fix(process_registry): drop OOMPolicy=kill from systemd scope units

[aravindtri]

Problem

On Ubuntu 20.04 (systemd 245), cron jobs silently never run. The availability probe for restart-safe dispatch fails with:

Unknown assignment: OOMPolicy=kill
systemd-run --user --scope is unavailable

systemd <250 rejects OOMPolicy=kill on --scope units.

Fix

Remove OOMPolicy=kill from _systemd_scope_argv. MemoryMax + MemoryAccounting still bound the worker cgroup, and the kernel OOM killer still applies — no memory-limit regression.

Verification

• systemd 245 (Ubuntu 20.04): scope spawn + probe now succeed; cron jobs dispatch.
• Coverage: single code path shared by probe and real spawns.

[crazyief]

Please use GitHub Commit suggestion (not copy-paste / amend). That records Co-authored-by: crazyief.

Suggestion: add unit tests for the OOMPolicy removal

The PR removes OOMPolicy=kill from _systemd_scope_argv but has zero test coverage for this function. I verified these tests pass 4/4 on the lab against the PR worktree.

diff --git a/tests/tools/test_systemd_scope_argv.py b/tests/tools/test_systemd_scope_argv.py
new file mode 100644
--- /dev/null
+++ b/tests/tools/test_systemd_scope_argv.py
@@ -0,0 +1,38 @@
+"""Tests for _systemd_scope_argv OOMPolicy removal (#103762)."""
+
+from __future__ import annotations
+
+import pytest
+
+
+class TestSystemdScopeArgv:
+    """_systemd_scope_argv must produce valid systemd-run argv without
+    OOMPolicy=kill (systemd <250 rejects it on scope units)."""
+
+    def test_oom_policy_not_present(self):
+        from tools.process_registry import _systemd_scope_argv
+        argv = _systemd_scope_argv("systemd-run", "test-unit", "echo", "hello")
+        oom_flags = [a for a in argv if "OOMPolicy" in a]
+        assert len(oom_flags) == 0
+
+    def test_memory_properties_still_present(self):
+        from tools.process_registry import _systemd_scope_argv
+        argv = _systemd_scope_argv("systemd-run", "test-unit", "echo", "hello")
+        assert any("MemoryMax" in a for a in argv)
+        assert any("MemoryAccounting" in a for a in argv)
+
+    def test_includes_collect_and_unit(self):
+        from tools.process_registry import _systemd_scope_argv
+        argv = _systemd_scope_argv("systemd-run", "test-unit", "echo", "hello")
+        assert "--collect" in argv
+        assert "--unit" in argv
+        assert "test-unit" in argv
+
+    def test_passes_through_command(self):
+        from tools.process_registry import _systemd_scope_argv
+        argv = _systemd_scope_argv("systemd-run", "test-unit", "echo", "hello")
+        assert "echo" in argv
+        assert "hello" in argv
+```

[alt-glitch]

This was generated by AI during triage.

Duplicate of #102357, which makes the same change (drop OOMPolicy=kill from the systemd-run --scope argv in tools/process_registry.py) for #102486. #102508 and #102655 are also open for this cluster; a maintainer should pick one.

[kshitijk4poor]

Thanks @aravindtri — the Ubuntu 20.04 / systemd 245 evidence here is useful: it shows transient --scope units reject OOMPolicy=kill well below the v253 boundary some of the sibling PRs assumed, which is what settled the cluster on "drop the property" rather than "version-gate it".

Closing as a duplicate of #102357 (@gkd2323c, opened 2026-09-03), which makes the same change — remove OOMPolicy=kill from both the availability probe and _build_systemd_scope_argv in tools/process_registry.py — and is the earliest submission for #102486. #102357 is the one being reviewed and salvaged for the cluster; your systemd-245 data point is being cited in its landing notes. If #102357 does not land for some reason this will be reopened rather than lost.