Re: [CentOS-devel] [rdo-list] [centos-ci] Artifacts server does not support ranges / resume of downloads?
2016-06-28 4:25 GMT+02:00 Gerard Braad me@gbraad.nl:
Hi all,
Usually I download the undercloud images using:
https://gist.github.com/gbraad/45cbe30415b0dc631f5e8d20beaffebf
which targets: 'artifacts.ci.centos.org' but now get
"HTTP server doesn't seem to support byte ranges. Cannot resume."
I need to resume a download about 10 times on the office connection (China), so what happened? I have a workaround by storing the files at another server first in Japan or the west-coast which supports resuming.
Although, being able to resume the download of these files is preferable.
regards,
Gerard
Since this part of infrastructure is managed by CentOS Core Team, CC'ing centos-devel list.
--
Gerard Braad | http://gbraad.nl [ Doing Open Source Matters ]
rdo-list mailing list rdo-list@redhat.com https://www.redhat.com/mailman/listinfo/rdo-list
To unsubscribe: rdo-list-unsubscribe@redhat.com
Hi,
On Wed, Jun 29, 2016 at 6:51 PM, Haïkel hguemar@fedoraproject.org wrote:
2016-06-28 4:25 GMT+02:00 Gerard Braad me@gbraad.nl:
https://gist.github.com/gbraad/45cbe30415b0dc631f5e8d20beaffebf
Since this part of infrastructure is managed by CentOS Core Team, CC'ing centos-devel list.
I tried as John(trown) suggested and use the buildlogs.centos.org at the moment, which seems to perform a 302. to buildlogs.cdn.centos.org. Due to my location (China) I have to add this entry to my /etc/hosts as:
185.59.223.23 buildlogs.cdn.centos.org
It seems this host allows for peaks that go over 2M/s, but overall does not improve much as times out more often.
Did something in the configuration of artifacts.ci.centos.org change? Currently it allows ranges for resume. And I'd rather download at 500k/s (to 2M/s) speed, as long as it is more reliable...
regards,
Gerard
FYI, most of the CentOS Core Team are at RH Summit.
On Jun 29 12:51, Haïkel wrote:
2016-06-28 4:25 GMT+02:00 Gerard Braad me@gbraad.nl:
Hi all,
Usually I download the undercloud images using:
https://gist.github.com/gbraad/45cbe30415b0dc631f5e8d20beaffebf
which targets: 'artifacts.ci.centos.org' but now get
"HTTP server doesn't seem to support byte ranges. Cannot resume."
I need to resume a download about 10 times on the office connection (China), so what happened? I have a workaround by storing the files at another server first in Japan or the west-coast which supports resuming.
Although, being able to resume the download of these files is preferable.
regards,
Gerard
Since this part of infrastructure is managed by CentOS Core Team, CC'ing centos-devel list.
--
Gerard Braad | http://gbraad.nl [ Doing Open Source Matters ]
We haven't changed anything related to the infrastructure on that machine here, I'm going to chalk this up to something in the route along the way to artifacts.ci (perhaps a proxy in the middle?)
I did a few tests and got the usual response to http range requests.
Cheers!
-- Brian Stinson CentOS CI Infrastructure Team
Hi,
On Fri, Jul 1, 2016 at 1:43 AM, Brian Stinson brian@bstinson.com wrote:
(perhaps a proxy in the middle?) I did a few tests and got the usual response to http range requests.
I was able to download the images again... and yes, it was able to do range requests. (In the situation of China it is impossible to know what it is in-between).
regards,
Gerard
From adi.gangidi at rackspace.com Thu Jul 14 01:23:20 2016 Date: Thu, 14 Jul 2016 01:23:20 +0000
Hello nice people on CentOS list,
?Default governor on CentOS 7 ppc64le OS bootup is "conservative" . This has about 20% effect on benchmark numbers, if you don't change the governor to ondemand manually (for specific benchmarks). I tested this on our barreleye power 8 server.
This even though kernel config is set for "ondemand" as default governor:
[root at Barreleye-1 ~]# cat /boot/config-3.10.0-327.el7.ppc64le | grep -i ondemand CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND=y CONFIG_CPU_FREQ_GOV_ONDEMAND=y
I filed a bug on this with details and am posting here for info and also to accelerate some response on the bug: https://bugs.centos.org/view.php?id=11094
Best Adi Gangidi Rackspace
Hello, Adi.
Any chance that other software is overriding the default CPU governor setting? For example, tuned is a good candidate.
Hey Murilo
I think so too. Tuned is causing this. But I feel like this shouldn't be the case.
It would seem like right thing to do is to ensure post-install out of box governor should be, what's being set in kernel config (on-demand).
Best Adi ________________________________________ From: Murilo Opsfelder Araújo muriloo@linux.vnet.ibm.com Sent: Thursday, July 14, 2016 6:15 PM To: centos-devel@centos.org Cc: Adi Gangidi Subject: Re: [CentOS-devel] Incorrect default governor on CentOS ppc64el builds
From adi.gangidi at rackspace.com Thu Jul 14 01:23:20 2016 Date: Thu, 14 Jul 2016 01:23:20 +0000
Hello nice people on CentOS list,
?Default governor on CentOS 7 ppc64le OS bootup is "conservative" . This has about 20% effect on benchmark numbers, if you don't change the governor to ondemand manually (for specific benchmarks). I tested this on our barreleye power 8 server.
This even though kernel config is set for "ondemand" as default governor:
[root at Barreleye-1 ~]# cat /boot/config-3.10.0-327.el7.ppc64le | grep -i ondemand CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND=y CONFIG_CPU_FREQ_GOV_ONDEMAND=y
I filed a bug on this with details and am posting here for info and also to accelerate some response on the bug: https://bugs.centos.org/view.php?id=11094
Best Adi Gangidi Rackspace
Hello, Adi.
Any chance that other software is overriding the default CPU governor setting? For example, tuned is a good candidate.
-- Murilo
On 07/25/2016 12:07 PM, Adi Gangidi wrote:
Hey Murilo
I think so too. Tuned is causing this. But I feel like this shouldn't be the case.
It would seem like right thing to do is to ensure post-install out of box governor should be, what's being set in kernel config (on-demand).
Best Adi
The tuned package is also installed by default on a RHEL 7.2 ppc64le minimal installation:
# rpm -q tuned tuned-2.5.1-4.el7_2.3.noarch
# rpm -q redhat-release-server redhat-release-server-7.2-9.el7.ppc64le
I don't believe this is a bug with CentOS; it is just honoring what RHEL installs by default.
Perhaps someone from Red Hat on the list can explain the rationale of installing tuned by default.
Sure, if someone can please explain the rationale / requirement for tuned to be installed by default I will close the bug.
--Adi ________________________________________ From: Murilo Opsfelder Araújo muriloo@linux.vnet.ibm.com Sent: Tuesday, August 2, 2016 7:18 AM To: Adi Gangidi; centos-devel@centos.org Subject: Re: [CentOS-devel] Incorrect default governor on CentOS ppc64el builds
On 07/25/2016 12:07 PM, Adi Gangidi wrote:
Hey Murilo
I think so too. Tuned is causing this. But I feel like this shouldn't be the case.
It would seem like right thing to do is to ensure post-install out of box governor should be, what's being set in kernel config (on-demand).
Best Adi
The tuned package is also installed by default on a RHEL 7.2 ppc64le minimal installation:
# rpm -q tuned tuned-2.5.1-4.el7_2.3.noarch
# rpm -q redhat-release-server redhat-release-server-7.2-9.el7.ppc64le
I don't believe this is a bug with CentOS; it is just honoring what RHEL installs by default.
Perhaps someone from Red Hat on the list can explain the rationale of installing tuned by default.
-- Murilo
On Tue, Aug 2, 2016 at 7:18 AM, Murilo Opsfelder Araújo < muriloo@linux.vnet.ibm.com> wrote:
On 07/25/2016 12:07 PM, Adi Gangidi wrote:
Hey Murilo
I think so too. Tuned is causing this. But I feel like this shouldn't be
the case.
It would seem like right thing to do is to ensure post-install out of
box governor should be, what's being set in kernel config
(on-demand).
Best Adi
The tuned package is also installed by default on a RHEL 7.2 ppc64le minimal installation:
# rpm -q tuned tuned-2.5.1-4.el7_2.3.noarch
# rpm -q redhat-release-server redhat-release-server-7.2-9.el7.ppc64le
I don't believe this is a bug with CentOS; it is just honoring what RHEL installs by default.
Perhaps someone from Red Hat on the list can explain the rationale of installing tuned by default.
Tuned chooses its profile based on the OS variant. See https://git.fedorahosted.org/cgit/tuned.git/tree/recommend.conf
The RHEL Server and Compute Node variants will use the throughput-performance profile, which sets the ondemand governor, while the Workstation and Client variants use the balanced profile, which sets the conservative governor.
CentOS does not set any variant to /etc/system-release-cpe, so it ends up with the default profile being the balanced.
I guess we could add the "server" string to /etc/system-release-cpe ? At least for ppc64 and ppc64le, I'm sure that is what we want anyway.
[]'s Gustavo
-- Murilo
CentOS-devel mailing list CentOS-devel@centos.org https://lists.centos.org/mailman/listinfo/centos-devel
-
Adi Gangidi -
Brian Stinson -
Gerard Braad -
Gustavo Luiz Duarte -
Haïkel -
Murilo Opsfelder Araújo