e}' to be a subdirectory for a task"
    )

• In the corresponding “with cwd” test, assert that entries are files, not directories:

entries = os.listdir("executorlib_cache")
self.assertEqual(len(entries), 2)
for e in entries:
    self.assertTrue(
        os.path.isfile(os.path.join("executorlib_cache", e)),
        f"Expected '{e}' to be a file in the shared cwd"
    )

• Add a new test to submit two separate jobs and verify each gets its own subdirectory:

+    def test_executor_multiple_jobs_no_cwd(self):
+        with FluxClusterExecutor(
+            resource_dict={"cores": 2},
+            block_allocation=False,
+            cache_directory="executorlib_cache",
+        ) as exe:
+            cloudpickle_register(ind=1)
+            # Submit two distinct jobs
+            fs1 = exe.submit(mpi_funct, 1)
+            fs2 = exe.submit(mpi_funct, 2)
+
+            # Ensure both complete correctly
+            self.assertEqual(fs1.result(), [(1, 2, 0), (1, 2, 1)])
+            self.assertEqual(fs2.result(), [(2, 2, 0), (2, 2, 1)])
+
+            # Verify separate subdirectories for each task of each job
+            entries = os.listdir("executorlib_cache")
+            # Expect 4 subdirectories: 2 tasks × 2 jobs
+            self.assertEqual(len(entries), 4)
+            for e in entries:
+                path = os.path.join("executorlib_cache", e)
+                self.assertTrue(os.path.isdir(path), f"{path} should be a directory")

This will ensure both the per‐task directory logic and multi‐job isolation are properly validated.

🤖 Prompt for AI Agents 
In tests/test_fluxclusterexecutor.py around lines 47 to 58, enhance the
test_executor_no_cwd by asserting that each entry in the "executorlib_cache"
directory is a subdirectory, not just checking the count. Similarly, in the
corresponding "with cwd" test, assert that entries are files rather than
directories. Additionally, add a new test that submits two separate jobs and
verifies each job gets its own subdirectory under "executorlib_cache" to ensure
multi-job workspace isolation is properly tested.


Expand Down
Loading