Handle when child process is killed by OOM killer

CHANGELOG.md

@@ -14,6 +14,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Fixed
 
 * Fix a bug where `a_pixels_per_b_pixel` summary statistics where equal to the `b_pixels_per_a_pixel` statistics.
+* `collapse` will return exit code 137 when one of the child processes is killed by the system (e.g. because it is
+   to much memory). This allows e.g. Nextflow to retry the process with more memory automatically.
 
 ## [0.17.1] - 2024-05-27
 

src/pixelator/cli/collapse.py

@@ -3,6 +3,7 @@
 Copyright © 2022 Pixelgen Technologies AB.
 """
 
+import sys
 from collections import defaultdict
 from concurrent import futures
 from pathlib import Path
@@ -25,6 +26,38 @@
 )
 
 
+def _handle_errors(jobs, executor):
+    for job in jobs:
+        exception = job.exception()
+        if exception is None:
+            continue
+
+        logger.error(
+            "Found an issue in the process pool. Trying to determine what went wrong and set the correct exit code. Exception was: %s",
+            exception,
+        )
+        process_map = executor._processes
+        for pid in process_map.keys():
+            exit_code = process_map[pid].exitcode
+            if exit_code is not None and exit_code != 0:
+                logger.error(
+                    "The child process in the process pool returned a non-zero exit code: %s.",
+                    exit_code,
+                )
+                # If we have an out of memory exception, make sure we exit with that.
+                if abs(exit_code) == 9:
+                    logger.error(
+                        "One of the child processes was killed (exit code: 9). Usually this is caused "
+                        "by a child process using to much memory. We will return an exit code of 137 "
+                        "to indicate that the process was killed by the out of memory killer."
+                    )
+                    sys.exit(137)
+        logger.error(
+            "Was unable to determine what when wrong in process pool. Will raise original exception."
+        )
+        raise exception
+
+
 @click.command(
     "collapse",
     short_help=(
@@ -238,12 +271,12 @@ def collapse(
                             min_count=min_count,
                         )
                     )
+            jobs = list(futures.as_completed(jobs))
+            _handle_errors(jobs, executor)
 
         total_input_reads = 0
         tmp_files = []
-        for job in futures.as_completed(jobs):
-            if job.exception() is not None:
-                raise job.exception()
+        for job in jobs:
             # the worker returns a path to a file (temp antibody edge list)
             tmp_file, input_reads_count = job.result()
             if tmp_file is not None:

src/pixelator/logging.py

@@ -214,6 +214,11 @@ def log_exception(exc_type, exc_value, traceback_obj):
                 # and thus we ignore them here.
                 return False
 
+            if issubclass(exc_type, SystemExit):
+                # SystemExit is raised when the application has been explicitly
+                # directed to exit, so we don't what a trace dumped for that.
+                return False
+
             self._root_logger.critical(
                 "Unhandled exception of type: {}".format(exc_type.__name__)
             )