openstack/glance
Code Issues Proposed changes

Use (# of CPUs) glance workers by default

Browse Source 
The config docs have historically recommended that the number of
glance  workers should be set to the number of CPUs available for
best performance, so we should make this the default.

Commit 75c96a48fc7e5dfb59d8258142b01422f81b0253 did the same thing in
Nova in Icehouse and the plan is to do the same thing for Cinder
and Trove workers.

The config files are updated to match the help string in the code.

There is no upgrade impact since glance-api.conf previously
hard-coded the workers value to 1 so anyone upgrading to this
will still get whatever value was set in glance-api.conf prior
to this change.

DocImpact: glance workers will now be equal to the number of
           CPUs available by default if not explicitly specified
           in glance-api.conf and/or glance-registry.conf.

UpgradeImpact: There is no upgrade impact to glance-api workers
           since glance-api.conf previously hard-coded the workers
           value to 1 so anyone upgrading to tihs will still get
           whatever value was set in glance-api.conf prior to
           this change. There is an upgrade impact to the
           glance-registry workers since glance-registry.conf
           did not hard-code the workers value to 1 before change
           I0cee0d284eef9ce5dcb26720499f2c4d31eaca0f, which is
           overwritten here. So anyone upgrading to this change
           that does not have workers specified in
           glance-registry.conf will now be running multiple
           workers by default when they restart the glance
           registry service.

Closes-Bug: #1333325

Change-Id: I6795c6e22268bb3fb67331edc7af641aefa904cc
This commit is contained in:
Matt Riedemann 2014-06-25 14:18:36 -07:00
parent dde2cbafd3
commit cb7b189f4c
4 changed files with 13 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
doc/source/configuring.rst
View File

@ -167,7 +167,7 @@ performance (especially if using SSL with compression enabled). Typically
it is recommended to have one worker process per CPU. The value `0`
will prevent any new processes from being created.
Optional. Default: ``1``
Optional. Default: The number of CPUs available will be used by default.
* ``db_auto_create=False``

10
etc/glance-api.conf
View File

@ -55,12 +55,10 @@ backlog = 4096
# package, it is also possible to use: glance.db.registry.api
# data_api = glance.db.sqlalchemy.api
# Number of Glance API worker processes to start.
# On machines with more than one CPU increasing this value
# may improve performance (especially if using SSL with
# compression turned on). It is typically recommended to set
# this value to the number of CPUs present on your machine.
workers = 1
# The number of child process workers that will be
# created to service API requests. The default will be
# equal to the number of CPUs available. (integer value)
#workers = None
# Maximum line size of message headers to be accepted.
# max_header_line may need to be increased when using large tokens

10
etc/glance-registry.conf
View File

@ -29,12 +29,10 @@ backlog = 4096
# package.
#data_api = glance.db.sqlalchemy.api
# Number of Glance Registry worker processes to start.
# On machines with more than one CPU increasing this value
# may improve performance (especially if using SSL with
# compression turned on). It is typically recommended to set
# this value to the number of CPUs present on your machine.
workers = 1
# The number of child process workers that will be
# created to service Registry requests. The default will be
# equal to the number of CPUs available. (integer value)
#workers = None
# Enable Registry API versions individually or simultaneously
#enable_v1_registry = True

6
glance/common/wsgi.py
View File

@ -47,6 +47,7 @@ from glance.common import utils
from glance.openstack.common import gettextutils
from glance.openstack.common import jsonutils
import glance.openstack.common.log as logging
from glance.openstack.common import processutils
bind_opts = [
@ -74,9 +75,10 @@ socket_opts = [
]
eventlet_opts = [
cfg.IntOpt('workers', default=1,
cfg.IntOpt('workers', default=processutils.get_worker_count(),
help=_('The number of child process workers that will be '
'created to service requests.')),
'created to service requests. The default will be '
'equal to the number of CPUs available.')),
cfg.StrOpt('eventlet_hub', default='poll',
help=_('Name of eventlet hub to use. Traditionally, we have '
'only supported \'poll\', however \'selects\' may be '