openstack
/
swift
Code Issues Proposed changes

Add the max_clients parameter to bound clients 

The new max_clients parameter allows one full control over the maximum
number of client requests that will be handled by a given worker for
any of the proxy, account, container or object servers.

Lowering the number of clients handled per worker, and raising the
number of workers can lessen the impact that a CPU intensive, or
blocking, request can have on other requests served by the same
worker.

If the maximum number of clients is set to one, then a given worker
will not perform another accept(2) call while processing, allowing
other workers a chance to process it.

DocImpact
Signed-off-by: Peter Portante <peter.portante@redhat.com>

Change-Id: Ic01430f7a6c5ff48d7aa349dc86a5f8ac463a420
This commit is contained in:
Peter Portante 2013-04-19 14:15:15 -04:00
parent 55c817ce9d
commit 2d42b37303
10 changed files with 122 additions and 26 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
doc/manpages/account-server.conf.5
View File

@ -114,6 +114,12 @@ Logging level. The default is INFO.
Enables request logging. The default is True.
.IP "\fB set log_address\fR
Logging address. The default is /dev/log.
.IP \fBmax_clients\fR
Maximum number of clients one worker can process simultaneously (it will
actually accept(2) N + 1). Setting this to one (1) will only handle one request
at a time, without accepting another request concurrently. By increasing the
number of workers to a much higher value, one can reduce the impact of slow file system
operations in one request from negatively impacting other requests. The default is 1024.
.RE
.PD
@ -253,5 +259,3 @@ and
.SH "SEE ALSO"
.BR swift-account-server(1),

8
doc/manpages/container-server.conf.5
View File

@ -120,6 +120,12 @@ Logging address. The default is /dev/log.
Request timeout to external services. The default is 3 seconds.
.IP \fBconn_timeout\fR
Connection timeout to external services. The default is 0.5 seconds.
.IP \fBmax_clients\fR
Maximum number of clients one worker can process simultaneously (it will
actually accept(2) N + 1). Setting this to one (1) will only handle one request
at a time, without accepting another request concurrently. By increasing the
number of workers to a much higher value, one can reduce the impact of slow file system
operations in one request from negatively impacting other requests. The default is 1024.
.RE
.PD
@ -278,5 +284,3 @@ and
.SH "SEE ALSO"
.BR swift-container-server(1),

8
doc/manpages/object-server.conf.5
View File

@ -120,6 +120,12 @@ Logging address. The default is /dev/log.
Request timeout to external services. The default is 3 seconds.
.IP \fBconn_timeout\fR
Connection timeout to external services. The default is 0.5 seconds.
.IP \fBmax_clients\fR
Maximum number of clients one worker can process simultaneously (it will
actually accept(2) N + 1). Setting this to one (1) will only handle one request
at a time, without accepting another request concurrently. By increasing the
number of workers to a much higher value, one can reduce the impact of slow file system
operations in one request from negatively impacting other requests. The default is 1024.
.RE
.PD
@ -270,5 +276,3 @@ and
.SH "SEE ALSO"
.BR swift-object-server(1),

6
doc/manpages/proxy-server.conf.5
View File

@ -537,6 +537,12 @@ object. The default is 10 segments.
.IP \fBrate_limit_segments_per_sec\fR
Once segment rate-limiting kicks in for an object, limit segments served to N
per second. The default is 1.
.IP \fBmax_clients\fR
Maximum number of clients one worker can process simultaneously (it will
actually accept(2) N + 1). Setting this to one (1) will only handle one request
at a time, without accepting another request concurrently. By increasing the
number of workers to a much higher value, one can reduce the impact of slow file system
operations in one request from negatively impacting other requests. The default is 1024.
.RE
.PD

89
doc/source/deployment_guide.rst
View File

@ -57,7 +57,7 @@ Web Front End Options
---------------------
Swift comes with an integral web front end. However, it can also be deployed
as a request processor of an Apache2 using mod_wsgi as described in
as a request processor of an Apache2 using mod_wsgi as described in
:doc:`Apache Deployment Guide <apache_deployment_guide>`.
.. _ring-preparing:
@ -241,6 +241,16 @@ bind_ip 0.0.0.0 IP Address for server to bind to
bind_port 6000 Port for server to bind to
bind_timeout 30 Seconds to attempt bind before giving up
workers 1 Number of workers to fork
max_clients 1024 Maximum number of clients one worker can
process simultaneously (it will actually
accept(2) N + 1). Setting this to one (1)
will only handle one request at a time,
without accepting another request
concurrently. By increasing the number of
workers to a much higher value, one can
reduce the impact of slow file system
operations in one request from negatively
impacting other requests.
disable_fallocate false Disable "fast fail" fallocate checks if the
underlying filesystem does not support it.
log_custom_handlers None Comma-separated list of functions to call
@ -358,6 +368,16 @@ bind_ip 0.0.0.0 IP Address for server to bind to
bind_port 6001 Port for server to bind to
bind_timeout 30 Seconds to attempt bind before giving up
workers 1 Number of workers to fork
max_clients 1024 Maximum number of clients one worker can
process simultaneously (it will actually
accept(2) N + 1). Setting this to one (1)
will only handle one request at a time,
without accepting another request
concurrently. By increasing the number of
workers to a much higher value, one can
reduce the impact of slow file system
operations in one request from negatively
impacting other requests.
user swift User to run as
disable_fallocate false Disable "fast fail" fallocate checks if the
underlying filesystem does not support it.
@ -440,7 +460,7 @@ log_name container-auditor Label used when logging
log_facility LOG_LOCAL0 Syslog log facility
log_level INFO Logging level
interval 1800 Minimum time for a pass to take
containers_per_second 200 Maximum containers audited per second.
containers_per_second 200 Maximum containers audited per second.
Should be tuned according to individual
system specs. 0 is unlimited.
===================== ================= =======================================
@ -468,6 +488,16 @@ bind_ip 0.0.0.0 IP Address for server to bind to
bind_port 6002 Port for server to bind to
bind_timeout 30 Seconds to attempt bind before giving up
workers 1 Number of workers to fork
max_clients 1024 Maximum number of clients one worker can
process simultaneously (it will actually
accept(2) N + 1). Setting this to one (1)
will only handle one request at a time,
without accepting another request
concurrently. By increasing the number of
workers to a much higher value, one can
reduce the impact of slow file system
operations in one request from negatively
impacting other requests.
user swift User to run as
db_preallocation off If you don't mind the extra disk space usage in
overhead, you can turn this on to preallocate
@ -527,9 +557,9 @@ log_name account-auditor Label used when logging
log_facility LOG_LOCAL0 Syslog log facility
log_level INFO Logging level
interval 1800 Minimum time for a pass to take
accounts_per_second 200 Maximum accounts audited per second.
accounts_per_second 200 Maximum accounts audited per second.
Should be tuned according to individual
system specs. 0 is unlimited.
system specs. 0 is unlimited.
==================== =============== =======================================
[account-reaper]
@ -572,6 +602,21 @@ bind_timeout 30 Seconds to attempt bind before
giving up
swift_dir /etc/swift Swift configuration directory
workers 1 Number of workers to fork
max_clients 1024 Maximum number of clients one
worker can process
simultaneously (it will
actually accept(2) N +
1). Setting this to one (1)
will only handle one request at
a time, without accepting
another request
concurrently. By increasing the
number of workers to a much
higher value, one can reduce
the impact of slow file system
operations in one request from
negatively impacting other
requests.
user swift User to run as
cert_file Path to the ssl .crt. This
should be enabled for testing
@ -580,9 +625,9 @@ key_file Path to the ssl .key. This
should be enabled for testing
purposes only.
cors_allow_origin This is a list of hosts that
are included with any CORS
request by default and
returned with the
are included with any CORS
request by default and
returned with the
Access-Control-Allow-Origin
header in addition to what
the container has set.
@ -785,12 +830,12 @@ should also be monitored to ensure that the times do not vary too much.
General Service Tuning
----------------------
Most services support either a worker or concurrency value in the settings.
This allows the services to make effective use of the cores available. A good
starting point to set the concurrency level for the proxy and storage services
to 2 times the number of cores available. If more than one service is
sharing a server, then some experimentation may be needed to find the best
balance.
Most services support either a `worker` or `concurrency` value in the
settings. This allows the services to make effective use of the cores
available. A good starting point to set the concurrency level for the proxy
and storage services to 2 times the number of cores available. If more than
one service is sharing a server, then some experimentation may be needed to
find the best balance.
At Rackspace, our Proxy servers have dual quad core processors, giving us 8
cores. Our testing has shown 16 workers to be a pretty good balance when
@ -798,9 +843,21 @@ saturating a 10g network and gives good CPU utilization.
Our Storage servers all run together on the same servers. These servers have
dual quad core processors, for 8 cores total. We run the Account, Container,
and Object servers with 8 workers each. Most of the background jobs are run
at a concurrency of 1, with the exception of the replicators which are run at
a concurrency of 2.
and Object servers with 8 workers each. Most of the background jobs are run at
a concurrency of 1, with the exception of the replicators which are run at a
concurrency of 2.
The `max_clients` parameter can be used to adjust the number of client
requests an individual worker accepts for processing. The fewer requests being
processed at one time, the less likely a request that consumes the worker's
CPU time, or blocks in the OS, will negatively impact other requests. The more
requests being processed at one time, the more likely one worker can utilize
network and disk capacity.
On systems that have more cores, and more memory, where one can afford to run
more workers, raising the number of workers and lowering the maximum number of
clients serviced per worker can lessen the impact of CPU intensive or stalled
requests.
The above configuration setting should be taken as suggestions and testing
of configuration settings should be done to ensure best utilization of CPU,

1
etc/account-server.conf-sample
View File

@ -47,6 +47,7 @@ use = egg:swift#account
# set log_requests = True
# set log_address = /dev/log
# auto_create_account_prefix = .
# max_clients = 1024
[filter:healthcheck]
use = egg:swift#healthcheck

1
etc/container-server.conf-sample
View File

@ -53,6 +53,7 @@ use = egg:swift#container
# conn_timeout = 0.5
# allow_versions = False
# auto_create_account_prefix = .
# max_clients = 1024
[filter:healthcheck]
use = egg:swift#healthcheck

1
etc/object-server.conf-sample
View File

@ -62,6 +62,7 @@ use = egg:swift#object
# Content-Type, etag, Content-Length, or deleted
# allowed_headers = Content-Disposition, Content-Encoding, X-Delete-At, X-Object-Manifest, X-Static-Large-Object
# auto_create_account_prefix = .
# max_clients = 1024
[filter:healthcheck]
use = egg:swift#healthcheck

1
etc/proxy-server.conf-sample
View File

@ -37,6 +37,7 @@
# Use a comma separated list of full url (http://foo.bar:1234,https://foo.bar)
# cors_allow_origin =
# eventlet_debug = false
# max_clients = 1024
[pipeline:main]
pipeline = catch_errors healthcheck proxy-logging cache slo ratelimit tempauth container-quotas account-quotas proxy-logging proxy-server

25
swift/common/wsgi.py
View File

@ -105,6 +105,22 @@ def get_socket(conf, default_port=8080):
return sock
class RestrictedGreenPool(GreenPool):
"""
Works the same as GreenPool, but if the size is specified as one, then the
spawn_n() method will invoke waitall() before returning to prevent the
caller from doing any other work (like calling accept()).
"""
def __init__(self, size=1024):
super(RestrictedGreenPool, self).__init__(size=size)
self._rgp_do_wait = (size == 1)
def spawn_n(self, *args, **kwargs):
super(RestrictedGreenPool, self).spawn_n(*args, **kwargs)
if self._rgp_do_wait:
self.waitall()
# TODO: pull pieces of this out to test
def run_wsgi(conf_file, app_section, *args, **kwargs):
"""
@ -132,7 +148,7 @@ def run_wsgi(conf_file, app_section, *args, **kwargs):
# redirect errors to logger and close stdio
capture_stdio(logger)
def run_server():
def run_server(max_clients):
wsgi.HttpProtocol.default_request_version = "HTTP/1.0"
# Turn off logging requests by the underlying WSGI software.
wsgi.HttpProtocol.log_request = lambda *a: None
@ -147,7 +163,7 @@ def run_wsgi(conf_file, app_section, *args, **kwargs):
eventlet.debug.hub_exceptions(eventlet_debug)
app = loadapp('config:%s' % conf_file,
global_conf={'log_name': log_name})
pool = GreenPool(size=1024)
pool = RestrictedGreenPool(size=max_clients)
try:
wsgi.server(sock, app, NullLogger(), custom_pool=pool)
except socket.error, err:
@ -155,10 +171,11 @@ def run_wsgi(conf_file, app_section, *args, **kwargs):
raise
pool.waitall()
max_clients = int(conf.get('max_clients', '1024'))
worker_count = int(conf.get('workers', '1'))
# Useful for profiling [no forks].
if worker_count == 0:
run_server()
run_server(max_clients)
return
def kill_children(*args):
@ -184,7 +201,7 @@ def run_wsgi(conf_file, app_section, *args, **kwargs):
if pid == 0:
signal.signal(signal.SIGHUP, signal.SIG_DFL)
signal.signal(signal.SIGTERM, signal.SIG_DFL)
run_server()
run_server(max_clients)
logger.notice('Child %d exiting normally' % os.getpid())
return
else: