Merge pull request ceph#45786 from adk3798/staggered-upgrade

mgr/cephadm: staggered upgrade

Reviewed-by: Anthony D'Atri <[email protected]>
Reviewed-by: Redouane Kachach <[email protected]>

doc/cephadm/upgrade.rst

@@ -188,3 +188,100 @@ you need. For example, the following command upgrades to a development build:
   ceph orch upgrade start --image quay.io/ceph-ci/ceph:recent-git-branch-name
 
 For more information about available container images, see :ref:`containers`.
+
+Staggered Upgrade
+=================
+
+Some users may prefer to upgrade components in phases rather than all at once.
+The upgrade command, starting in 16.2.10 and 17.2.1 allows parameters
+to limit which daemons are upgraded by a single upgrade command. The options in
+include ``daemon_types``, ``services``, ``hosts`` and ``limit``. ``daemon_types``
+takes a comma-separated list of daemon types and will only upgrade daemons of those
+types. ``services`` is mutually exclusive with ``daemon_types``, only takes services
+of one type at a time (e.g. can't provide an OSD and RGW service at the same time), and
+will only upgrade daemons belonging to those services. ``hosts`` can be combined
+with ``daemon_types`` or ``services`` or provided on its own. The ``hosts`` parameter
+follows the same format as the command line options for :ref:`orchestrator-cli-placement-spec`.
+``limit`` takes an integer > 0 and provides a numerical limit on the number of
+daemons cephadm will upgrade. ``limit`` can be combined with any of the other
+parameters. For example, if you specify to upgrade daemons of type osd on host
+Host1 with ``limit`` set to 3, cephadm will upgrade (up to) 3 osd daemons on
+Host1.
+
+Example: specifying daemon types and hosts:
+
+.. prompt:: bash #
+
+  ceph orch upgrade start --image <image-name> --daemon-types mgr,mon --hosts host1,host2
+
+Example: specifying services and using limit:
+
+.. prompt:: bash #
+
+  ceph orch upgrade start --image <image-name> --services rgw.example1,rgw.example2 --limit 2
+
+.. note::
+
+   Cephadm strictly enforces an order to the upgrade of daemons that is still present
+   in staggered upgrade scenarios. The current upgrade ordering is
+   ``mgr -> mon -> crash -> osd -> mds -> rgw -> rbd-mirror -> cephfs-mirror -> iscsi -> nfs``.
+   If you specify parameters that would upgrade daemons out of order, the upgrade
+   command will block and note which daemons will be missed if you proceed.
+
+.. note::
+
+  Upgrade commands with limiting parameters will validate the options before beginning the
+  upgrade, which may require pulling the new container image. Do not be surprised
+  if the upgrade start command takes a while to return when limiting parameters are provided.
+
+.. note::
+
+   In staggered upgrade scenarios (when a limiting parameter is provided) monitoring
+   stack daemons including Prometheus and node-exporter are refreshed after the Manager
+   daemons have been upgraded. Do not be surprised if Manager upgrades thus take longer
+   than expected. Note that the versions of monitoring stack daemons may not change between
+   Ceph releases, in which case they are only redeployed.
+
+Upgrading to a version that supports staggered upgrade from one that doesn't
+----------------------------------------------------------------------------
+
+While upgrading from a version that already supports staggered upgrades the process
+simply requires providing the necessary arguments. However, if you wish to upgrade
+to a version that supports staggered upgrade from one that does not, there is a
+workaround. It requires first manually upgrading the Manager daemons and then passing
+the limiting parameters as usual.
+
+.. warning::
+  Make sure you have multiple running mgr daemons before attempting this procedure.
+
+To start with, determine which Manager is your active one and which are standby. This
+can be done in a variety of ways such as looking at the ``ceph -s`` output. Then,
+manually upgrade each standby mgr daemon with:
+
+.. prompt:: bash #
+
+  ceph orch daemon redeploy mgr.example1.abcdef --image <new-image-name>
+
+.. note::
+
+   If you are on a very early version of cephadm (early Octopus) the ``orch daemon redeploy``
+   command may not have the ``--image`` flag. In that case, you must manually set the
+   Manager container image ``ceph config set mgr container_image <new-image-name>`` and then
+   redeploy the Manager ``ceph orch daemon redeploy mgr.example1.abcdef``
+
+At this point, a Manager fail over should allow us to have the active Manager be one
+running the new version.
+
+.. prompt:: bash #
+
+  ceph mgr fail
+
+Verify the active Manager is now one running the new version. To complete the Manager
+upgrading:
+
+.. prompt:: bash #
+
+  ceph orch upgrade start --image <new-image-name> --daemon-types mgr
+
+You should now have all your Manager daemons on the new version and be able to
+specify the limiting parameters for the rest of the upgrade.

qa/suites/orch/cephadm/upgrade/3-upgrade/.qa

@@ -0,0 +1 @@
+../.qa/

qa/suites/orch/cephadm/upgrade/3-upgrade/staggered.yaml

@@ -0,0 +1,111 @@
+tasks:
+- cephadm.shell:
+    env: [sha1]
+    mon.a:
+      - radosgw-admin realm create --rgw-realm=r --default
+      - radosgw-admin zonegroup create --rgw-zonegroup=default --master --default
+      - radosgw-admin zone create --rgw-zonegroup=default --rgw-zone=z --master --default
+      - radosgw-admin period update --rgw-realm=r --commit
+      - ceph orch apply rgw r z --placement=2 --port=8000
+      - sleep 180
+      - ceph config set mon mon_warn_on_insecure_global_id_reclaim false --force
+      - ceph config set mon mon_warn_on_insecure_global_id_reclaim_allowed false --force
+      - ceph config set global log_to_journald false --force
+      # get some good info on the state of things pre-upgrade. Useful for debugging
+      - ceph orch ps
+      - ceph versions
+      - ceph -s
+      - ceph orch ls
+      # doing staggered upgrade requires mgr daemons being on a version that contains the staggered upgrade code
+      # until there is a stable version that contains it, we can test by manually upgrading a mgr daemon
+      - ceph config set mgr container_image quay.ceph.io/ceph-ci/ceph:$sha1
+      - ceph orch daemon redeploy "mgr.$(ceph mgr dump -f json | jq .standbys | jq .[] | jq -r .name)"
+      - ceph orch ps --refresh
+      - sleep 180
+      # gather more possible debugging info
+      - ceph orch ps
+      - ceph versions
+      - ceph -s
+      # check that there are two different versions found for mgr daemon (which implies we upgraded one)
+      - ceph versions | jq -e '.mgr | length == 2'
+      - ceph mgr fail
+      - sleep 180
+      # now try upgrading the other mgr
+      # we should now have access to --image flag for the daemon redeploy command
+      - ceph orch daemon redeploy "mgr.$(ceph mgr dump -f json | jq .standbys | jq .[] | jq -r .name)" --image quay.ceph.io/ceph-ci/ceph:$sha1
+      - ceph orch ps --refresh
+      - sleep 180
+      # gather more possible debugging info
+      - ceph orch ps
+      - ceph versions
+      - ceph -s
+      - ceph mgr fail
+      - sleep 180
+      # gather more debugging info
+      - ceph orch ps
+      - ceph versions
+      - ceph -s
+      # now that both mgrs should have been redeployed with the new version, we should be back on only 1 version for the mgrs
+      - ceph versions | jq -e '.mgr | length == 1'
+      - ceph mgr fail
+      - sleep 180
+      # debugging info
+      - ceph orch ps
+      - ceph versions
+      # to make sure mgr daemons upgrade is fully completed, including being deployed by a mgr on new new version
+      # also serves as an early failure if manually upgrading the mgrs failed as --daemon-types won't be recognized
+      - ceph orch upgrade start --image quay.ceph.io/ceph-ci/ceph:$sha1 --daemon-types mgr
+      - while ceph orch upgrade status | jq '.in_progress' | grep true && ! ceph orch upgrade status | jq '.message' | grep Error ; do ceph orch ps ; ceph versions ; ceph orch upgrade status ; sleep 30 ; done
+      # verify only one version found for mgrs and that their version hash matches what we are upgrading to
+      - ceph versions | jq -e '.mgr | length == 1'
+      - ceph versions | jq -e '.mgr | keys' | grep $sha1
+      # verify overall we still se two versions, basically to make sure --daemon-types wans't ignored and all daemons upgraded
+      - ceph versions | jq -e '.overall | length == 2'
+      # check that exactly two daemons have been upgraded to the new image (our 2 mgr daemons)
+      - ceph orch upgrade check quay.ceph.io/ceph-ci/ceph:$sha1 | jq -e '.up_to_date | length == 2'
+      # upgrade only the mons on one of the two hosts
+      - ceph orch upgrade start --image quay.ceph.io/ceph-ci/ceph:$sha1 --daemon-types mon --hosts $(ceph orch ps | grep mgr.x | awk '{print $2}')
+      - while ceph orch upgrade status | jq '.in_progress' | grep true && ! ceph orch upgrade status | jq '.message' | grep Error ; do ceph orch ps ; ceph versions ; ceph orch upgrade status ; sleep 30 ; done
+      - ceph orch ps
+      # verify tow different version seen for mons
+      - ceph versions | jq -e '.mon | length == 2'
+      # upgrade mons on the other hosts
+      - ceph orch upgrade start --image quay.ceph.io/ceph-ci/ceph:$sha1 --daemon-types mon --hosts $(ceph orch ps | grep mgr.y | awk '{print $2}')
+      - while ceph orch upgrade status | jq '.in_progress' | grep true && ! ceph orch upgrade status | jq '.message' | grep Error ; do ceph orch ps ; ceph versions ; ceph orch upgrade status ; sleep 30 ; done
+      - ceph orch ps
+      # verify all mons now on same version and version hash matches what we are upgrading to
+      - ceph versions | jq -e '.mon | length == 1'
+      - ceph versions | jq -e '.mon | keys' | grep $sha1
+      # verify exactly 5 daemons are now upgraded (2 mgrs, 3 mons)
+      - ceph orch upgrade check quay.ceph.io/ceph-ci/ceph:$sha1 | jq -e '.up_to_date | length == 5'
+      # upgrade exactly 2 osd daemons
+      - ceph orch upgrade start --image quay.ceph.io/ceph-ci/ceph:$sha1 --daemon-types osd --limit 2
+      - while ceph orch upgrade status | jq '.in_progress' | grep true && ! ceph orch upgrade status | jq '.message' | grep Error ; do ceph orch ps ; ceph versions ; ceph orch upgrade status ; sleep 30 ; done
+      - ceph orch ps
+      # verify two different versions now seen for osds
+      - ceph versions | jq -e '.osd | length == 2'
+      # verify exactly 7 daemons have been upgraded (2 mgrs, 3 mons, 2 osds)
+      - ceph orch upgrade check quay.ceph.io/ceph-ci/ceph:$sha1 | jq -e '.up_to_date | length == 7'
+      # upgrade one more osd
+      - ceph orch upgrade start --image quay.ceph.io/ceph-ci/ceph:$sha1 --daemon-types crash,osd --limit 1
+      - while ceph orch upgrade status | jq '.in_progress' | grep true && ! ceph orch upgrade status | jq '.message' | grep Error ; do ceph orch ps ; ceph versions ; ceph orch upgrade status ; sleep 30 ; done
+      - ceph orch ps
+      - ceph versions | jq -e '.osd | length == 2'
+      # verify now 8 daemons have been upgraded
+      - ceph orch upgrade check quay.ceph.io/ceph-ci/ceph:$sha1 | jq -e '.up_to_date | length == 8'
+      # upgrade the rest of the osds
+      - ceph orch upgrade start --image quay.ceph.io/ceph-ci/ceph:$sha1 --daemon-types crash,osd
+      - while ceph orch upgrade status | jq '.in_progress' | grep true && ! ceph orch upgrade status | jq '.message' | grep Error ; do ceph orch ps ; ceph versions ; ceph orch upgrade status ; sleep 30 ; done
+      - ceph orch ps
+      # verify all osds are now on same version and version hash matches what we are upgrading to
+      - ceph versions | jq -e '.osd | length == 1'
+      - ceph versions | jq -e '.osd | keys' | grep $sha1
+      # upgrade the rgw daemons using --services
+      - ceph orch upgrade start --image quay.ceph.io/ceph-ci/ceph:$sha1 --services rgw.r.z
+      - while ceph orch upgrade status | jq '.in_progress' | grep true && ! ceph orch upgrade status | jq '.message' | grep Error ; do ceph orch ps ; ceph versions ; ceph orch upgrade status ; sleep 30 ; done
+      - ceph orch ps
+      # verify all rgw daemons on same version and version hash matches what we are upgrading to
+      - ceph versions | jq -e '.rgw | length == 1'
+      - ceph versions | jq -e '.rgw | keys' | grep $sha1
+      # run upgrade one more time with no filter parameters to make sure anything left gets upgraded
+      - ceph orch upgrade start --image quay.ceph.io/ceph-ci/ceph:$sha1

src/pybind/mgr/cephadm/module.py

@@ -60,7 +60,7 @@
 from .upgrade import CephadmUpgrade
 from .template import TemplateMgr
 from .utils import CEPH_IMAGE_TYPES, RESCHEDULE_FROM_OFFLINE_HOSTS_TYPES, forall_hosts, \
-    cephadmNoImage
+    cephadmNoImage, CEPH_UPGRADE_ORDER
 from .configchecks import CephadmConfigChecks
 from .offline_watcher import OfflineHostWatcher
 
@@ -2692,10 +2692,37 @@ def upgrade_ls(self, image: Optional[str], tags: bool, show_all_versions: Option
         return self.upgrade.upgrade_ls(image, tags, show_all_versions)
 
     @handle_orch_error
-    def upgrade_start(self, image: str, version: str) -> str:
+    def upgrade_start(self, image: str, version: str, daemon_types: Optional[List[str]] = None, host_placement: Optional[str] = None,
+                      services: Optional[List[str]] = None, limit: Optional[int] = None) -> str:
         if self.inventory.get_host_with_state("maintenance"):
             raise OrchestratorError("upgrade aborted - you have host(s) in maintenance state")
-        return self.upgrade.upgrade_start(image, version)
+        if daemon_types is not None and services is not None:
+            raise OrchestratorError('--daemon-types and --services are mutually exclusive')
+        if daemon_types is not None:
+            for dtype in daemon_types:
+                if dtype not in CEPH_UPGRADE_ORDER:
+                    raise OrchestratorError(f'Upgrade aborted - Got unexpected daemon type "{dtype}".\n'
+                                            f'Viable daemon types for this command are: {utils.CEPH_TYPES + utils.GATEWAY_TYPES}')
+        if services is not None:
+            for service in services:
+                if service not in self.spec_store:
+                    raise OrchestratorError(f'Upgrade aborted - Got unknown service name "{service}".\n'
+                                            f'Known services are: {self.spec_store.all_specs.keys()}')
+        hosts: Optional[List[str]] = None
+        if host_placement is not None:
+            all_hosts = list(self.inventory.all_specs())
+            placement = PlacementSpec.from_string(host_placement)
+            hosts = placement.filter_matching_hostspecs(all_hosts)
+            if not hosts:
+                raise OrchestratorError(
+                    f'Upgrade aborted - hosts parameter "{host_placement}" provided did not match any hosts')
+
+        if limit is not None:
+            if limit < 1:
+                raise OrchestratorError(
+                    f'Upgrade aborted - --limit arg must be a positive integer, not {limit}')
+
+        return self.upgrade.upgrade_start(image, version, daemon_types, hosts, services, limit)
 
     @handle_orch_error
     def upgrade_pause(self) -> str:

src/pybind/mgr/cephadm/services/osd.py

@@ -298,7 +298,8 @@ def generate_previews(self, osdspecs: List[DriveGroupSpec], for_host: str) -> Li
 
                 # driveselection for host
                 cmds: List[str] = self.driveselection_to_ceph_volume(ds,
-                                                                     osd_id_claims.filtered_by_host(host),
+                                                                     osd_id_claims.filtered_by_host(
+                                                                         host),
                                                                      preview=True)
                 if not cmds:
                     logger.debug("No data_devices, skipping DriveGroup: {}".format(

src/pybind/mgr/cephadm/tests/test_cephadm.py

@@ -166,9 +166,11 @@ def test_re_add_host_receive_loopback(self, resolve_ip, cephadm_module):
         resolve_ip.side_effect = ['192.168.122.1', '127.0.0.1', '127.0.0.1']
         assert wait(cephadm_module, cephadm_module.get_hosts()) == []
         cephadm_module._add_host(HostSpec('test', '192.168.122.1'))
-        assert wait(cephadm_module, cephadm_module.get_hosts()) == [HostSpec('test', '192.168.122.1')]
+        assert wait(cephadm_module, cephadm_module.get_hosts()) == [
+            HostSpec('test', '192.168.122.1')]
         cephadm_module._add_host(HostSpec('test'))
-        assert wait(cephadm_module, cephadm_module.get_hosts()) == [HostSpec('test', '192.168.122.1')]
+        assert wait(cephadm_module, cephadm_module.get_hosts()) == [
+            HostSpec('test', '192.168.122.1')]
         with pytest.raises(OrchestratorError):
             cephadm_module._add_host(HostSpec('test2'))
 
@@ -894,7 +896,8 @@ def test_driveselection_to_ceph_volume(self, cephadm_module, devices, preview, e
             ds = DriveSelection(dg, Devices([Device(path) for path in devices]))
             preview = preview
             out = cephadm_module.osd_service.driveselection_to_ceph_volume(ds, [], preview)
-            assert all(any(cmd in exp_cmd for exp_cmd in exp_commands) for cmd in out), f'Expected cmds from f{out} in {exp_commands}'
+            assert all(any(cmd in exp_cmd for exp_cmd in exp_commands)
+                       for cmd in out), f'Expected cmds from f{out} in {exp_commands}'
 
     @pytest.mark.parametrize(
         "devices, preview, exp_commands",
@@ -919,7 +922,8 @@ def test_raw_driveselection_to_ceph_volume(self, cephadm_module, devices, previe
             ds = DriveSelection(dg, Devices([Device(path) for path in devices]))
             preview = preview
             out = cephadm_module.osd_service.driveselection_to_ceph_volume(ds, [], preview)
-            assert all(any(cmd in exp_cmd for exp_cmd in exp_commands) for cmd in out), f'Expected cmds from f{out} in {exp_commands}'
+            assert all(any(cmd in exp_cmd for exp_cmd in exp_commands)
+                       for cmd in out), f'Expected cmds from f{out} in {exp_commands}'
 
     @mock.patch("cephadm.serve.CephadmServe._run_cephadm", _run_cephadm(
         json.dumps([