Do not call persistent_set() from libnvme

The persistent_set() API from libnvme had absolutely no effect on connections. Most importantly, it did not set a default KATO on persistent DC connections. I/O connections, on the other hand, are assigned a default 2-minute KATO by the kernel regardless of the persistent_set() function. Signed-off-by: Martin Belanger <[email protected]>
linux-nvme · Mar 28, 2022 · 2e76a8e · 2e76a8e
1 parent a708be9
commit 2e76a8e
Show file tree

Hide file tree

Showing 2 changed files with 14 additions and 3 deletions.
diff --git a/etc/stas/stafd.conf b/etc/stas/stafd.conf
@@ -33,11 +33,11 @@
 
 # kato:        Keep Alive Timeout (KATO): This field specifies the timeout value
 #              for the Keep Alive feature in seconds. The default value for this
-#              field is 120 seconds (2 minutes).
+#              field is 30 seconds.
 #              Type:  Unsigned integer
 #              Range: 0..N
 #              Unit:  Seconds
-#kato=120
+#kato=30
 
 # persistent-connections: Whether connections to Discovery Controllers (DC)
 #                         are persistent. If stafd is stopped, the connections

diff --git a/staslib/stas.py b/staslib/stas.py
@@ -30,6 +30,8 @@
 from libnvme import nvme
 from staslib import defs
 
+DC_KATO_DEFAULT = 30 # seconds
+
 #*******************************************************************************
 def check_if_allowed_to_continue():
     ''' @brief Let's perform some basic checks before going too far. There are
@@ -1160,7 +1162,6 @@ def _try_to_connect(self):
                                host_traddr=self.tid.host_traddr if self.tid.host_traddr else None,
                                host_iface=host_iface)
         self._ctrl.discovery_ctrl_set(self._discovery_ctrl)
-        self._ctrl.persistent_set(True)
 
         # Audit existing nvme devices. If we find a match, then
         # we'll just borrow that device instead of creating a new one.
@@ -1177,6 +1178,16 @@ def _try_to_connect(self):
                     'data_digest': CNF.data_digest }
             if CNF.kato is not None:
                 cfg['keep_alive_tmo'] = CNF.kato
+            elif self._discovery_ctrl:
+                # All the connections to Controllers (I/O and Discovery) are
+                # persistent. Persistent connections MUST configure the KATO.
+                # The kernel already assigns a default 2-minute KATO to I/O
+                # controller connections, but it doesn't assign one to
+                # Discovery controller (DC) connections. Here we set the default
+                # DC connection KATO to match the default set by nvme-cli on
+                # persistent DC connections (i.e. 30 sec).
+                cfg['keep_alive_tmo'] = DC_KATO_DEFAULT
+
             LOG.debug('Controller._try_to_connect()       - %s Connecting to nvme control with cfg=%s', self.id, cfg)
             self._connect_op = AsyncOperationWithRetry(self._on_connect_success, self._on_connect_fail,
                                                        self._ctrl.connect, self._host, cfg)