4 ms·
I am curious why you'd need to run monitoring on your own instances, since the assumption is that given a set of resources (say X RAM and Y CPU), you can suppor
by why-el 10y ago
I am curious why you'd need to run monitoring on your own instances, since the assumption is that given a set of resources (say X RAM and Y CPU), you can support Z users. At least I remember they advertised such a number.
Now, if they somehow get to have these monitoring numbers automatically sent back to a centralized server where they can gather more insights about various installations, that would be killer, but I am not sure about the feasibility, giving the privacy implications (although plenty of open source software does it).
- sytse 10y ago> I am curious why you'd need to run monitoring on your own instances, since the assumption is that given a set of resources (say X RAM and Y CPU), you can support Z users. At least I remember they advertised such a number. We indeed publish a list of requirements https://docs.gitlab.com/ce/install/requirements.html https://docs.gitlab.com/ce/install/requirements.html But this is not enough. Not all hardware is created equal (one core is faster than another) and not all users are created equal (one might push many more branches per day). So to help people with the performance of their GitLab server we think the integrated metrics will be a large benefit. But another important reason to add this is to ensure that GitLab has metrics about applications that are deployed with GitLab. > Now, if they somehow get to have these monitoring numbers automatically sent back to a centralized server where they can gather more insights about various installations, that would be killer, but I am not sure about the feasibility, giving the privacy implications (although plenty of open source software does it). We're very conscious of the privacy implications of sending data about GitLab usage back. We're doing that not with Prometheus but with a usage ping in GitLab EE. We're working on bringing that usage ping to CE, for our reasoning see https://gitlab.com/gitlab-com/www-gitlab-com/merge_requests/4428 https://gitlab.com/gitlab-com/www-gitlab-com/merge_requests/... BTW People can use Prometheus monitoring in three forms: on-premise centralized, on-premises federated, and as a SaaS On-premises centralized is what we just shipped. Prometheus is easy to federate, where some of the metrics of a server are included in another one. In the future we might give every deployed application its own Prometheus server in a pod. If you want to use Prometheus as a SaaS we recommend Weaveworks https://www.weave.works/solution/prometheus-monitoring/ https://www.weave.works/solution/prometheus-monitoring/
- why-el 10y agoGreat, thank you for the detailed answer. We use and love Gitlab. ;)
- netingle 10y ago> If you want to use Prometheus as a SaaS we recommend Weaveworks https://www.weave.works/solution/prometheus-monitoring/ https://www.weave.works/solution/prometheus-monitoring/ Which is also open source! https://github.com/weaveworks/cortex https://github.com/weaveworks/cortex