mgr/orch: Fix DaemonDescription().daemon_id() for RGWs

Turns out, users put dots into their RGW service names.

Fixes: https://tracker.ceph.com/issues/44934
Signed-off-by: Sebastian Wagner <sebastian.wagner@suse.com>

diff --git a/src/pybind/mgr/cephadm/tests/test_spec.py b/src/pybind/mgr/cephadm/tests/test_spec.py
new file mode 100644 (file)
index 0000000..de5eb15
--- /dev/null
+++ b/src/pybind/mgr/cephadm/tests/test_spec.py
@@ -0,0 +1,290 @@
+import json
+
+import pytest
+
+from ceph.deployment.service_spec import ServiceSpec, RGWSpec, PlacementSpec
+from orchestrator import DaemonDescription, OrchestratorError
+
+
+def test_spec_octopus():
+    # https://tracker.ceph.com/issues/44934
+    # Those are real user data from early octopus.
+    # Please do not modify those JSON values.
+    specs_text = """[
+{
+  "placement": {
+    "count": 1
+  },
+  "service_type": "alertmanager"
+},
+{
+  "placement": {
+    "host_pattern": "*"
+  },
+  "service_type": "crash"
+},
+{
+  "placement": {
+    "count": 1
+  },
+  "service_type": "grafana"
+},
+{
+  "placement": {
+    "count": 2
+  },
+  "service_type": "mgr"
+},
+{
+  "placement": {
+    "count": 5
+  },
+  "service_type": "mon"
+},
+{
+  "placement": {
+    "host_pattern": "*"
+  },
+  "service_type": "node-exporter"
+},
+{
+  "placement": {
+    "count": 1
+  },
+  "service_type": "prometheus"
+},
+{
+  "placement": {
+    "hosts": [
+      {
+        "hostname": "ceph-001",
+        "network": "",
+        "name": ""
+      }
+    ]
+  },
+  "service_type": "rgw",
+  "service_id": "default-rgw-realm.eu-central-1.1",
+  "rgw_realm": "default-rgw-realm",
+  "rgw_zone": "eu-central-1",
+  "subcluster": "1"
+}
+]
+"""
+    dds_text = """[
+    {
+        "hostname": "ceph-001",
+        "container_id": "d94d7969094d",
+        "container_image_id": "0881eb8f169f5556a292b4e2c01d683172b12830a62a9225a98a8e206bb734f0",
+        "container_image_name": "docker.io/prom/alertmanager:latest",
+        "daemon_id": "ceph-001",
+        "daemon_type": "alertmanager",
+        "version": "0.20.0",
+        "status": 1,
+        "status_desc": "running",
+        "last_refresh": "2020-04-03T15:31:48.725856",
+        "created": "2020-04-02T19:23:08.829543",
+        "started": "2020-04-03T07:29:16.932838" 
+    },
+    {
+        "hostname": "ceph-001",
+        "container_id": "c4b036202241",
+        "container_image_id": "204a01f9b0b6710dd0c0af7f37ce7139c47ff0f0105d778d7104c69282dfbbf1",
+        "container_image_name": "docker.io/ceph/ceph:v15",
+        "daemon_id": "ceph-001",
+        "daemon_type": "crash",
+        "version": "15.2.0",
+        "status": 1,
+        "status_desc": "running",
+        "last_refresh": "2020-04-03T15:31:48.725903",
+        "created": "2020-04-02T19:23:11.390694",
+        "started": "2020-04-03T07:29:16.910897" 
+    },
+    {
+        "hostname": "ceph-001",
+        "container_id": "5b7b94b48f31",
+        "container_image_id": "87a51ecf0b1c9a7b187b21c1b071425dafea0d765a96d5bc371c791169b3d7f4",
+        "container_image_name": "docker.io/ceph/ceph-grafana:latest",
+        "daemon_id": "ceph-001",
+        "daemon_type": "grafana",
+        "version": "6.6.2",
+        "status": 1,
+        "status_desc": "running",
+        "last_refresh": "2020-04-03T15:31:48.725950",
+        "created": "2020-04-02T19:23:52.025088",
+        "started": "2020-04-03T07:29:16.847972" 
+    },
+    {
+        "hostname": "ceph-001",
+        "container_id": "9ca007280456",
+        "container_image_id": "204a01f9b0b6710dd0c0af7f37ce7139c47ff0f0105d778d7104c69282dfbbf1",
+        "container_image_name": "docker.io/ceph/ceph:v15",
+        "daemon_id": "ceph-001.gkjwqp",
+        "daemon_type": "mgr",
+        "version": "15.2.0",
+        "status": 1,
+        "status_desc": "running",
+        "last_refresh": "2020-04-03T15:31:48.725807",
+        "created": "2020-04-02T19:22:18.648584",
+        "started": "2020-04-03T07:29:16.856153" 
+    },
+    {
+        "hostname": "ceph-001",
+        "container_id": "3d1ba9a2b697",
+        "container_image_id": "204a01f9b0b6710dd0c0af7f37ce7139c47ff0f0105d778d7104c69282dfbbf1",
+        "container_image_name": "docker.io/ceph/ceph:v15",
+        "daemon_id": "ceph-001",
+        "daemon_type": "mon",
+        "version": "15.2.0",
+        "status": 1,
+        "status_desc": "running",
+        "last_refresh": "2020-04-03T15:31:48.725715",
+        "created": "2020-04-02T19:22:13.863300",
+        "started": "2020-04-03T07:29:17.206024" 
+    },
+    {
+        "hostname": "ceph-001",
+        "container_id": "36d026c68ba1",
+        "container_image_id": "e5a616e4b9cf68dfcad7782b78e118be4310022e874d52da85c55923fb615f87",
+        "container_image_name": "docker.io/prom/node-exporter:latest",
+        "daemon_id": "ceph-001",
+        "daemon_type": "node-exporter",
+        "version": "0.18.1",
+        "status": 1,
+        "status_desc": "running",
+        "last_refresh": "2020-04-03T15:31:48.725996",
+        "created": "2020-04-02T19:23:53.880197",
+        "started": "2020-04-03T07:29:16.880044" 
+    },
+    {
+        "hostname": "ceph-001",
+        "container_id": "faf76193cbfe",
+        "container_image_id": "204a01f9b0b6710dd0c0af7f37ce7139c47ff0f0105d778d7104c69282dfbbf1",
+        "container_image_name": "docker.io/ceph/ceph:v15",
+        "daemon_id": "0",
+        "daemon_type": "osd",
+        "version": "15.2.0",
+        "status": 1,
+        "status_desc": "running",
+        "last_refresh": "2020-04-03T15:31:48.726088",
+        "created": "2020-04-02T20:35:02.991435",
+        "started": "2020-04-03T07:29:19.373956" 
+    },
+    {
+        "hostname": "ceph-001",
+        "container_id": "f82505bae0f1",
+        "container_image_id": "204a01f9b0b6710dd0c0af7f37ce7139c47ff0f0105d778d7104c69282dfbbf1",
+        "container_image_name": "docker.io/ceph/ceph:v15",
+        "daemon_id": "1",
+        "daemon_type": "osd",
+        "version": "15.2.0",
+        "status": 1,
+        "status_desc": "running",
+        "last_refresh": "2020-04-03T15:31:48.726134",
+        "created": "2020-04-02T20:35:17.142272",
+        "started": "2020-04-03T07:29:19.374002" 
+    },
+    {
+        "hostname": "ceph-001",
+        "container_id": "2708d84cd484",
+        "container_image_id": "358a0d2395fe711bb8258e8fb4b2d7865c0a9a6463969bcd1452ee8869ea6653",
+        "container_image_name": "docker.io/prom/prometheus:latest",
+        "daemon_id": "ceph-001",
+        "daemon_type": "prometheus",
+        "version": "2.17.1",
+        "status": 1,
+        "status_desc": "running",
+        "last_refresh": "2020-04-03T15:31:48.726042",
+        "created": "2020-04-02T19:24:10.281163",
+        "started": "2020-04-03T07:29:16.926292" 
+    },
+    {
+        "hostname": "ceph-001",
+        "daemon_id": "default-rgw-realm.eu-central-1.1.ceph-001.ytywjo",
+        "daemon_type": "rgw",
+        "status": 1,
+        "status_desc": "starting" 
+    }
+]"""
+    specs_json = json.loads(specs_text)
+    dds_json = json.loads(dds_text)
+    specs = [ServiceSpec.from_json(j) for j in specs_json]
+    dds = [DaemonDescription.from_json(j) for j in dds_json]
+
+    # just some verification that we can sill read old octopus specs
+    def remove_service_name(j):
+        if 'service_name' in j:
+            j_c = j.copy()
+            del j_c['service_name']
+            return j_c
+        return j
+    assert specs_json == [remove_service_name(s.to_json()) for s in specs]
+    assert dds_json == [d.to_json() for d in dds]
+
+
+@pytest.mark.parametrize("spec,dd,valid",
+[
+    (
+        # https://tracker.ceph.com/issues/44934
+        RGWSpec(
+            rgw_realm="default-rgw-realm",
+            rgw_zone="eu-central-1",
+            subcluster='1',
+        ),
+        DaemonDescription(
+            daemon_type='rgw',
+            daemon_id="default-rgw-realm.eu-central-1.1.ceph-001.ytywjo",
+            hostname="ceph-001",
+        ),
+        True
+    ),
+    (
+        # no subcluster
+        RGWSpec(
+            rgw_realm="default-rgw-realm",
+            rgw_zone="eu-central-1",
+        ),
+        DaemonDescription(
+            daemon_type='rgw',
+            daemon_id="default-rgw-realm.eu-central-1.ceph-001.ytywjo",
+            hostname="ceph-001",
+        ),
+        True
+    ),
+    (
+        # with tld
+        RGWSpec(
+            rgw_realm="default-rgw-realm",
+            rgw_zone="eu-central-1",
+            subcluster='1',
+        ),
+        DaemonDescription(
+            daemon_type='rgw',
+            daemon_id="default-rgw-realm.eu-central-1.1.host.domain.tld.ytywjo",
+            hostname="host.domain.tld",
+        ),
+        True
+    ),
+    (
+        # without host
+        RGWSpec(
+            service_type='rgw',
+            rgw_realm="default-rgw-realm",
+            rgw_zone="eu-central-1",
+            subcluster='1',
+        ),
+        DaemonDescription(
+            daemon_type='rgw',
+            daemon_id="default-rgw-realm.eu-central-1.1.hostname.ytywjo",
+            hostname=None,
+        ),
+        False
+    ),
+])
+def test_rgw_service_name(spec: RGWSpec, dd: DaemonDescription, valid):
+    if valid:
+        assert spec.service_name() == dd.service_name()
+    else:
+        with pytest.raises(OrchestratorError):
+            dd.service_name()
+

diff --git a/src/pybind/mgr/orchestrator/_interface.py b/src/pybind/mgr/orchestrator/_interface.py
index a02572a074e3c3ad7395156b73fd6a52846b80bc..d5bf3a62a81d3e41920e0bfa68435b5fd056ab3c 100644 (file)
--- a/src/pybind/mgr/orchestrator/_interface.py
+++ b/src/pybind/mgr/orchestrator/_interface.py
@@ -1280,8 +1280,16 @@ class DaemonDescription(object):
 
     def service_id(self):
         if self.daemon_type == 'rgw':
-            v = self.daemon_id.split('.')
-            return '.'.join(v[0:2])
+            if self.hostname and self.hostname in self.daemon_id:
+                pre, post_ = self.daemon_id.split(self.hostname)
+                return pre[:-1]
+            else:
+                # daemon_id == "realm.zone.host.random"
+                v = self.daemon_id.split('.')
+                if len(v) == 4:
+                    return '.'.join(v[0:2])
+                # subcluster or fqdn? undecidable.
+                raise OrchestratorError(f"DaemonDescription: Cannot calculate service_id: {v}")
         if self.daemon_type in ['mds', 'nfs', 'iscsi']:
             return self.daemon_id.split('.')[0]
         return self.daemon_type