nvme: run nvme connect-all --nbft before running the actual probes

When running in an environment where a NBFT is available, we want
probert to collect information about the remote NVMe drives too.

This can be done by calling the `nvme connect-all --nbft` command. That
said, running the command in the NVMe probe code itself will be too late
for other probes (e.g., blockdev, filesystem, ...) to pick-up
information about the remote NVMe drives.

Instead, we run the command before dispatching the probes ; in a
function called "activate_devices". In the long run, I would like to
move some code that is currently in the probes (e.g., vgchange
--activate=y) to the activate_devices function ; to avoid side effects
in the probe code.

Signed-off-by: Olivier Gayot <[email protected]>

probert/lvm.py

@@ -196,6 +196,8 @@ async def probe(context=None, **kw):
         older LVM2 stacks, the LVM probe may be incomplete.
     """
     # scan and activate lvm vgs/lvs
+    # TODO it feels like the activation code should be moved to
+    # Storage.activate_devices.
     lvm_scan()
     activate_volgroups()
 

probert/nvme.py

@@ -13,10 +13,11 @@
 # along with this program.  If not, see <http://www.gnu.org/licenses/>.
 
 import logging
+import subprocess
 
 import pyudev
 
-from probert.utils import udev_get_attributes
+from probert.utils import arun, udev_get_attributes
 
 log = logging.getLogger('probert.nvme')
 
@@ -32,3 +33,11 @@ async def probe(context=None, **kw):
         nvme_controllers[controller.sys_name] = props
 
     return nvme_controllers
+
+
+async def connect_nbft() -> None:
+    cmd = ['nvme', 'connect-all', '--nbft']
+    try:
+        await arun(cmd)
+    except (subprocess.CalledProcessError, FileNotFoundError):
+        log.error('Failed to run cmd: %s', cmd)

probert/storage.py

@@ -230,6 +230,10 @@ async def probe(self, probe_types=None, *, parallelize=False):
             print('Unavilable probe types: %s' % not_avail)
             return self.results
 
+        # Run operations that have side-effects, before we start reading udev
+        # cache.
+        await self.activate_devices()
+
         probed_data = {}
 
         async def run_probe(ptype):
@@ -250,3 +254,20 @@ async def run_probe(ptype):
 
         self.results = probed_data
         return probed_data
+
+    async def activate_devices(probe_types: list[str]) -> None:
+        """Before querying udev, we want to let it know of the existence of
+        some block devices (which may not be currently visible). This includes
+        connecting to NVMe drives using the NVMe/TCP protocol, but also
+        activating LVM VGs to make LVM LVs visible.
+        Typically, commands must be run in a specific order to make all the
+        devices visible; so running the commands in the code of the probes
+        themselves (which run in an undefined order) feels wrong.
+        Ideally, the code of the probes should have no side effect, and we
+        should move operations that have side effect here."""
+        if "nvme" in probe_types:
+            await nvme.connect_nbft()
+
+        # TODO we should probably move the LVM VG activation code here ; so
+        # that blockdev, filesystem_sizing, OS detection and other probes
+        # include information about LVs.