When running yum in a situation where there's a lot of data churn on the mirrors, I often encounter cases where all mirrors don't have the same set of metadata stuff available to download, and it causes yum download errors.
I tried to make yum specifically use the public mirror from the metalink that has the highest priority. I know this mirror is up to date, I maintain it :) So there's no reason for my boxes not to use it... I verified that the "failover" option is correctly set to "priority".
But while running a "yum update", I noticed that yum still tried to download metadata urls from other mirrors, that were not uptodate, which caused download errors. I searched again, and found that urlgrabber launches by default up to five connections in parallel. This value can be controlled by the max_connections option in yum.conf, so I decreased this value to "1".
And again, I noticed that yum still tried to use other mirrors. I found that urlgrabber also uses internally a "default_speed" value to discard the current mirror if the estimated download speed is below the "default_speed". This mirror shuffling occurs _even_ if the current download did succeed. This default_speed is relatively high (1e6 # 1Mbit, line 1002 in grabber.py).
According to my speed tests, I think this value expresses bytes rather than bits, so the default_speed is rather closer to 10Mbit instead. With a download speed of 380kB/s (kBytes) displayed by the yum progress bar, the estimated speed for my mirror around line 2291 of grabber.py is 510454, which is below the default_speed, and causes another mirror to be selected.
Also, when the box is on a slow network, below this default_speed, no mirror can achieve the download speed expected by urlgrabber, and that certainly may cause many "mirror rotations", useless in this case.
I suggest to fix the default_speed value in urlgrabber (125e4 for 1Mbit), and ideally, this value could be added as an option in the yum config file (for people on permanently slow networks) ?
doh, 125e3 for 1Mbit
Right, it's 1MBps, not 1Mbit. default_speed is used only as the first estimate for never-tried-before mirrors, but when there's a lot of them it might still cause problems..
OTOh when set too low, Yum uses only first few mirrors even if there are faster ones available. And while the 1st condition fixes automatically once all mirrors have been tried (and found to be inferior), the 2nd one is permanent... that's why I set the default speed relatively high. Anyway, it's on the top of the list of urlgrabber options that should be exposed in yum.conf. Eg:
This is the speed estimate for mirrors Yum has not used yet. Setting this slightly below the speed of your favorite mirror makes Yum much less promisuous. The default is 1MBps.
python-urlgrabber-3.10.1-0.fc20 has been submitted as an update for Fedora 20.
* should fix your issue,
* was pushed to the Fedora 20 testing repository,
* should be available at your local mirror within two days.
Update it with:
# su -c 'yum update --enablerepo=updates-testing python-urlgrabber-3.10.1-0.fc20'
as soon as you are able to.
Please go to the following url:
then log in and leave karma (feedback).
python-urlgrabber-3.10.1-0.fc20 has been pushed to the Fedora 20 stable repository. If problems still persist, please make note of it in this bug report.