auto update: restart instead of stop+start

Commit f131eaa74a changed restart to a stop+start motivated by
comments in the systemd man pages that restart behaves different than
stop+start, for instance, that it keeps certain resources open and
treats timers differently.  Yet, the actually fix for #17607 in the very
same commit was dealing with an ENOENT of the CID file on container
removal.

As it turns out in in #18926, changing to stop+start regressed on
restarting dependencies when auto updating a systemd unit.  Hence, move
back to using restart to make sure that dependent systemd units are
restarted as well.

An alternative could be recommending to use `BindsTo=` in Quadlet files
but this seems less common than `Requires=` and hence more risky to
cause issues on user sites.

Fixes: #18926
Signed-off-by: Valentin Rothberg <vrothberg@redhat.com>
This commit is contained in:
Valentin Rothberg 2023-06-19 08:52:55 +02:00
parent feea6663ee
commit 47e0557d57
2 changed files with 31 additions and 36 deletions

View File

@ -325,16 +325,8 @@ func (t *task) rollbackImage() error {
// restartSystemdUnit restarts the systemd unit the container is running in.
func (u *updater) restartSystemdUnit(ctx context.Context, unit string) error {
if err := u.stopSystemdUnit(ctx, unit); err != nil {
return err
}
return u.startSystemdUnit(ctx, unit)
}
// startSystemdUnit starts the systemd unit the container is running in.
func (u *updater) startSystemdUnit(ctx context.Context, unit string) error {
restartChan := make(chan string)
if _, err := u.conn.StartUnitContext(ctx, unit, "replace", restartChan); err != nil {
if _, err := u.conn.RestartUnitContext(ctx, unit, "replace", restartChan); err != nil {
return err
}
@ -348,28 +340,7 @@ func (u *updater) startSystemdUnit(ctx context.Context, unit string) error {
return nil
default:
return fmt.Errorf("error starting systemd unit %q expected %q but received %q", unit, "done", result)
}
}
// stopSystemdUnit stop the systemd unit the container is running in.
func (u *updater) stopSystemdUnit(ctx context.Context, unit string) error {
restartChan := make(chan string)
if _, err := u.conn.StopUnitContext(ctx, unit, "replace", restartChan); err != nil {
return err
}
// Wait for the restart to finish and actually check if it was
// successful or not.
result := <-restartChan
switch result {
case "done":
logrus.Infof("Successfully stopped systemd unit %q", unit)
return nil
default:
return fmt.Errorf("error stopping systemd unit %q expected %q but received %q", unit, "done", result)
return fmt.Errorf("error restarting systemd unit %q expected %q but received %q", unit, "done", result)
}
}

View File

@ -53,6 +53,7 @@ function generate_service() {
local command=$3
local extraArgs=$4
local noTag=$5
local requires=$6
# Unless specified, set a default command.
if [[ -z "$command" ]]; then
@ -73,9 +74,14 @@ function generate_service() {
else
label=""
fi
if [[ -n "$requires" ]]; then
requires="--requires=$requires"
fi
run_podman create $extraArgs --name $cname $label $target_img $command
(cd $UNIT_DIR; run_podman generate systemd --new --files --name $cname)
(cd $UNIT_DIR; run_podman generate systemd --new --files --name $requires $cname)
echo "container-$cname" >> $SNAME_FILE
run_podman rm -t 0 -f $cname
@ -161,23 +167,41 @@ function _confirm_update() {
run_podman events --filter type=system --since $since --stream=false
is "$output" ""
# Generate two units. The first "parent" to be auto updated, the second
# "child" depends on/requires the "parent" and is expected to get restarted
# as well on auto updates (regression test for #18926).
generate_service alpine image
ctr_parent=$cname
_wait_service_ready container-$ctr_parent.service
generate_service alpine image "" "" "" "container-$ctr_parent.service"
ctr_child=$cname
_wait_service_ready container-$ctr_child.service
run_podman container inspect --format "{{.ID}}" $ctr_child
old_child_id=$output
_wait_service_ready container-$cname.service
since=$(date --iso-8601=seconds)
run_podman auto-update --dry-run --format "{{.Unit}},{{.Image}},{{.Updated}},{{.Policy}}"
is "$output" ".*container-$cname.service,quay.io/libpod/alpine:latest,pending,registry.*" "Image update is pending."
is "$output" ".*container-$ctr_parent.service,quay.io/libpod/alpine:latest,pending,registry.*" "Image update is pending."
run_podman events --filter type=system --since $since --stream=false
is "$output" ".* system auto-update"
since=$(date --iso-8601=seconds)
run_podman auto-update --rollback=false --format "{{.Unit}},{{.Image}},{{.Updated}},{{.Policy}}"
is "$output" "Trying to pull.*" "Image is updated."
is "$output" ".*container-$cname.service,quay.io/libpod/alpine:latest,true,registry.*" "Image is updated."
is "$output" ".*container-$ctr_parent.service,quay.io/libpod/alpine:latest,true,registry.*" "Image is updated."
run_podman events --filter type=system --since $since --stream=false
is "$output" ".* system auto-update"
_confirm_update $cname $ori_image
# Confirm that the update was successful and that the child container/unit
# has been restarted as well.
_confirm_update $ctr_parent $ori_image
run_podman container inspect --format "{{.ID}}" $ctr_child
assert "$output" != "$old_child_id" \
"child container/unit has not been restarted during update"
run_podman container inspect --format "{{.ID}}" $ctr_child
run_podman container inspect --format "{{.State.Status}}" $ctr_child
is "$output" "running" "child container is in running state"
}
@test "podman auto-update - label io.containers.autoupdate=image with rollback" {