mgr/cephadm: skip keepalive deployment until haproxy has been deployed

We need the haproxy daemons in order to be able to populate
the keepalive conf with targets for keepalive to check
the health of. Before deploying in parallel, this worked
as we would do prepare_create and deployment for each
daemon sequentially and would deploy haproxy before keepalived.
Now that we're doing all the prepare_create steps first
and then deploying in parallel, keepalived's prepare_create
will fail unless we've already  deployed haproxy daemons
for that service.

Signed-off-by: Adam King <[email protected]>
(cherry picked from commit 96ddd28)

Resolves: rhbz#2372821

Author: adk3798

src/pybind/mgr/cephadm/serve.py

@@ -1129,6 +1129,17 @@ def prep_daemon_specs_for_creation_by_service(self, spec: ServiceSpec, slots_to_
             if service_type == 'rgw':
                 self.mgr.rgw_service.set_realm_zg_zone(cast(RGWSpec, spec))
 
+            if service_type == 'ingress':
+                ingress_spec = cast(IngressSpec, spec)
+                if slot.daemon_type == 'keepalived' and not ingress_spec.keepalive_only:
+                    if not self.mgr.cache.get_daemons_by_service(spec.service_name()):
+                        # prepare_create for keepalive will fail in this case as
+                        # we need to deploy haproxy daemons first in order to have targets
+                        # for keepalive to check the health of. For that reason,
+                        # we'll skip the keepalive daemon(s) and deploy them once
+                        # the serve loop comes back around after deploying haproxy
+                        continue
+
             # deploy new daemon
             daemon_id = slot.name