From c197bbdccadf799be2d86da28f1d2302ccf5e1e8 Mon Sep 17 00:00:00 2001
From: Takashi Kajinami <kajinamit@oss.nttdata.com>
Date: Mon, 27 Nov 2023 11:33:12 +0900
Subject: [PATCH] RBD: Use rados_connect_timeout to override timeout

This imports the existing rados_connect_timeout option handling in
cinder so that users can use a different timeouts for Glance
specifically, instead of relying on the global ceph.conf options.

This parameter was initially deprecated in Zed release by [1] and has
had no effect since then, but this change restores the parameter with
the logic to override the timeout value in RADOS client.

[1] b1d0feeba4e16046c7840bd4966f31a0e7527e67

Related-Bug: #1983499
Change-Id: Ib370f527c06dc85bcfd9df6ca1efb2a4e8cb5e7d
---
 glance_store/_drivers/rbd.py                  | 25 +++++++++++--------
 ...ados_connect_timeout-39e5074bc1a3b65b.yaml | 19 ++++++++++++++
 2 files changed, 34 insertions(+), 10 deletions(-)
 create mode 100644 releasenotes/notes/fix-rados_connect_timeout-39e5074bc1a3b65b.yaml

diff --git a/glance_store/_drivers/rbd.py b/glance_store/_drivers/rbd.py
index ba2defa3..5e18af8f 100644
--- a/glance_store/_drivers/rbd.py
+++ b/glance_store/_drivers/rbd.py
@@ -134,14 +134,7 @@ Related options:
     * rbd_store_user
 
 """),
-    cfg.IntOpt('rados_connect_timeout', default=0,
-               deprecated_for_removal=True,
-               deprecated_since='Zed',
-               deprecated_reason="""
-This option has not had any effect in years. Users willing to set a timeout for
-connecting to the Ceph cluster should use 'client_mount_timeout' in Ceph's
-configuration file.
-""",
+    cfg.IntOpt('rados_connect_timeout', default=-1,
                help="""
 Timeout value for connecting to Ceph cluster.
 
@@ -149,8 +142,8 @@ This configuration option takes in the timeout value in seconds used
 when connecting to the Ceph cluster i.e. it sets the time to wait for
 glance-api before closing the connection. This prevents glance-api
 hangups during the connection to RBD. If the value for this option
-is set to less than or equal to 0, no timeout is set and the default
-librados value is used.
+is set to less than 0, no timeout is set and the default librados value
+is used.
 
 Possible Values:
     * Any integer value
@@ -302,6 +295,18 @@ class Store(driver.Store):
     def get_connection(self, conffile, rados_id):
         client = rados.Rados(conffile=conffile, rados_id=rados_id)
 
+        if self.backend_group:
+            timeout = getattr(self.conf,
+                              self.backend_group).rados_connect_timeout
+        else:
+            timeout = self.conf.glance_store.rados_connect_timeout
+
+        if timeout >= 0:
+            t = str(timeout)
+            client.conf_set('rados_osd_op_timeout', t)
+            client.conf_set('rados_mon_op_timeout', t)
+            client.conf_set('client_mount_timeout', t)
+
         try:
             client.connect()
         except (rados.Error, rados.ObjectNotFound) as e:
diff --git a/releasenotes/notes/fix-rados_connect_timeout-39e5074bc1a3b65b.yaml b/releasenotes/notes/fix-rados_connect_timeout-39e5074bc1a3b65b.yaml
new file mode 100644
index 00000000..27aef5be
--- /dev/null
+++ b/releasenotes/notes/fix-rados_connect_timeout-39e5074bc1a3b65b.yaml
@@ -0,0 +1,19 @@
+---
+features:
+  - |
+    RBD driver: the ``rados_connect_timeout`` config option has been
+    un-deprecated and its behavior has been improved.  A value of ``0``
+    is now respected as disabling timeout in requests, while a value less
+    than zero indicates that glance_store will not set a timeout but
+    instead will use whatever timeouts are set in the Ceph configuration
+    file.
+
+upgrade:
+  - |
+    RBD driver: the default value of the ``rados_connect_timeout`` option
+    has been changed from 0 to -1, so that the RBD driver will by default
+    use the timeout values defined in ``ceph.conf``. Be aware that
+    setting this option to 0 disables timeouts (that is, the RBD driver
+    will make requests with a timeout of zero, and all requests wait forever),
+    thereby overriding any timeouts that are set in the Ceph configuration
+    file.