Re: [systemd-devel] [PATCH] hostnamectl: correct IDs for remote hosts
On Sun, 01.06.14 22:03, Mantas Mikulėnas (graw...@gmail.com) wrote: On Sun, Jun 1, 2014 at 9:01 AM, Lennart Poettering lenn...@poettering.net wrote: On Sat, 31.05.14 18:21, Rico Sagner (sag...@b1-systems.de) wrote: Heya! I think that the two ids would probably be better exposed by PID 1 istelf, instead of hostnamed. It's a bit difficult to come up with a rule which props should be exposed from hostnamed and which ones from PID1, but I think the rule should be something along the lines of if it's more an automatically determined property than a user or vendor chosen one, and if PID 1 knows/caches it anyway then it belongs into PID 1... So, could you please move the machine/boot IDs into PID 1 instead of hostnamed, next to where the Virtualization/Architecture properties are defined? Out of curiosity, wouldn't the existing org.freedesktop.DBus.Peer.GetMachineId() work here? True. That sounds like a good option. And in fact, we actually have a high-level wrapper call for that: sd_get_owner_machine_id(). Lennart -- Lennart Poettering, Red Hat ___ systemd-devel mailing list systemd-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/systemd-devel
Re: [systemd-devel] [PATCH] hostnamectl: correct IDs for remote hosts
On Mon, 02.06.14 13:31, Simon McVittie (simon.mcvit...@collabora.co.uk) wrote: On 01/06/14 20:03, Mantas Mikulėnas wrote: Out of curiosity, wouldn't the existing org.freedesktop.DBus.Peer.GetMachineId() work here? In principle Peer.GetMachineId() returns the D-Bus machine ID /var/lib/dbus/machine-id, which in rare cases won't match the systemd machine ID /etc/machine-id, unfortunately. (Perhaps systemd's Peer implementation prefers the systemd machine ID anyway, though.) yes, we prefer /etc/machine-id. If the ids differ then that's a bug really, and we shouldn't try to work around it... Lennart -- Lennart Poettering, Red Hat ___ systemd-devel mailing list systemd-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/systemd-devel
Re: [systemd-devel] [PATCH] hostnamectl: correct IDs for remote hosts
On 01/06/14 20:03, Mantas Mikulėnas wrote: Out of curiosity, wouldn't the existing org.freedesktop.DBus.Peer.GetMachineId() work here? In principle Peer.GetMachineId() returns the D-Bus machine ID /var/lib/dbus/machine-id, which in rare cases won't match the systemd machine ID /etc/machine-id, unfortunately. (Perhaps systemd's Peer implementation prefers the systemd machine ID anyway, though.) On a distribution that runs the correct setup steps in packages' dpkg postinst, RPM %post or equivalent, those rare cases are: * systemd was installed *before* dbus (i.e. systemd-machine-id-setup was run before dbus-uuidgen) * the first dbus version to be installed was one with Bug #77941, i.e. 1.8.0 or older (that bug was fixed in 1.8.2) In practice, most systems won't get into that situation, because installing dbus before systemd is likely to be more common: the sort of users and vendors who will install systemd are probably the same sort of users and vendors who would already have installed dbus, and systemd packages are likely to have a weak dependency on dbus anyway (e.g. in Debian, systemd Depends: libdbus-1-3 which Recommends: dbus). S ___ systemd-devel mailing list systemd-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/systemd-devel
Re: [systemd-devel] [PATCH] hostnamectl: correct IDs for remote hosts
On Sat, 31.05.14 18:21, Rico Sagner (sag...@b1-systems.de) wrote: Heya! I think that the two ids would probably be better exposed by PID 1 istelf, instead of hostnamed. It's a bit difficult to come up with a rule which props should be exposed from hostnamed and which ones from PID1, but I think the rule should be something along the lines of if it's more an automatically determined property than a user or vendor chosen one, and if PID 1 knows/caches it anyway then it belongs into PID 1... So, could you please move the machine/boot IDs into PID 1 instead of hostnamed, next to where the Virtualization/Architecture properties are defined? Also, please use ay instead of s as type for the property (i.e. a byte-array of the 16 raw uuid bytes instead of an ascii-formatted string). This is how we encoded UUIDs so far as bus properties, see machined's machine uuid property (src/machine/machine-dbus.c) for example. Thanks! Lennart -- Lennart Poettering, Red Hat ___ systemd-devel mailing list systemd-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/systemd-devel
Re: [systemd-devel] [PATCH] hostnamectl: correct IDs for remote hosts
On Sun, Jun 1, 2014 at 9:01 AM, Lennart Poettering lenn...@poettering.net wrote: On Sat, 31.05.14 18:21, Rico Sagner (sag...@b1-systems.de) wrote: Heya! I think that the two ids would probably be better exposed by PID 1 istelf, instead of hostnamed. It's a bit difficult to come up with a rule which props should be exposed from hostnamed and which ones from PID1, but I think the rule should be something along the lines of if it's more an automatically determined property than a user or vendor chosen one, and if PID 1 knows/caches it anyway then it belongs into PID 1... So, could you please move the machine/boot IDs into PID 1 instead of hostnamed, next to where the Virtualization/Architecture properties are defined? Out of curiosity, wouldn't the existing org.freedesktop.DBus.Peer.GetMachineId() work here? Also, please use ay instead of s as type for the property (i.e. a byte-array of the 16 raw uuid bytes instead of an ascii-formatted string). This is how we encoded UUIDs so far as bus properties, see machined's machine uuid property (src/machine/machine-dbus.c) for example. Thanks! Lennart -- Mantas Mikulėnas graw...@gmail.com ___ systemd-devel mailing list systemd-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/systemd-devel