Improved Upgrade Speeds Now Possible!


TL; DR - Change your /etc/ostree/remotes.d/fedora-workstation.conf file to match the following:

$ cat /etc/ostree/remotes.d/fedora-workstation.conf 
[remote "fedora-workstation"]

A common complaint from Silverblue users, especially those outside the United States, has been that the upgrade speeds have been terribly slow. This was ultimately due to how the mirroring of the ostree content was being done. The Fedora Infrastructure team knew of the problem, however there were always bigger fires to put out. Thankfully, a member of our community (and engineer on Fedora CoreOS) stepped up to the challenge and did an exhaustive investigation that resulted in changes around how mirroring was being done for the Fedora ostree content.

@sinnykumari led the effort to find the problems with the current mirroring scheme and helped come up with a new solution that resulted in exceptionally faster speeds for Silverblue upgrades.

You can read about the investigative work that Sinny did in the upstream ostree issue and see the kinds of speed improvements that we have seen.

Ultimately, we want to make the config mentioned above to be the default config for all ostree based hosts, so we are considering shipping a new RPM that owns the default ostree remote config. This would allow us to update the default remote configuration more easily in the future, if needed.

Additionally, we are considering shipping a systemd service that would modify the existing fedora-workstation config to match the newer, faster configuration. These kinds of discussions are occurring in two places:

We would love to hear feedback from the community about how the config changes affects your upgrade experience. Additionally, please feel free to weigh in on the two issues regarding the default remote config file.



Tested and it worked great for me. rpm-ostree upgrade that resulted in no-op took < 10sec to complete.


Just tried it out, works perfectly!


Is this using the technique I read about where it downloads the normal RPMs from the normal mirrors, and then uses those to reconstruct the OSTree commit? Or is this something else?


No, this is using the traditional method of downloading the individual files from the ostree repo. The changes made were around how the certain files were being served up via CDN. There should be more details in the ostree issue that I linked in the original post.

I believe you are probably thinking of rojig, which hasn’t been fully implemented yet, but is being tracked by this upstream issue:


Tested, it works like a charm!


Yeah! It works great!


thanks! works for me, seems to be faster


While people are testing things out, you can also enable HTTP2 by adding http2=true into the remote config.


[remote "fedora-workstation"]