Aloha and thanks for all your hard work on the pending release.
I am confused and a little nervous. I was just beginning a net-install of 5.4, and suddenly it disappeared from my mirror.
I see that the 5.x tree is now gone. My rsync job from msync.centos is still running, but, my timestamp has rolled back to May 8.
Do I need to do something, NOW, to minimize the lost functionality of my mirror, or are we all along for the ride, now?
--scott
On Thu, 13 May 2010, R. Scott Belford wrote:
I am confused and a little nervous. I was just beginning a net-install of 5.4, and suddenly it disappeared from my mirror.
The wiki counsels against installing across the net for a large variety of reasons
http://wiki.centos.org/HowTos/PXE/InternetInstallation amended to reflect yet another
-- Russ herrold
On Thu, May 13, 2010 at 11:24 AM, R P Herrold herrold@centos.org wrote:
On Thu, 13 May 2010, R. Scott Belford wrote:
I am confused and a little nervous. I was just beginning a net-install of 5.4, and suddenly it disappeared from my mirror.
The wiki counsels against installing across the net for a large variety of reasons
http://wiki.centos.org/HowTos/PXE/InternetInstallation amended to reflect yet another
Respectfully, I tirelessly manage the mirror at UH in Hawaii, mirror.hosef.org, and I expect to eat my own cooking. I've been doing network installs via http for years, and I cache locally on one of the squid servers in my LAN. I had no reason to expect that a machine I manage over bandwidth I moderately influence would stop serving a 5.4 directory because an errant rsync job deleted our entire 5.x tree. I agree that this another reason to be wary of network installs, but I would hardly chalk this one up to yet another example of why not. The failure has nothing to do with the wiki.
-- Russ herrold
--scott
On Thu, 13 May 2010, R. Scott Belford wrote:
umm did you read the new section about 'skew' ?
agree that this another reason to be wary of network installs, but I would hardly chalk this one up to yet another example of why not. The failure has nothing to do with the wiki.
The sever management team are human, working across networks, on remote machines that come and go at times, within limiting constraints ... keeping a disk space 'bloom' under control comes to mind
This is an elaborate dance, and predictable and comtrollable only if, say, we do a few dry runs to get a stripted roll out and check tested for a set of test cases. No doubt there would then be a unanticipated growth in a result archive, and the cost of testing would go unrewarded
Of course this would mean that we have perhaps another week's delay 'til bitflip release. It's not gonna happen
That means 'skew happens', no matter how perfect the tools. I believe Ralph mentioned a 'hung' puppet process
If a person wants to install off a hot mirror, rather than a quiescent local copy, they earned their pain.
-- Russ herrold
On Thu, May 13, 2010 at 5:04 PM, R P Herrold herrold@owlriver.com wrote:
If a person wants to install off a hot mirror, rather than a quiescent local copy, they earned their pain.
If you are a mirror manager, R P Herrold, then I appreciate your time and energy. I am just a guy who takes responsibility for our mirror here in our IP space, and I noticed that the cron jobs I've been running for years without issue suddenly deleted ALL of the 5.x tree, including isos, and set our mirror timestamp to 5-8-2010. It's of no consequence, I've been trying to follow all of today's threads, but I am assuming that it is my duty and responsibility to report it here in case my welcomed information can be helpful to someone else in need. You are not being very welcoming or kind, and I don't understand why not or what is wrong with reporting errors.
With Great Respect and Admiration for everyone who contributes to the success and distribution of CentOS.
-- Russ herrold
--scott
On Fri, May 14, 2010 at 6:55 AM, R. Scott Belford scott@hosef.org wrote:
If you are a mirror manager, R P Herrold, then I appreciate your time and energy. I am just a guy who takes responsibility for our mirror here in our IP space, and I noticed that the cron jobs I've been running for years without issue suddenly deleted ALL of the 5.x tree, including isos, and set our mirror timestamp to 5-8-2010.
I really wonder how that could have happened, at least 5.4 is available everywhere. The issues we have seen here with systems wiping the 5.5 tree was non-updated config files which excluded 5.5 from rsync. Those were 2 machines out of X.
Cheers,
Ralph
On Thu, May 13, 2010 at 11:37 PM, Ralph Angenendt ralph.angenendt@gmail.com wrote:
On Fri, May 14, 2010 at 6:55 AM, R. Scott Belford scott@hosef.org wrote:
If you are a mirror manager, R P Herrold, then I appreciate your time and energy. I am just a guy who takes responsibility for our mirror here in our IP space, and I noticed that the cron jobs I've been running for years without issue suddenly deleted ALL of the 5.x tree, including isos, and set our mirror timestamp to 5-8-2010.
I really wonder how that could have happened, at least 5.4 is available everywhere. The issues we have seen here with systems wiping the 5.5 tree was non-updated config files which excluded 5.5 from rsync. Those were 2 machines out of X.
I wish I could provide more information or that I had been fiddling with things and could point to a PEBCAK. My cron job and script have not changed for at least a year. All I know is that I was shelled in to the mirror, could see a long cron job running from the msync pool, and observed that the 5.5 directory was getting full. I did a network install of 5.4, and when I started another, it failed. I checked the mirror, and and we had a 5-8-2010 timestamp, no directories above 5.3, and all the 5.x directories were empty. I checked my shell, and the cron job had stopped.
I changed to the same upstream mirror we use for debian and ubuntu, and it's slowly refilling. I don't think I'll have any logs of interest. I wish I could report more than an anomaly, but, it's gone and soon forgotten.
Cheers,
Ralph
Aloha
--scott