Remove threads_per_disk setting

This patch removes the threads_per_disk setting. It was already a deprecated
setting and by default set to 0, which effectively meant to not use a per-disk
thread pool at all. Users are encouraged to use servers_per_port instead.

DocImpact

Change-Id: Ie76be5c8a74d60a1330627caace19e06d1b9383c

doc/manpages/object-server.conf.5

@@ -88,10 +88,9 @@ The default is 86400.
 .IP \fBexpiring_objects_account_name\fR
 The default is 'expiring_objects'.
 .IP \fBservers_per_port\fR
-Make object-server run this many worker processes per unique port of
-"local" ring devices across all storage policies.  This can help provide
-the isolation of threads_per_disk without the severe overhead.  The default
-value of 0 disables this feature.
+Make object-server run this many worker processes per unique port of "local"
+ring devices across all storage policies. The default value of 0 disables this
+feature.
 .IP \fBlog_name\fR
 Label used when logging. The default is swift.
 .IP \fBlog_facility\fR

doc/source/deployment_guide.rst

@@ -151,11 +151,6 @@ service a request for any disk, and a slow I/O request blocks the eventlet hub,
 a single slow disk can impair an entire storage node.  This also prevents
 object servers from fully utilizing all their disks during heavy load.
 
-The :ref:`threads_per_disk <object-server-options>` option was one way to
-address this, but came with severe performance overhead which was worse
-than the benefit of I/O isolation.  Any clusters using threads_per_disk should
-switch to using `servers_per_port`.
-
 Another way to get full I/O isolation is to give each disk on a storage node a
 different port in the storage policy rings.  Then set the
 :ref:`servers_per_port <object-server-default-options>`
@@ -547,15 +542,6 @@ allowed_headers                Content-Disposition,   Comma separated list of he
                                X-Static-Large-Object  Content-Type, etag, Content-Length, or deleted
 auto_create_account_prefix     .                      Prefix used when automatically
                                                       creating accounts.
-threads_per_disk               0                      Size of the per-disk thread pool
-                                                      used for performing disk I/O. The
-                                                      default of 0 means to not use a
-                                                      per-disk thread pool.
-                                                      This option is no longer
-                                                      recommended and the
-                                                      :ref:`servers_per_port
-                                                      <server-per-port-configuration>`
-                                                      should be used instead.
 replication_server                                    Configure parameter for creating
                                                       specific server. To handle all verbs,
                                                       including replication verbs, do not

etc/object-server.conf-sample

@@ -16,10 +16,9 @@ bind_port = 6000
 # ignored.
 # workers = auto
 #
-# Make object-server run this many worker processes per unique port of
-# "local" ring devices across all storage policies.  This can help provide
-# the isolation of threads_per_disk without the severe overhead.  The default
-# value of 0 disables this feature.
+# Make object-server run this many worker processes per unique port of "local"
+# ring devices across all storage policies. The default value of 0 disables this
+# feature.
 # servers_per_port = 0
 #
 # Maximum concurrent requests per worker
@@ -106,10 +105,6 @@ use = egg:swift#object
 #
 # auto_create_account_prefix = .
 #
-# A value of 0 means "don't use thread pools". A reasonable starting point is
-# 4.
-# threads_per_disk = 0
-#
 # Configure parameter for creating specific server
 # To handle all verbs, including replication verbs, do not specify
 # "replication_server" (this is the default). To only handle replication,

swift/common/manager.py

@@ -55,7 +55,7 @@ WARNING_WAIT = 3  # seconds to wait after message that may just be a warning
 
 MAX_DESCRIPTORS = 32768
 MAX_MEMORY = (1024 * 1024 * 1024) * 2  # 2 GB
-MAX_PROCS = 8192  # workers * disks * threads_per_disk, can get high
+MAX_PROCS = 8192  # workers * disks, can get high
 
 
 def setup_env():

swift/obj/diskfile.py

@@ -536,9 +536,7 @@ class BaseDiskFileManager(object):
             conf.get('replication_one_per_device', 'true'))
         self.replication_lock_timeout = int(conf.get(
             'replication_lock_timeout', 15))
-        threads_per_disk = int(conf.get('threads_per_disk', '0'))
-        self.threadpools = defaultdict(
-            lambda: ThreadPool(nthreads=threads_per_disk))
+        self.threadpools = defaultdict(lambda: ThreadPool(nthreads=0))
 
         self.use_splice = False
         self.pipe_size = None