Description of problem: Dashboard has been removed. To me, this is a fundamental and the most useful feature of cockpit. Specifically, dashboard provides a view of all monitored machines on a common timeline. Dashboard makes it very easy to spot, at a glance, any resource problem on one or more machines. All of the other features, that show or change behaviour, whether graphical or text listing, are only available on a per machine basis. Hence there is no great value in cockpit because one might as well connect that machine of concern and perform the same monitoring/adjusting tasks natively on that machine. https://cockpit-project.org/blog/cockpit-234.html The dashboard has been removed Due to limited functionality, the dashboard page has been removed. Connecting to additional hosts is directly available in the Shell without the need for a separate package. The dashboard graphs were not providing enough information to be useful and were implemented with a performance cost on the servers and the browser. Individual machine metrics are available on a separate page on the Overview page. Multi-host monitoring is handled better by other tools (e.g. Grafana), which provide more useful data. I can see the rationale which led to the dashboard being removed, but I wonder if the perspective for consideration was unsufficiently universal. Let me elaborate on a point by point basis: Limited functionality: dashboard did one thing, it did it well; it presented a simulataneous view of multiple machines. Not enough information to be useful: but dashboard did provide sufficient information to be useful for quickly identifying a hot-spot or resource problem on a single page. Connection to additional hosts: connecting to additional hosts, individually via the Shell is indeed possible, but getting information collectively and displaying that graphically is a different matter. Performance cost: any collective monitoring facility is going to have a performance impact on each of the servers being monitored, the real question is if that cost is acceptable given the gains. In my opinion, being able to spot a problem before becoming critical or causing a crash is definitly justifiable. The performance cost on the browser is a different question. Indeed the cost on the browser server is very high. However, by deploying the cockpit dashboard a server dedicated purely to admin and monitoring means good monitoring performance and low impact on the machines being monitored. Individual metrics: these are still available, but the dashboard feature is about collective rather than individaul metrics and hence I fail to understand why this is stated as a reason for removing the dashboard. Better handled by other tools: even if the display might be handled better, the fundamental property of collecting metrics from multiple machines and displaying a (albeit minimal) collective summary was something that cockpit-dashboard did very well, and is unique to cockpit. The suggested Grafana does not appear to do the metric gathering from multiple machines like cockpit does. Cockpit-dashboard provides a self-contained gathering and display of collective metrics without requiring significant effort to comprehend, configure and control the graphic interface. I would welcome counter arguments from the developers or other suggestions to meet this single requirement: display metrics from multiple servers on a single page timeline. Version-Release number of selected component (if applicable): systemctl info cockpit Installed Packages Name : cockpit Version : 234 Release : 1.fc33 Architecture : x86_64 Size : 55 k Source : cockpit-234-1.fc33.src.rpm Repository : @System From repo : updates Summary : Web Console for Linux servers URL : https://cockpit-project.org/ License : LGPLv2+ Description : The Cockpit Web Console enables users to administer GNU/Linux servers using a : web browser. : : It offers network configuration, log inspection, diagnostic reports, SELinux : troubleshooting, interactive command-line sessions, and more. How reproducible: 100% on latest version 234 Steps to Reproduce: 1. https://localhost:9090/dashboard Actual results: Not found! Expected results: Graphs showing a summary of activity for all monitored machines, ideally a vertical stack (with a corresponding horizontal timeline) with one graph each for CPU, Memory, Network and Disk I/O. Additional graphs for % disk space across all monitored machines would be useful (but with the x-axis as disk partition rather than time) That way, a single glance can show immediately if any one machine is suffering a resource problem. Additional info: Graphical views have been gradually changing over the last year or so, and dropping the most useful features. Moving graphs from being a vertical stack to side-by-side removes the ability to visibly align events on a common timeline. When graphs are stacked veritically, the corresponding event for each graph aligns vertically and, for instance, high Disk I/O activity and high Network activity often occur at the same time.
We of course won't stop anyone from using the old package version, or taking the code and maintain it in a separate project. We in the cockpit team won't do that by ourselves, our stanza is still what was written in the release note; but of course we will be glad to help anyone with setting up a separate project. It certainly needs some thought, a better design. Cockpit does a really poor job at being a multi-server monitoring tool (who stares at a browser graph all day long?), and does not aspire to become one. That starts with the insufficient automatic authentication, continues with the rather expensive way of collecting/showing the data, and reaches up to the glaring security problem that every connected host is technically able to do arbitrary things on any other connected host (i.e. the multiple hosts are not isolated from one another). However, we do plan to work on a better Grafana integration, we track that in our internal Jira. I am sorry if that is disappointing.
@Martin Thanks for your response. I have the luxury of multiple large screens (up to 4K) so having a small window permanently open in one corner showing a summary of critical metrics for all machines is of no cost. I don't stare at this browser graph all day :-) , but do look at this frequently, such as when I change tasks or when I am waiting for a compute task to finish. I'm paranoid about security (both in the sense of countering untrusion and in the sense of reliability) so a quick glance at the cockpit-dashboard gives that warm feeling that all is well. And I think I understand the security risks you mention not to be worried about those. I don't really care how you restore this dashboard collective summary, whether via Grafana integration or anything else. But I do look forward to this fundamental feature being reinstated. If you could point me at an equivalent that does the job that cockpit-dashboard used to do, then I'd be equally satisfied. Meanwhile, I rolled back to version 213. I can see, at a glance, that a backup job is running (disk & network activity) and taking longer than normal. Very useful. I shall now look at the backup logs on that server to satisfy myself that all is well.
A while ago I wrote a blog post how to do proper metrics collection and graphing of multiple machines: https://cockpit-project.org/blog/pcp-grafana.html I forgot to update that bug afterwards, sorry. This is what we support now, the old and rather limited Dashboard is not going to come back. I realize that might be inconvenient for some users, but it was really not a thing that we could support in good faith.