[Unit]
# Rootless podman drives containers through each account's own systemd manager
# (user@<uid>.service). Normally logind starts those at boot for every lingering
# account and nothing here is needed. On a host where `user@.service` is masked
# — CageFS 7.6.39+ does this deliberately, CloudLinux CLOS-4517 — logind cannot,
# so a lingering account comes up with no manager and its containers stay down
# until somebody happens to run an ea-podman command. This unit is that trigger.
# See EA4-319 and docs/container-shell-access.md.
Description=Start the per-user systemd managers ea-podman containers need
Documentation=https://webpros.atlassian.net/browse/EA4-319
# logind owns the linger markers this reads and the managers it starts, so it
# has to be up first. Ordering only: a box without logind has nothing to fix.
After=systemd-logind.service systemd-user-sessions.service
Wants=systemd-logind.service
# The CLI is compiled at install time (bin/compile.sh, run from %post), so on a
# half-installed box there is nothing to run.
ConditionPathExists=/opt/cpanel/ea-podman/bin/ea-podman
[Service]
Type=oneshot
# So `systemctl status` still shows the sweep ran, and it is not re-run on every
# `daemon-reload`.
RemainAfterExit=yes
ExecStart=/opt/cpanel/ea-podman/bin/ea-podman ensure_user_sessions --quiet
# A single account whose manager will not start must not hold the boot open: the
# sweep polls for at most ~10s in total, so anything near this ceiling means
# systemd itself is wedged, and failing is better than waiting forever. Nothing
# is ordered after this unit, so a failure here costs only the containers.
TimeoutStartSec=300
[Install]
WantedBy=multi-user.target
|