48 lines
1.6 KiB
Bash
48 lines
1.6 KiB
Bash
|
#!/usr/bin/env bash
|
||
|
|
# heavy-prefix.sh — containment helpers for batch/long-running workloads.
|
||
|
|
#
|
||
|
|
# Sourcing this file has no side effects beyond defining functions.
|
||
|
|
#
|
||
|
|
# Purpose: keep heavy godot/python workloads (RL training, autoplay batches,
|
||
|
|
# proof renders, claude-player workers) inside the heavy-tests.slice cgroup
|
||
|
|
# so they cannot starve sshd / interactive work on apricot. Background: on
|
||
|
|
# 2026-05-18 / 2026-05-19 the box wedged when ~3000 godot workers spawned
|
||
|
|
# outside any cgroup; CPUWeight=20 on the slice lets sshd preempt.
|
||
|
|
|
||
|
|
_heavy_have_systemd_run() {
|
||
|
|
[[ "$(uname -s)" == "Linux" ]] && command -v systemd-run >/dev/null 2>&1
|
||
|
|
}
|
||
|
|
|
||
|
|
# Replace the current shell with the given command, wrapped in a transient
|
||
|
|
# scope under heavy-tests.slice. Falls back to a direct exec on non-Linux.
|
||
|
|
# Args: <unit-name> <cmd> [args...]
|
||
|
|
heavy_exec() {
|
||
|
|
local unit="${1:?heavy_exec: unit name required}"
|
||
|
|
shift
|
||
|
|
if _heavy_have_systemd_run; then
|
||
|
|
exec systemd-run --user \
|
||
|
|
--slice=heavy-tests.slice \
|
||
|
|
--scope --quiet --collect \
|
||
|
|
--unit="${unit}" \
|
||
|
|
-- nice -n 10 ionice -c 3 "$@"
|
||
|
|
fi
|
||
|
|
exec "$@"
|
||
|
|
}
|
||
|
|
|
||
|
|
# Start the given command as a detached transient .service under the slice.
|
||
|
|
# Returns immediately; logs go to the journal (journalctl --user -u <unit>).
|
||
|
|
# Args: <unit-name> <cmd> [args...]
|
||
|
|
heavy_service() {
|
||
|
|
local unit="${1:?heavy_service: unit name required}"
|
||
|
|
shift
|
||
|
|
if _heavy_have_systemd_run; then
|
||
|
|
systemd-run --user \
|
||
|
|
--slice=heavy-tests.slice \
|
||
|
|
--unit="${unit}" \
|
||
|
|
--collect --quiet \
|
||
|
|
-- "$@"
|
||
|
|
return $?
|
||
|
|
fi
|
||
|
|
nohup "$@" >/dev/null 2>&1 &
|
||
|
|
}
|