From b37e93d21263a65d11a508ee52c9f63a91c9d273 Mon Sep 17 00:00:00 2001
From: Cristian Matiut <cmatiut@cloudbasesolutions.com>
Date: Thu, 15 Jan 2026 11:19:07 +0000
Subject: [PATCH 1/2] Increase default retry_on_error and get_status sleep time

---
 coriolis/providers/replicator.py | 2 +-
 coriolis/utils.py                | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/coriolis/providers/replicator.py b/coriolis/providers/replicator.py
index b3ec9add..f69a251e 100644
--- a/coriolis/providers/replicator.py
+++ b/coriolis/providers/replicator.py
@@ -162,7 +162,7 @@ def _get_session(self):
         sess.verify = self._creds["ca_cert"]
         return sess
 
-    @utils.retry_on_error()
+    @utils.retry_on_error(sleep_seconds=10)
     def get_status(self, device=None, brief=True):
         uri = "%s/api/v1/dev" % (self._base_uri)
         if device is not None:
diff --git a/coriolis/utils.py b/coriolis/utils.py
index 097ce52e..21320467 100644
--- a/coriolis/utils.py
+++ b/coriolis/utils.py
@@ -166,7 +166,7 @@ def get_single_result(lis):
     return lis[0]
 
 
-def retry_on_error(max_attempts=5, sleep_seconds=0,
+def retry_on_error(max_attempts=5, sleep_seconds=1,
                    terminal_exceptions=[]):
     def _retry_on_error(func):
         @functools.wraps(func)

From ec8e2b8192b649ddb89493d20fa0a27a092cfba0 Mon Sep 17 00:00:00 2001
From: Cristian Matiut <cmatiut@cloudbasesolutions.com>
Date: Mon, 19 Jan 2026 13:31:20 +0000
Subject: [PATCH 2/2] Catch get_status errors, close ssh on connection error

---
 coriolis/providers/replicator.py | 65 +++++++++++++++++++-------------
 1 file changed, 39 insertions(+), 26 deletions(-)

diff --git a/coriolis/providers/replicator.py b/coriolis/providers/replicator.py
index f69a251e..b4df88ff 100644
--- a/coriolis/providers/replicator.py
+++ b/coriolis/providers/replicator.py
@@ -162,7 +162,7 @@ def _get_session(self):
         sess.verify = self._creds["ca_cert"]
         return sess
 
-    @utils.retry_on_error(sleep_seconds=10)
+    @utils.retry_on_error(sleep_seconds=5)
     def get_status(self, device=None, brief=True):
         uri = "%s/api/v1/dev" % (self._base_uri)
         if device is not None:
@@ -285,8 +285,14 @@ def _reconnect_ssh(self):
         return self._ssh
 
     def init_replicator(self):
-        self._credentials = utils.retry_on_error()(
-            self._setup_replicator)(self._ssh)
+        try:
+            self._credentials = utils.retry_on_error(sleep_seconds=5)(
+                self._setup_replicator)(self._ssh)
+        except Exception:
+            LOG.warn("Failed to setup replicator, trying to reconnect ssh")
+            self._reconnect_ssh()
+            self._credentials = utils.retry_on_error(sleep_seconds=5)(
+                self._setup_replicator)(self._ssh)
         utils.retry_on_error()(
             self._init_replicator_client)(self._credentials)
         LOG.debug(
@@ -372,26 +378,29 @@ def attach_new_disk(
         new_disks_status = None
         new_device_paths = None
         for i in range(retry_count):
-            new_disks_status = self._cli.get_status()
-            new_device_paths = [dev['device-path']
-                                for dev in new_disks_status]
-            LOG.debug(
-                "Polled devices while waiting for disk '%s' to attach "
-                "(try %d/%d): %s", disk_id, i + 1, retry_count,
-                new_device_paths)
-
-            # check for missing/multiple new device paths:
-            missing_device_paths = (
-                set(previous_device_paths) - set(new_device_paths))
-            if missing_device_paths:
-                LOG.warn(
-                    "The following devices from the previous disk state qeury "
-                    "are no longer detected: %s", [
-                        dev for dev in previous_disks_status
-                        if dev['device-path'] in missing_device_paths])
-
-            new_device_paths = set(
-                new_device_paths) - set(previous_device_paths)
+            try:
+                new_disks_status = self._cli.get_status()
+                new_device_paths = [dev['device-path']
+                                    for dev in new_disks_status]
+                LOG.debug(
+                    "Polled devices while waiting for disk '%s' to attach "
+                    "(try %d/%d): %s", disk_id, i + 1, retry_count,
+                    new_device_paths)
+
+                # check for missing/multiple new device paths:
+                missing_device_paths = (
+                    set(previous_device_paths) - set(new_device_paths))
+                if missing_device_paths:
+                    LOG.warn(
+                        "The following devices from the previous disk state "
+                        "qeury are no longer detected: %s", [
+                            dev for dev in previous_disks_status
+                            if dev['device-path'] in missing_device_paths])
+
+                new_device_paths = set(
+                    new_device_paths) - set(previous_device_paths)
+            except Exception:
+                LOG.debug("Failed to get new device status")
             if new_device_paths:
                 break
             else:
@@ -466,7 +475,7 @@ def update_state(self, state, restart=False):
             self.restart()
             self._cli._test_connection()
 
-    @utils.retry_on_error()
+    @utils.retry_on_error(sleep_seconds=5)
     def _get_ssh_client(self, args):
         """
         gets a paramiko SSH client
@@ -474,8 +483,12 @@ def _get_ssh_client(self, args):
         try:
             ssh = paramiko.SSHClient()
             ssh.set_missing_host_key_policy(paramiko.AutoAddPolicy())
-            ssh.connect(**args)
-            return ssh
+            try:
+                ssh.connect(**args)
+                return ssh
+            except Exception:
+                ssh.close()
+                raise
         except paramiko.ssh_exception.SSHException as ex:
             raise exception.CoriolisException(
                 "Failed to setup SSH client: %s" % str(ex)) from ex