From d962cc96cd0a425e93d15ad7fd7aaac9c7a2ea28 Mon Sep 17 00:00:00 2001
From: Tobias Wolf <wolf@b1-systems.de>
Date: Mon, 4 Nov 2024 15:35:02 +0100
Subject: [PATCH] Add support to configure which systemd unit services or
 targets to use

Signed-off-by: Tobias Wolf <wolf@b1-systems.de>
---
 README.md                                | 10 +++---
 config.example.osism.yaml                |  1 -
 config.example.yaml                      |  6 ++++
 src/rookify/config.schema.yaml           |  8 +++++
 src/rookify/modules/ceph.py              | 43 ++++++++++++++++++++++++
 src/rookify/modules/migrate_mds/main.py  |  5 ++-
 src/rookify/modules/migrate_mgrs/main.py |  5 ++-
 src/rookify/modules/migrate_mons/main.py |  5 ++-
 src/rookify/modules/migrate_osds/main.py |  4 ++-
 src/rookify/modules/migrate_rgws/main.py |  5 ++-
 10 files changed, 82 insertions(+), 10 deletions(-)

diff --git a/README.md b/README.md
index 99656ec..8a0df87 100644
--- a/README.md
+++ b/README.md
@@ -103,17 +103,19 @@ OSD devices: Not analyzed yet
 
 #### Troubleshooting
 
-**ssh-issues:**
-- make sure the id-rsa keys are "clean" and do not contain unexpected strings like "\<\<EOF". E.g. call `ssh-keygen -p -N "" -f ssh.key` to convert and reformat the keyfile to the expected format.
-- allow direnv (`direnv allow`) to use `.envrc` or copy and execute the command from the file: this switches off the ssh-agent, which sometimes has too many keys loaded
+**missing Ceph systemd targets:**
+- please consult your distribution's documentation in case documented Ceph systemd unit services or targets are missing
 
 **frozen state:**
 - if the rookify process freezes, check your connections. In the OSISM testbed especially check the vpn-connection (in testbed repository try `make vpn-*`)
 
+**ssh-issues:**
+- make sure the id-rsa keys are "clean" and do not contain unexpected strings like "\<\<EOF". E.g. call `ssh-keygen -p -N "" -f ssh.key` to convert and reformat the keyfile to the expected format.
+- allow direnv (`direnv allow`) to use `.envrc` or copy and execute the command from the file: this switches off the ssh-agent, which sometimes has too many keys loaded
+
 ## Support
 
 For issues, questions, or contributions, please open an issue or pull request in the GitHub repository. We welcome community feedback and contributions to enhance rookify.
 
-
 ## License
 This project is licensed under the Apache 2.0 License - see the LICENSE file for details.
diff --git a/config.example.osism.yaml b/config.example.osism.yaml
index 3159a83..cdd93ac 100644
--- a/config.example.osism.yaml
+++ b/config.example.osism.yaml
@@ -42,7 +42,6 @@ rook:
 
 migration_modules:
 - example
-- create_rook_cluster
 - migrate_osds
 - migrate_osd_pools
 - migrate_mds
diff --git a/config.example.yaml b/config.example.yaml
index 54ede15..2769b3f 100644
--- a/config.example.yaml
+++ b/config.example.yaml
@@ -10,6 +10,12 @@ logging:
 ceph:
   config: ./.ceph/ceph.conf
   keyring: ./.ceph/ceph.client.admin.keyring
+  systemd_file_name_templates:
+    mds: "ceph-{fsid}@mds.{host}.service"
+    mgr: "ceph-{fsid}@mgr.{host}.service"
+    mon: "ceph-{fsid}@mon.{host}.service"
+    osd: "ceph-{fsid}@osd.{host}.service"
+    rgw: "ceph-{fsid}@rgw.{host}.service"
 
 # fill in correct path to private key
 ssh:
diff --git a/src/rookify/config.schema.yaml b/src/rookify/config.schema.yaml
index ae30c41..8f9da4f 100644
--- a/src/rookify/config.schema.yaml
+++ b/src/rookify/config.schema.yaml
@@ -10,6 +10,7 @@ logging:
 ceph:
   config: str()
   keyring: str()
+  systemd_file_name_templates: include("systemd_file_name_templates", required=False)
 
 ssh:
   private_key: str()
@@ -39,3 +40,10 @@ migration_modules: list(str())
 ssh_host:
     address: ip()
     user: str()
+---
+systemd_file_name_templates:
+    mds: str(required=False)
+    mgr: str(required=False)
+    mon: str(required=False)
+    osd: str(required=False)
+    rgw: str(required=False)
diff --git a/src/rookify/modules/ceph.py b/src/rookify/modules/ceph.py
index ddc3ed2..5458e36 100644
--- a/src/rookify/modules/ceph.py
+++ b/src/rookify/modules/ceph.py
@@ -16,6 +16,13 @@ def __init__(self, config: Dict[str, Any]):
         except rados.ObjectNotFound as err:
             raise ModuleException(f"Could not connect to ceph: {err}")
 
+        status_data = self.mon_command("status")
+
+        self._fsid = status_data["fsid"]
+        self._systemd_file_name_templates = config.get(
+            "systemd_file_name_templates", {}
+        )
+
     def __getattr__(self, name: str) -> Any:
         return getattr(self.__ceph, name)
 
@@ -32,6 +39,42 @@ def _json_command(self, handler: Any, *args: Any) -> Any:
 
         return data
 
+    def get_systemd_mds_file_name(self, host: str) -> str:
+        return self._get_systemd_template_file_name(
+            self._systemd_file_name_templates.get("mds", "ceph-mds.target"),
+            host,
+        )
+
+    def get_systemd_mgr_file_name(self, host: str) -> str:
+        return self._get_systemd_template_file_name(
+            self._systemd_file_name_templates.get("mgr", "ceph-mgr.target"),
+            host,
+        )
+
+    def get_systemd_mon_file_name(self, host: str) -> str:
+        return self._get_systemd_template_file_name(
+            self._systemd_file_name_templates.get("mon", "ceph-mon.target"),
+            host,
+        )
+
+    def get_systemd_osd_file_name(self, host: str, osd_id: int) -> str:
+        file_name_template: str = self._systemd_file_name_templates.get(
+            "osd", "ceph-osd@{0:d}.service".format(osd_id)
+        )
+
+        return file_name_template.format(fsid=self._fsid, host=host, osd_id=osd_id)
+
+    def get_systemd_rgw_file_name(self, host: str) -> str:
+        return self._get_systemd_template_file_name(
+            self._systemd_file_name_templates.get("mon", "ceph-radosgw.target"),
+            host,
+        )
+
+    def _get_systemd_template_file_name(
+        self, file_name_template: str, host: str
+    ) -> str:
+        return file_name_template.format(fsid=self._fsid, host=host)
+
     def get_osd_pool_configurations_from_map(
         self, dump_data: Dict[str, Any]
     ) -> Dict[str, Any]:
diff --git a/src/rookify/modules/migrate_mds/main.py b/src/rookify/modules/migrate_mds/main.py
index 163bc56..d9f7694 100644
--- a/src/rookify/modules/migrate_mds/main.py
+++ b/src/rookify/modules/migrate_mds/main.py
@@ -83,7 +83,10 @@ def execute(self) -> None:
 
     def _disable_mds(self, mds_host: str) -> None:
         result = self.ssh.command(
-            mds_host, "sudo systemctl disable --now ceph-mds.target"
+            mds_host,
+            "sudo systemctl disable --now {0}".format(
+                self.ceph.get_systemd_mds_file_name(mds_host)
+            ),
         )
 
         if result.failed:
diff --git a/src/rookify/modules/migrate_mgrs/main.py b/src/rookify/modules/migrate_mgrs/main.py
index 8dce03e..9803975 100644
--- a/src/rookify/modules/migrate_mgrs/main.py
+++ b/src/rookify/modules/migrate_mgrs/main.py
@@ -44,7 +44,10 @@ def _migrate_mgr(self, mgr_host: str) -> None:
         self.logger.info("Migrating ceph-mgr daemon at host'{0}'".format(mgr_host))
 
         result = self.ssh.command(
-            mgr_host, "sudo systemctl disable --now ceph-mgr.target"
+            mgr_host,
+            "sudo systemctl disable --now {0}".format(
+                self.ceph.get_systemd_mgr_file_name(mgr_host)
+            ),
         )
 
         if result.failed:
diff --git a/src/rookify/modules/migrate_mons/main.py b/src/rookify/modules/migrate_mons/main.py
index 92e08a0..f3e425e 100644
--- a/src/rookify/modules/migrate_mons/main.py
+++ b/src/rookify/modules/migrate_mons/main.py
@@ -51,7 +51,10 @@ def _migrate_mon(self, mon: Dict[str, Any]) -> None:
         self.logger.info("Migrating ceph-mon daemon '{0}'".format(mon["name"]))
 
         result = self.ssh.command(
-            mon["name"], "sudo systemctl disable --now ceph-mon.target"
+            mon["name"],
+            "sudo systemctl disable --now {0}".format(
+                self.ceph.get_systemd_mon_file_name(mon["name"])
+            ),
         )
 
         if result.failed:
diff --git a/src/rookify/modules/migrate_osds/main.py b/src/rookify/modules/migrate_osds/main.py
index aa743d9..d2212b9 100644
--- a/src/rookify/modules/migrate_osds/main.py
+++ b/src/rookify/modules/migrate_osds/main.py
@@ -158,7 +158,9 @@ def migrate_osds(self, host: str, osd_ids: List[int]) -> None:
 
             result = self.ssh.command(
                 host,
-                "sudo systemctl disable --now ceph-osd@{0:d}.service".format(osd_id),
+                "sudo systemctl disable --now {0}".format(
+                    self.ceph.get_systemd_osd_file_name(host, osd_id)
+                ),
             )
 
             if result.failed:
diff --git a/src/rookify/modules/migrate_rgws/main.py b/src/rookify/modules/migrate_rgws/main.py
index fc11c55..aaa4e25 100644
--- a/src/rookify/modules/migrate_rgws/main.py
+++ b/src/rookify/modules/migrate_rgws/main.py
@@ -78,7 +78,10 @@ def _migrate_rgw(self, rgw_host: str) -> None:
 
         if is_migration_required:
             result = self.ssh.command(
-                rgw_host, "sudo systemctl disable --now ceph-radosgw.target"
+                rgw_host,
+                "sudo systemctl disable --now {0}".format(
+                    self.ceph.get_systemd_rgw_file_name(rgw_host)
+                ),
             )
 
             if result.failed: