Andreas: "systemctl --type=service" shows only all active services, while Services in Cockpit shows all active and inactive services.

The command line interface needs --all in order to also show the inactive services.
Since a smooth transition from web ui to cli would be good, it could make sense to do a similar thing for the web ui. Show only active services by default, and a button to allow to also see inactive services.

This includes the ability to view device- and service- specific log entries, time lines, and support for searching and filtering.

Just use the suffix of the UDisks path directly. This requires unexporting old stuff before exporting new stuff to avoid conflicts.

This is arguably udisks job:

​https://bugs.freedesktop.org/show_bug.cgi?id=60113

Partitions are created and formatted with one operation, while logical volumes are created in a separate step from formatting them.

Formatting needs many parameters, but partition creation only needs the size, so it made sense to put them into one dialog. (I think I copied that from GNOME Disks.)

Logical volume creation might be complex in terms of UI, just like raid creation, so it is probably better off as a seperate step from formatting, which is also complex.

To unify this, we might separate partition creation from formatting; or we might combine lv/raid creation and formatting and just pay attention that we have a manageable UI for this complex task (similar to Anaconda, maybe).

Right now, when specifying a exact point in time when to do a shutdown or restart (as opposed to specifying a delay), the UI mimics the venerable /usr/bin/shutdown" interface.

We should have a nice date / time picker there, with calendar and everything.

We should select one from the symlinks, but care must be taken to only use valid ones.

(For example, if two filesystems have the same label L, they will both have /dev/disk/by-label/L among their symlinks, but the actual symlink of course only points to one device node. This should be a known problem and a bit of Googling is probably in order.)

Need to review and complete it.

• Make it cleanly cancellable or disable cancelling
• Also represent externally started pvmoves as jobs
• Allow interrupted pvmoves to be resumed and aborted

The content on the Shutdown page is very wide. Maybe we can make this a dialog instead?

Service page should allow sending signals to processes

Andreas: I had a chat with Lennart where he suggested we should hide all but the services in the service pane, since it was a bit too technical and a bit overkill for Cockpit to also list the other types.

We can have information about Wants, WantedBy, Before, After, etc, complete with links to the other services.

We can also have information about conflicts, such as when enabling a unit file fails because its alias is already installed by some other unit file.

When a lost connection comes back, the retrieve a complete object / interface / properties tree from the server and merge it with the remembered state from before the loss, synthesizing signals as appropriate.

This needs testing.

Cockpit should be self-documenting.

Service page should show resource graphs

Maybe by exposing the mtime of the image as a D-Bus property and using LocalStorage as a cache.

Not just whether someone is currently logged or when the last time was.

We have some network monitoring support, but configuration has not been covered at all yet. For this, we need to talk to NetworkManager.

mvollmer: I expect it to seek to the start of the given boot, but it doesn't seem to have any effect. There is a FIXME to that effect in the systemd sources, but it still looks like the necessary code is there.

mvollmer: When selinux is 'enforcing' and cockpitd and realmd have been started in certain ways (which include both being auto-launched by D-Bus), cockpitd does not receive replies from realmd, and times out.

This happens to all replies, such as the initial call to retrieve property values.

I am not entirely sure which ways to start them work, and which don't. I think if it at least one of them is started from the command line, they can communicate. All other ways fail IIRC, like D-Bus autolaunching, systemctl start, and D-Bus autolaunching via systemd.

Which one starts first also doesn't seem to make a difference.

I haven't observed this with any other D-Bus service that cockpitd uses.
We use our own custom realmd package, and I might have broken it.
In any case, disabling selinux makes everything work.

The cgroup-show code is a butchered version of code in systemctl. It would be good to share this code with the systemd project, maybe by simply having identical source files, or maybe by making libsystemd-shared a externally useable library.

This is also related to logs: the journald API is only really accessible from inside the systemctl source tree, if I am not mistaken. We might want to change that, or not.

Lennart proposed to collapse instanced services (like getty's) into one. This would help keep our list of services from getting too long.

When SSH doesn't know the host key of the machine that we want to run cockpit-agent on, we should allow the user to explicitly trust this key.

In particular, we want to make sure that we only pass acceptable passphrases to udisks so that we don't get cryptic error messages.

Currently entrys in the Journal are named after their commandline names, but it might be a good idea to name them after their human names instead.

So:

• cockpit-ws -> Cockpit Web Server
• NetworkManager -> Network Manager
• setroubleshoot -> SELinux Troubleshooter

etc. Is this possible?

We parse unit files ourselves, so we might have to give up on that or tweak it.

Currently, it just does a linear search over all log entries. We should cache the current list of day headers and their positions, and then only search that on each scroll event.

The current Services UI is almost pure systemd, which is nice and powerful, but we might be able to put some nice abstractions over it with little effort.

For example, the user can choose the default target by enabling exactly one of the targets that install themselves as an alias for "default.target". Once you know this, it is fairly easy to do with raw systemd. Still, we can make a "Default Server Profile" page that presents these targets as radio buttons.

Cockpit attempts to cancel join and leave operations, but support in realmd is still incomplete.

Realmd has support for cancelling operations in the Example provider. This is not yet in upstream, and there is no patch yet to support cancelling of operations in the real providers.

See ​https://bugs.freedesktop.org/show_bug.cgi?id=57614

(For better or worse, depending on who you talk to) LVM is an important part of Linux on the server, so Cockpit should support it. This is complicated by the fact that LVM does still not integrate very well with the modern udev world, and is not supported by udisks.

Our options for supporting it are:

• bypass udisks
• get lvm support into udisks

We should just use the journal for this.

Some dialogs can be closed by clicking outside of them, and some only via an action button.

Right now, a big ugly "Disconnected" popup pops up with a "Reconnect" button that doesn't seem to do anything. Let's hear it from the designers how this should be done nicely.

Migrated from another bugzilla: Instead of raw POSIX functions.

glusterfs is a distributed filesystem, which can be used in nas. Supporting this could be done either by adding support for it to udisks, or by directly using the glusterfs api. Adding it in udisks would have the advantage that we get desktop support for free (not sure how relevant that is).

jQuery seems to do something smart when the browser window is resized and a popup is open. For the "Join a Domain" dialog, resizing the browser window results in all fields being reset to the default domain, as if the dialog had just been opened.

It would be desirable to have some support for monitoring performance. The framework called Performance Co-Pilot (pcp), which could serve as a basis for this.

​http://oss.sgi.com/projects/pcp/index.html

It is possible to do 'stupid' things with RAIDs in Cockpit that can't be undone with it.

When starting a RAID fails, this might leave inactive raids behind that udisks2 doesn't seem to handle well. Those arrays can keep disks busy, but udisks2 will not allow inactive arrays to be stopped. (It either ingnores them or pretends they are not running.)

See ​https://bugs.freedesktop.org/show_bug.cgi?id=60359

Also, udisks2 sometimes gets out of sync and shows RAIDs that don't exist anymore at all, and will disappear from udisk2 after a restart of it.

Maybe there are other things...

Report on selected services on the multi-server dashboard.

Go over

​http://stackoverflow.com/questions/549/the-definitive-guide-to-forms-based-website-authentication

and see whether there is anything in it for us.

Right now, the list is hard coded. It should be configurable.

In the extreme, we might offer support for discovering packages that would add more filesystem types, pretty much like totem can go looking for video codec packages.

POSTing about 4000 bytes to /authrequest takes about a second, for example. The slowness might be outside of the request handler actually.

Continue journal queries after disconnection

Enrolling in an AD domain was one of them. We do have support for it, but we may want to revisit it and make sure it makes sense.

Use js event delegation instead of binding each button individually

It would be nice to get a quick overview of the characteristics of the machine already on the Dashboard. This would include the most crucial parts of information of the machine.

Some quick ideas on what this would be:

• Host name
• Operating System
• Domain
• Uptime

It would also be nice to put the identifying picture here.

We allow ourselves the use of GCC extensions, such as attribute((cleanup)) to simplify control flow due to temporary allocations.

This should be documented, including the replacement of the common "if (error) goto out;" idiom with "if (error) return;"

Then all code should be cleaned up to use this pattern. This might need extensions to libgsystem, or a more local approach.

Represent mdraid syncing and scrubbing as jobs

The ServerTimeUpdate signal is specified, but neither implemented nor used.