Yesterday, we deployed a new version of Copr.
First we migrated to new Fedora Cloud instance, which give us more reliable infrastructure. Previously we used 12 builders, but due problems with old OpenStack we were forced to lower it to 6 builders. Today we are back to 12 builders and we are ready to increase the limit [1]. We are monitoring the new instance and we will very likely increase the number of builders to 20 soon. This should be enough for some time.
Second change is reuse of VM. If builder finishes building of your package and there is in queue some other task from you, then it is picked up and build on the same VM without re-provisioning of that VM. The logic is that you probably do not want to do cracking of your own packages, so we can reuse that VM for yours packages. Packages are still build in different chroot, but the same machine. Of course - if there is only packages of another users in the queue, then the builder VM is terminated and clean one VM is spin up.
Last change is limitation of users. If you send dozens of task to Copr, then you are limited to just 4 parallel builds. Other builders are used for other users next in queue.
Because we have more resources available. I recall that there we some use case where you requested more memory or disk. However I could not find the details. If you have such use case, now is right time to contact me as I can increase the limits if there is good reason. Oh and I forgot that we have new storage for your repositories. Previously we had just 800 GB and we were reaching the roof. On new location we have 4TB and it can be extended if needed.
And the conclusion? You should expect less delay. Not being blocked by other users. And I can finally say, what I wanted to say for long time: Please use Copr more! You can even set up builds in Copr after each commit in your project and Copr should handle that. Hopefully this will allow you to set up Continuous Integration on your project, or make it even better.
[1] During testing of new cloud we tried to spawn enormous numbers of builders and it seems that number 200 is our hard limit.
On 05/20/2015 03:37 PM, Miroslav Suchý wrote:
Yesterday, we deployed a new version of Copr.
First we migrated to new Fedora Cloud instance, which give us more reliable infrastructure. Previously we used 12 builders, but due problems with old OpenStack we were forced to lower it to 6 builders. Today we are back to 12 builders and we are ready to increase the limit [1]. We are monitoring the new instance and we will very likely increase the number of builders to 20 soon. This should be enough for some time.
Second change is reuse of VM. If builder finishes building of your package and there is in queue some other task from you, then it is picked up and build on the same VM without re-provisioning of that VM. The logic is that you probably do not want to do cracking of your own packages, so we can reuse that VM for yours packages. Packages are still build in different chroot, but the same machine. Of course - if there is only packages of another users in the queue, then the builder VM is terminated and clean one VM is spin up.
Last change is limitation of users. If you send dozens of task to Copr, then you are limited to just 4 parallel builds. Other builders are used for other users next in queue.
Because we have more resources available. I recall that there we some use case where you requested more memory or disk. However I could not find the details. If you have such use case, now is right time to contact me as I can increase the limits if there is good reason. Oh and I forgot that we have new storage for your repositories. Previously we had just 800 GB and we were reaching the roof. On new location we have 4TB and it can be extended if needed.
And the conclusion? You should expect less delay. Not being blocked by other users. And I can finally say, what I wanted to say for long time: Please use Copr more! You can even set up builds in Copr after each commit in your project and Copr should handle that. Hopefully this will allow you to set up Continuous Integration on your project, or make it even better.
[1] During testing of new cloud we tried to spawn enormous numbers of builders and it seems that number 200 is our hard limit.
Mirek, great announcement - thank you. With the infrastructure changes are there any plans for supporting additional architectures?
radek
Dne 21.5.2015 v 13:41 Radek Vokal napsal(a):
With the infrastructure changes are there any plans for supporting additional architectures?
Yes. I plan to return soon to PPC64LE arch. AARM64 is planned for long time. But is is complicated and not big priority - no ETA right now.
On Thu, 21 May 2015 15:06:53 +0200 Miroslav Suchý msuchy@redhat.com wrote:
Dne 21.5.2015 v 13:41 Radek Vokal napsal(a):
With the infrastructure changes are there any plans for supporting additional architectures?
Yes. I plan to return soon to PPC64LE arch.
We will need to find hardware for that.
AARM64 is planned for long time. But is is complicated and not big priority - no ETA right now.
Also for that. ;)
We do have armv7 hw in the same network, so it might be possible to add, but it will not be super fast.
kevin
copr-devel@lists.fedorahosted.org
-
Kevin Fenzi -
Miroslav Suchý -
Radek Vokal