Re: DNF and PackageKit background data usage
Vít Ondruch wrote: > While more convenient way is "dnf history undo last", which used to work > with YUM, but is broken in DNF, since it does not keep the cache around, > as Kevin pointed out. YUM was also set up with keepcache=0 in the Fedora package, against the upstream default of keepcache=1. In addition, even with keepcache=1, it would not find the package to revert to if it was no longer in the repos, I had to use rpm -Uvh --oldpackage by hand. No idea whether this is better in DNF, I didn't have the situation with DNF yet (and besides, I usually update with PackageKit, which has its own cache, and which doesn't do downgrades at all, leaving me with only the rpm -Uvh --oldpackage approach). Kevin Kofler ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
Adam Williamson wrote: > (Obviously the *right* way to do that general approach is some form of > snapshotting... No. If one update is broken, I want to revert that particular update, not my entire system. The Window$ "system restore" approach is just broken and a horrible idea to reimplement. > hi, ostree). And ostree is even worse, copying the smartphone approach of having to update the whole take-it-or-leave-it image at every change, and also losing the ability to customize the package set. And reverting is just as broken as in the client-side snapshot approach. There is no way I am going to use an ostree any time soon. Kevin Kofler ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Sun, 06 Nov 2016 23:57:07 - Samuel Rakitničanwrote: > > On Sat, 2016-11-05 at 05:01 +0100, Kevin Kofler wrote: > > > > "hunt down"? > > > > koji download-build (nvr) works fine. Just a side note: Kevin Kofler didn't say that. Please watch your attributions. > I think you are missing the point here. Reverting dnf history does > not work if packages are missing from repository. Besides "koji" > command is not installed on my machine, and hence not on default > install. Your suggestion for users is to install "RPM Development > Tools" group? > > I don't see how this command would help in what I usually do when I > need stuff that is missing from repo, other then a little bit more > convenience perhaps. That is to locate the packages in > koji.fedoraproject.org and pass the urls to curl. This command does that part... locates the package on koji and passes the urls to curl and downloads them for you. kevin pgpMkr2Y6_Tq4.pgp Description: OpenPGP digital signature ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
Dne 5.11.2016 v 07:10 Adam Williamson napsal(a): > On Sat, 2016-11-05 at 05:01 +0100, Kevin Kofler wrote: >> Otherwise, you have to hunt down the old builds >> directly in Koji > "hunt down"? > > koji download-build (nvr) works fine. While more convenient way is "dnf history undo last", which used to work with YUM, but is broken in DNF, since it does not keep the cache around, as Kevin pointed out. Vít ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 11/05/2016 05:01 AM, Kevin Kofler wrote: > Theodore Papadopoulo wrote: >> Add to this that these caches seem to be never cleaned, so that they >> grow up very large up to the point they prevent updating the system. I >> just found that the packagekit cache on my machine is about 7Gb !!! >> Probably because I do use packageit at all (dnf cache is much smaller at >> ~700 Mb). > > IMHO, this is a feature, and the DNF default behavior is the bug, a critical > data loss bug even! Given that Fedora does NOT keep old updates on its > mirrors, the only convenient way to revert a broken update is to keep ALL > old updates cached locally. Otherwise, you have to hunt down the old builds > directly in Koji. The version from the GA release is more often than not TOO > old, you really want the previous update. > > I consider it very broken to delete files that I might still need without > even asking me. Setting keepcache=1 is the one of the very first things I do > to any new Fedora installation. Well, if I understand well. You tell us that because fedora repos do not keep versions of old packages, all fedora machine have to keep them locally ?? And that just in case they want to revert a package. Honestly, why on earth would one ask every fedora machine to have copies of something that is not even kept in the original repo ? To me, if the other solutions that were proposed are not OK for some reason, it is the fedora repo policy of not keeping old updates that is wrong Given the usual thoughtfull posts you usually make in this list, I fear I have not well understood what you say or I'm missing some point Otherwise, let's not correct a bug by creating other bugs Theo. smime.p7s Description: S/MIME Cryptographic Signature ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 2016-11-07, Adam Williamsonwrote: > I don't believe the vast majority of people actually install old > package versions *ever*, unless they get very explicit instructions to > do so either because there was a big fail of some kind and we did our > usual emergency drill, or they hit some kind of very specific bug and > got very specific instructions to do so in Bugzilla or something. > Becuse they don't know how, they are lost without explicit instructions. > I'd say. I think the > idea that everyone needs to have a gigantic cache of every package > version they've ever installed stored locally Only few (usually one) previous ones are enough. If a buggy package reaches stable repository, downgrading this one package is usually the easiest temporary fix until package maintainer provides a fix. I recommend to look into repository of a rolling distribution. > they need to install an old one some time is...wrong. (Obviously the > *right* way to do that general approach is some form of > snapshotting...hi, ostree). Of course. User is waiting for a bug fix for his hot package, na atomic update delivers the fixed package with unrelated broken package. The result is as terrible as now. User get everything or nothing. That's maybe fine for laic users who want only to use. But unaccaptable for advanced users or developers. Is Fedora still for developers? -- Petr ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 2016-11-05, Adam Williamsonwrote: > On Sat, 2016-11-05 at 05:01 +0100, Kevin Kofler wrote: >> Otherwise, you have to hunt down the old builds >> directly in Koji > > "hunt down"? > > koji download-build (nvr) works fine. DNF history lists binary package NVRs. Koji works with source package NVRS. You cannot programatically map from the first one to the second. Moreover users use package manager. They do not know anything about Koji. -- Petr ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Sun, 2016-11-06 at 23:57 +, Samuel Rakitničan wrote: > > On Sat, 2016-11-05 at 05:01 +0100, Kevin Kofler wrote: > > > > "hunt down"? > > > > koji download-build (nvr) works fine. > > I think you are missing the point here. Reverting dnf history does > not work if packages are missing from repository. Besides "koji" > command is not installed on my machine, and hence not on default > install. Your suggestion for users is to install "RPM Development > Tools" group? I don't believe the vast majority of people actually install old package versions *ever*, unless they get very explicit instructions to do so either because there was a big fail of some kind and we did our usual emergency drill, or they hit some kind of very specific bug and got very specific instructions to do so in Bugzilla or something. I should probably have said that too to make it more clear, but basically, in the rather unusual case that you actually do need to get an old package, the koji CLI does the job fine, I'd say. I think the idea that everyone needs to have a gigantic cache of every package version they've ever installed stored locally just on the off-chance they need to install an old one some time is...wrong. (Obviously the *right* way to do that general approach is some form of snapshotting...hi, ostree). -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
> On Sat, 2016-11-05 at 05:01 +0100, Kevin Kofler wrote: > > "hunt down"? > > koji download-build (nvr) works fine. I think you are missing the point here. Reverting dnf history does not work if packages are missing from repository. Besides "koji" command is not installed on my machine, and hence not on default install. Your suggestion for users is to install "RPM Development Tools" group? I don't see how this command would help in what I usually do when I need stuff that is missing from repo, other then a little bit more convenience perhaps. That is to locate the packages in koji.fedoraproject.org and pass the urls to curl. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Sat, 2016-11-05 at 05:01 +0100, Kevin Kofler wrote: > Otherwise, you have to hunt down the old builds > directly in Koji "hunt down"? koji download-build (nvr) works fine. -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
Adam Williamson wrote: > 1) Both dnf and GNOME Software / PackageKit default to performing > fairly data-hungry transactions in the background, out of the box, > without telling you about it. GNOME's is particularly bad, as it will > happily download available updates in the background, which can be > gigabytes worth of data. DNF only updates its metadata caches (on a > systemd timer), but even that could be behaviour that users in certain > circumstances really really do not want. IMHO, both of these are misbehaviors: For the dnf case: I am not convinced that dnf's cronjob that always updates metadata rather than updating it when actually needed is a win. Sure, it helps interactivity, but at the expense of a lot of unnecessary downloads, especially if you use dnf rarely or not at all (because you use PackageKit). IMHO, updating metadata periodically only makes sense in user interfaces that are actually offering updates to the user. Updating the metadata eagerly just in case somebody will run the command later is a bad idea. For the GNOME Software case: IMHO, downloading entire packages in the background without asking, before even telling the user that there are updates at all, is totally unacceptable (on ANY type of Internet connection), and I am really glad that plasma-pk-updates does not do such a boneheaded thing. (I do think that updating metadata to offer updates is a helpful thing on an unmetered connection, but you are right that there needs to be a good way to avoid doing so without an explicit request on metered connections. I also don't think that the metered/unmetered property can reliably be autodetected as GNOME Software tries to do.) Kevin Kofler ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
Theodore Papadopoulo wrote: > Add to this that these caches seem to be never cleaned, so that they > grow up very large up to the point they prevent updating the system. I > just found that the packagekit cache on my machine is about 7Gb !!! > Probably because I do use packageit at all (dnf cache is much smaller at > ~700 Mb). IMHO, this is a feature, and the DNF default behavior is the bug, a critical data loss bug even! Given that Fedora does NOT keep old updates on its mirrors, the only convenient way to revert a broken update is to keep ALL old updates cached locally. Otherwise, you have to hunt down the old builds directly in Koji. The version from the GA release is more often than not TOO old, you really want the previous update. I consider it very broken to delete files that I might still need without even asking me. Setting keepcache=1 is the one of the very first things I do to any new Fedora installation. Kevin Kofler ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Thu, Nov 3, 2016 at 10:13 AM, Gerd Hoffmannwrote: > Hi, > >> > The metalinks already provide checksums and timestamps for the >> > metadata. So instead of dumbly going out and re-downloading the entire >> > metadata at hardcoded intervals, couldn't we rather just check if the >> > 'latest' metadata (according to the metalink) has changed since the >> > last time we ran, or something along those lines, and only actually >> > download the metadata if so? Wouldn't that save us a lot of wasted >> > metadata downloads? >> >> I suppose that's what is supposed to happen. And of course, now that I >> try to reproduce it, that is exactly what happens. >> >> I know I've seen it redownload the main repo occasionally but perhaps >> its just been bugs since fixed. Or then the main repo actually changed. >> But perhaps having bugs is the more likely explanation... Yeah, it was bug in DNF handling metalinks. It should have been fixed. > dnf seems to reload the complete metadata if you touch the repo file. > I see this happening after running fedrepos (to configure mirror and > proxy) on fresh installs. Not that it bothers me much in that specific > case as it re-downloads the metadata from local squid cache. But just > re-downloading repomd.xml surely should have been enough. Is this fedrepos changing URL and/or type of URL (metalink, baseurl or mirrorlist)? Honza ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
Hi, > > The metalinks already provide checksums and timestamps for the > > metadata. So instead of dumbly going out and re-downloading the entire > > metadata at hardcoded intervals, couldn't we rather just check if the > > 'latest' metadata (according to the metalink) has changed since the > > last time we ran, or something along those lines, and only actually > > download the metadata if so? Wouldn't that save us a lot of wasted > > metadata downloads? > > I suppose that's what is supposed to happen. And of course, now that I > try to reproduce it, that is exactly what happens. > > I know I've seen it redownload the main repo occasionally but perhaps > its just been bugs since fixed. Or then the main repo actually changed. > But perhaps having bugs is the more likely explanation... dnf seems to reload the complete metadata if you touch the repo file. I see this happening after running fedrepos (to configure mirror and proxy) on fresh installs. Not that it bothers me much in that specific case as it re-downloads the metadata from local squid cache. But just re-downloading repomd.xml surely should have been enough. cheers, Gerd ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 11/02/2016 08:15 PM, Adam Williamson wrote: On Sun, 2016-10-30 at 08:50 +0200, Panu Matilainen wrote: On a related note, why on earth is the main Fedora repo set to expire every two weeks? (and its -source and -debuginfo every week??) It's not supposed to change *ever* for a released distro version now is it? You know, this may be me being dumb, but this question prompted me to wonder...can't we do something better than metadata expiry and complete re-download for repos that use the mirrormanager metalinks? The metalinks already provide checksums and timestamps for the metadata. So instead of dumbly going out and re-downloading the entire metadata at hardcoded intervals, couldn't we rather just check if the 'latest' metadata (according to the metalink) has changed since the last time we ran, or something along those lines, and only actually download the metadata if so? Wouldn't that save us a lot of wasted metadata downloads? I suppose that's what is supposed to happen. And of course, now that I try to reproduce it, that is exactly what happens. I know I've seen it redownload the main repo occasionally but perhaps its just been bugs since fixed. Or then the main repo actually changed. But perhaps having bugs is the more likely explanation... - Panu - ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
Yeah, sorry, looks like either hyperkitty or I messed up. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
El 2/11/2016 12:16 p. m., "Adam Williamson"escribió: > > On Sun, 2016-10-30 at 08:50 +0200, Panu Matilainen wrote: > > On a related note, why on earth is the main Fedora repo set to expire > > every two weeks? (and its -source and -debuginfo every week??) It's not > > supposed to change *ever* for a released distro version now is it? > > You know, this may be me being dumb, but this question prompted me to > wonder...can't we do something better than metadata expiry and complete > re-download for repos that use the mirrormanager metalinks? > > The metalinks already provide checksums and timestamps for the > metadata. So instead of dumbly going out and re-downloading the entire > metadata at hardcoded intervals, couldn't we rather just check if the > 'latest' metadata (according to the metalink) has changed since the > last time we ran, or something along those lines, and only actually > download the metadata if so? Wouldn't that save us a lot of wasted > metadata downloads? +1 and +1 to have dnf and packagekit sharing the same metadata Why do not include a note in Fedora docs or release notes alerting users about the default configuration and posible high data download and instructions to disable background process if the user want? This could help in the very short time. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Wed, 2016-11-02 at 10:49 +, Christian Stadelmann wrote: > > On 10/30/2016 03:26 AM, Adam Williamson wrote: > > > > Fedora updates so often that attempts to pre-download anything updates > > related are pointless. Chances are you > > a) waste gobs of bandwidth downloading that changing data over and over > > again without ever using it > > b) when you actually *do* use it, it's typically already outdated and > > the update you're looking for isn't there, so you need to force manual > > refresh anyway Your quoting is incorrect. I did not write any of this. -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Sun, 2016-10-30 at 08:50 +0200, Panu Matilainen wrote: > On a related note, why on earth is the main Fedora repo set to expire > every two weeks? (and its -source and -debuginfo every week??) It's not > supposed to change *ever* for a released distro version now is it? You know, this may be me being dumb, but this question prompted me to wonder...can't we do something better than metadata expiry and complete re-download for repos that use the mirrormanager metalinks? The metalinks already provide checksums and timestamps for the metadata. So instead of dumbly going out and re-downloading the entire metadata at hardcoded intervals, couldn't we rather just check if the 'latest' metadata (according to the metalink) has changed since the last time we ran, or something along those lines, and only actually download the metadata if so? Wouldn't that save us a lot of wasted metadata downloads? -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 10/31/2016 05:10 PM, Adam Williamson wrote: > On Mon, 2016-10-31 at 15:09 +, Richard Hughes wrote: >> >> I think this kind of issue is really fixed the hard way, i.e. fixing >> bugs and adding unimplemented features rather than just adding complex >> UI workarounds. > > I think making it work as best as possible without interaction is > great, but I'd argue that fundamentally this needs interaction. At a > bare minimum the UI for users to mark connections as metered or > unmetered needs to be there. Is there actually UI for this already? I > could not find it. > > The question of what the default state should be is an interesting one, > and I bet your take on it differs substantially depending on your level > of 'internet connection privilege'. ;) While all of this is true, let's not forget that having two completely different mechanisms downloading the same information in cache is the really bad thing... smime.p7s Description: S/MIME Cryptographic Signature ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
> On 10/30/2016 03:26 AM, Adam Williamson wrote: > > Fedora updates so often that attempts to pre-download anything updates > related are pointless. Chances are you > a) waste gobs of bandwidth downloading that changing data over and over > again without ever using it > b) when you actually *do* use it, it's typically already outdated and > the update you're looking for isn't there, so you need to force manual > refresh anyway From my experience, this is just plain wrong. Even with updates-testing enabled there is hardly any package getting updated twice a week. And updating your OS less than once a week is careless if you run it every single or second day. So, no, automatic update downloads are the only right thing to do if you can't guarantee having no security bugs. > On a related note, why on earth is the main Fedora repo set to expire every > two weeks? (and its -source and -debuginfo every week??) It's not supposed to > change *ever* for a released distro version now is it? +1 ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Tue, Nov 1, 2016 at 9:59 AM, Miroslav Suchýwrote: > Dne 31.10.2016 v 12:55 Tom Hughes napsal(a): >> The problem, as I believe has been repeatedly explained, is that you don't >> really have any idea whether the connection >> is metered. Yes you may be trying to guess by considering things like 3G >> connections as metered but that is an extremely >> poor proxy for whether a connection is metered. > > https://bugzilla.redhat.com/show_bug.cgi?id=1112230 Then new DNF config option ~sync_metadata_unmetred_conn_only, set to True by default could be a solution. I am just skeptical about DNF depending on NetworkManager. In the long term we would like to share cache between PackageKit and DNF + download just metadata changes [1] which is the root cause and should reduce data transfer in general. Honza [1] https://github.com/rh-lab-q/deltametadata-prototype/wiki/Deltametadata-of-repo-md-files ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
NetworkManager's Metered property (was: Re: DNF and PackageKit background data usage)
On Mon, 2016-10-31 at 12:31 -0400, Matthew Miller wrote: > On Mon, Oct 31, 2016 at 09:23:06AM -0700, Adam Williamson wrote: > > > > For that matter, I'm fairly sure I've seen background update > > downloading happen when I was using an Android wifi tether > > connection. I'm pretty sure I remember it blowing my data cap one > > month when I was using my laptop on the bus. > > As far as I can see, the default is "unknown". You can set > "CONNECTION_METERED=yes" in the > /etc/sysconfig/network-scripts/ifcfg-[whatever] file to change it on > a > per-device basis. > > The man page for NetworkManager.conf doesn't document this as one of > the properties which can have its default overridden (and says that > anything not documented can't be overridden). I took that at its > word; > possibly worth testing anyway. :) Hi, In NetworkManager this works as follows: (1) There is the per-connection property "connection.metered" [1]. It can be "yes", "no", "unknown". "unknown" is the default and prompts NM to do some basic heuristics. $ nmcli -f connection.metered connection show $NAME $ nmcli connection update $NAME connection.metered yes (2) Then there is the Device's metered property in the D-Bus API [2]. If you activate a connection that has "connection.metered" explicitly set to "yes" or "no", that's it. Otherwise, NM will set the device property to "guess-yes" or "guess- no". For example, - WWAN connections are "guess-yes" - if there is a DHCP option ANDROID_METERED, it is "guess-yes". See [4]. $ nmcli -f GENERAL.METERED device show $DEVICE (3) Finally, there is the Manager's metered property [3]. This combines the metered property of multiple devices into one. PackageKit would only care about this $ busctl get-property org.freedesktop.NetworkManager /org/freedesktop/NetworkManager org.freedesktop.NetworkManager Metered best, Thomas [1] https://developer.gnome.org/NetworkManager/stable/nm-settings.html#nm-settings.property.connection.metered [2] https://developer.gnome.org/NetworkManager/stable/gdbus-org.freedesktop.NetworkManager.Device.html#gdbus-property-org-freedesktop-NetworkManager-Device.Metered [3] https://developer.gnome.org/NetworkManager/stable/gdbus-org.freedesktop.NetworkManager.html#gdbus-property-org-freedesktop-NetworkManager.Metered [4] https://cgit.freedesktop.org/NetworkManager/NetworkManager/tree/src/devices/nm-device.c?id=58482a8feccc58db8d5a7eae9a004ce691e0c502#n10272 signature.asc Description: This is a digitally signed message part ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
Dne 31.10.2016 v 12:55 Tom Hughes napsal(a): > The problem, as I believe has been repeatedly explained, is that you don't > really have any idea whether the connection > is metered. Yes you may be trying to guess by considering things like 3G > connections as metered but that is an extremely > poor proxy for whether a connection is metered. https://bugzilla.redhat.com/show_bug.cgi?id=1112230 -- Miroslav Suchy, RHCA Red Hat, Senior Software Engineer, #brno, #devexp, #fedora-buildsys ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, 2016-10-31 at 18:57 -0700, Adam Williamson wrote: > Huh, since F23 or so I've found the refresh button to be pretty > reliable. It does seem to actually force Software to go refresh the > metadata and download available updates, now. I think in the past it > just 'forced' a run of the refresh timer, which doesn't always > actually > go and check for updates, it has a bunch of conditionals that decide > whether it will. But that got changed at some point. The problem is that if the update fails for any reason, it reports that everything is up to date, last check for updates at the time the update failed. It is https://bugzilla.gnome.org/show_bug.cgi?id=763566. Note that bug was reported with F23, you can tell by the desktop background in my screenshot. It's still broken today I think. In practice, updates quite often fail for me. I think PackageKit gives up if a mirror is missing a package, which is not an infrequent occurrence, whereas dnf just moves on to the next mirror. (Or maybe that got fixed?) I've seen a post-install update fail five times in a row before. :( Michael ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, 2016-10-31 at 19:57 -0500, Michael Catanzaro wrote: > I think it's a GNOME Software bug that it says up to date when it's > not. I've complained about that for years. It will happily tell you > that your freshly-installed Fedora system is up to date months after > release, before the first update, often even after you manually check > for updates by pressing the Refresh button. :) Huh, since F23 or so I've found the refresh button to be pretty reliable. It does seem to actually force Software to go refresh the metadata and download available updates, now. I think in the past it just 'forced' a run of the refresh timer, which doesn't always actually go and check for updates, it has a bunch of conditionals that decide whether it will. But that got changed at some point. -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, 2016-10-31 at 15:35 -0600, Chris Murphy wrote: > The other thing I'm seeing is, I'll get a notification for software > updates, click on it, see the Restart & Install blue button in GNOME > Software, close/quite GNOME Software, and at some later time relaunch > GNOME Software and it says everything is up to date even though the > offline update wasn't run. And if I refresh, it appears to be > downloading a lot of data all over again - I just don't know what and > have no good way to troubleshoot this, but the refresh is taking a > long time, maybe 30 minutes. I think any time you run a dnf command, the prepared update gets invalidated. (That may be an oversimplification.) I don't think it should be downloading unchanged packages again, though. I think it's a GNOME Software bug that it says up to date when it's not. I've complained about that for years. It will happily tell you that your freshly-installed Fedora system is up to date months after release, before the first update, often even after you manually check for updates by pressing the Refresh button. :) Michael ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
31.10.2016 22:35 Chris Murphywrote: > [...] And if I refresh, it appears to be > downloading a lot of data all over again - I just don't know what and > have no good way to troubleshoot this, but the refresh is taking a > long time, maybe 30 minutes. That's definitely not the answer for the end users but 'pkmon' command may at least tell you what it's downloading and what's the progress. Regards, Rafal ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, Oct 31, 2016 at 3:35 PM, Chris Murphywrote: > On Mon, Oct 31, 2016 at 5:02 AM, Michael Catanzaro > wrote: >> On Sun, 2016-10-30 at 22:50 -0600, Chris Murphy wrote: >>> Since offline updates are the default, and packagekit downloads >>> everything currently needing updating, if the user doesn't ever do a >>> Restart & Install to proceed with offline updates, i.e. they only >>> ever >>> use dnf for updates and never disable packagekit updates, I'm pretty >>> sure the expected result is it just accumulates all of these >>> unapplied >>> updates. >> >> A 7 GB cache cannot be the expected result. PackageKit can only have >> one update prepared at a time, so it should only cache one prepared >> update at a time and should clear the cache before preparing the next >> one. Otherwise it's just a cache leak. > > Then there must be a leak as I have multiple rpms with slightly > different versions. > > https://paste.fedoraproject.org/466976/ > > The other thing I'm seeing is, I'll get a notification for software > updates, click on it, see the Restart & Install blue button in GNOME > Software, close/quite GNOME Software, and at some later time relaunch > GNOME Software and it says everything is up to date even though the > offline update wasn't run. And if I refresh, it appears to be > downloading a lot of data all over again - I just don't know what and > have no good way to troubleshoot this, but the refresh is taking a > long time, maybe 30 minutes. After refreshing, and choosing Restart & Install, I get an offline update. Following reboot from that, all the RPMs that were in /var/cache/PackageKit/25/metadata/updates-testing/packages before the update are still in that path. -- Chris Murphy ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, Oct 31, 2016 at 5:02 AM, Michael Catanzarowrote: > On Sun, 2016-10-30 at 22:50 -0600, Chris Murphy wrote: >> Since offline updates are the default, and packagekit downloads >> everything currently needing updating, if the user doesn't ever do a >> Restart & Install to proceed with offline updates, i.e. they only >> ever >> use dnf for updates and never disable packagekit updates, I'm pretty >> sure the expected result is it just accumulates all of these >> unapplied >> updates. > > A 7 GB cache cannot be the expected result. PackageKit can only have > one update prepared at a time, so it should only cache one prepared > update at a time and should clear the cache before preparing the next > one. Otherwise it's just a cache leak. Then there must be a leak as I have multiple rpms with slightly different versions. https://paste.fedoraproject.org/466976/ The other thing I'm seeing is, I'll get a notification for software updates, click on it, see the Restart & Install blue button in GNOME Software, close/quite GNOME Software, and at some later time relaunch GNOME Software and it says everything is up to date even though the offline update wasn't run. And if I refresh, it appears to be downloading a lot of data all over again - I just don't know what and have no good way to troubleshoot this, but the refresh is taking a long time, maybe 30 minutes. -- Chris Murphy ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, 2016-10-31 at 11:54 -0500, Michael Catanzaro wrote: > On Mon, 2016-10-31 at 18:15 +0200, Alexander Ploumistos wrote: > > This is more of an issue for Workstation and the spins destined for > > desktop use, so any possible solution needs to rely on > > Anaconda/Initial Setup and each DE. > > My $0.02 as a sometimes contributor to gnome-initial-setup: it already > has too many options that can be changed post-installation. Users > report confusion about how to answer the existing, very simple options > (enable geolocation, enable automatic crash reporting). We don't want > to add any more. Especially as this option needs to be network- > specific. I do think we need a graphical option somewhere, but the > right place is the Network system settings panel. This is something that needs to be changed after install anyway, it should be a setting that is available via network manager or similar as it depends on the actual interface use. I also experienced my smartphone tethered connection being clogged with PkgKit trying to do updates, luckily for me T-Mobile has unlimited data even on roaming ... or the bill could have been astronomical. Simo. -- Simo Sorce * Red Hat, Inc * New York ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, 2016-10-31 at 18:15 +0200, Alexander Ploumistos wrote: > This is more of an issue for Workstation and the spins destined for > desktop use, so any possible solution needs to rely on > Anaconda/Initial Setup and each DE. My $0.02 as a sometimes contributor to gnome-initial-setup: it already has too many options that can be changed post-installation. Users report confusion about how to answer the existing, very simple options (enable geolocation, enable automatic crash reporting). We don't want to add any more. Especially as this option needs to be network- specific. I do think we need a graphical option somewhere, but the right place is the Network system settings panel. Michael ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, 2016-10-31 at 12:25 -0400, Bastien Nocera wrote: > Otherwise your description doesn't seem to square with the fact that > > most everyone sees background update downloads OOTB. > > > > Even still, in fact, kparal mentioned them happening on hotel wifi, so > > why would that be the case if the default is 'metered'? > > Looks like I switched metered and unmetered around in my description ;) > > They're unmetered unless you either tag them as metered, or something > (like a DHCP server) tells NM they're metered. > > Sorry about that. Ah, indeed, that lines up better with the observed data =) Still, it's interesting that the DHCP metered indication thing doesn't seem to be working, per my recall and kparal's nmcli output? -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, Oct 31, 2016 at 09:23:06AM -0700, Adam Williamson wrote: > For that matter, I'm fairly sure I've seen background update > downloading happen when I was using an Android wifi tether > connection. I'm pretty sure I remember it blowing my data cap one > month when I was using my laptop on the bus. As far as I can see, the default is "unknown". You can set "CONNECTION_METERED=yes" in the /etc/sysconfig/network-scripts/ifcfg-[whatever] file to change it on a per-device basis. The man page for NetworkManager.conf doesn't document this as one of the properties which can have its default overridden (and says that anything not documented can't be overridden). I took that at its word; possibly worth testing anyway. :) -- Matthew MillerFedora Project Leader ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
No, there's no UI: https://bugzilla.gnome.org/show_bug.cgi?id=745747 - Original Message - > On Mon, 2016-10-31 at 15:09 +, Richard Hughes wrote: > > > > I think this kind of issue is really fixed the hard way, i.e. fixing > > bugs and adding unimplemented features rather than just adding complex > > UI workarounds. > > I think making it work as best as possible without interaction is > great, but I'd argue that fundamentally this needs interaction. At a > bare minimum the UI for users to mark connections as metered or > unmetered needs to be there. Is there actually UI for this already? I > could not find it. > > The question of what the default state should be is an interesting one, > and I bet your take on it differs substantially depending on your level > of 'internet connection privilege'. ;) > -- > Adam Williamson > Fedora QA Community Monkey > IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net > http://www.happyassassin.net > ___ > devel mailing list -- devel@lists.fedoraproject.org > To unsubscribe send an email to devel-le...@lists.fedoraproject.org > ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
- Original Message - > On Mon, 2016-10-31 at 08:18 -0400, Bastien Nocera wrote: > > > > How does a connection become "unmetered"? It can't just be on interface > > > type, > > > as I have metered connections on all interface types, so presumably you > > > use > > > some form of web service to distinguish "metered" from "unmetered" based > > > on > > > a list of known IP blocks? > > > > > > Or do you simply assume that all connections are metered until the user > > > says > > > otherwise? > > > > They're metered unless you either tag them as unmetered, or hints are > > provided > > to NetworkManager by what you're connected to. For example, Android > > tethering > > is automatically tagged as metered as Android provides a hint in its DHCP > > configuration. > > Does NetworkManager 'hint' that wired connections are unmetered? It doesn't say "definitely metered" or "definitely unmetered". We usually take that to mean it's unmetered. > Otherwise your description doesn't seem to square with the fact that > most everyone sees background update downloads OOTB. > > Even still, in fact, kparal mentioned them happening on hotel wifi, so > why would that be the case if the default is 'metered'? Looks like I switched metered and unmetered around in my description ;) They're unmetered unless you either tag them as metered, or something (like a DHCP server) tells NM they're metered. Sorry about that. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, 2016-10-31 at 08:18 -0400, Bastien Nocera wrote: > > How does a connection become "unmetered"? It can't just be on interface > > type, > > as I have metered connections on all interface types, so presumably you use > > some form of web service to distinguish "metered" from "unmetered" based on > > a list of known IP blocks? > > > > Or do you simply assume that all connections are metered until the user says > > otherwise? > > They're metered unless you either tag them as unmetered, or hints are provided > to NetworkManager by what you're connected to. For example, Android tethering > is automatically tagged as metered as Android provides a hint in its DHCP > configuration. Does NetworkManager 'hint' that wired connections are unmetered? Otherwise your description doesn't seem to square with the fact that most everyone sees background update downloads OOTB. Even still, in fact, kparal mentioned them happening on hotel wifi, so why would that be the case if the default is 'metered'? -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, Oct 31, 2016 at 09:06:50AM -0700, Adam Williamson wrote: > One of the bugs has a note I was not aware of before: there's > actually a DHCP value that can be sent by the server to denote a > metered connection. If that's actually widely respected, and set by > phone wifi AP applications and the like, it could be pretty accurate. > Not perfect, but better than heuristics. I've no idea how widely it's > used in the real world, though. For what it's worth, with wifi hotspot on my Nexus 6P running latest Android N: $ nmcli -f connection.metered connection show myhotspotssid connection.metered: unknown -- Matthew MillerFedora Project Leader ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, 2016-10-31 at 09:13 -0700, Adam Williamson wrote: > On Mon, 2016-10-31 at 08:18 -0400, Bastien Nocera wrote: > > > > How does a connection become "unmetered"? It can't just be on interface > > > type, > > > as I have metered connections on all interface types, so presumably you > > > use > > > some form of web service to distinguish "metered" from "unmetered" based > > > on > > > a list of known IP blocks? > > > > > > Or do you simply assume that all connections are metered until the user > > > says > > > otherwise? > > > > They're metered unless you either tag them as unmetered, or hints are > > provided > > to NetworkManager by what you're connected to. For example, Android > > tethering > > is automatically tagged as metered as Android provides a hint in its DHCP > > configuration. > > Does NetworkManager 'hint' that wired connections are unmetered? > Otherwise your description doesn't seem to square with the fact that > most everyone sees background update downloads OOTB. > > Even still, in fact, kparal mentioned them happening on hotel wifi, so > why would that be the case if the default is 'metered'? For that matter, I'm fairly sure I've seen background update downloading happen when I was using an Android wifi tether connection. I'm pretty sure I remember it blowing my data cap one month when I was using my laptop on the bus. -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, 2016-10-31 at 15:09 +, Richard Hughes wrote: > > I think this kind of issue is really fixed the hard way, i.e. fixing > bugs and adding unimplemented features rather than just adding complex > UI workarounds. I think making it work as best as possible without interaction is great, but I'd argue that fundamentally this needs interaction. At a bare minimum the UI for users to mark connections as metered or unmetered needs to be there. Is there actually UI for this already? I could not find it. The question of what the default state should be is an interesting one, and I bet your take on it differs substantially depending on your level of 'internet connection privilege'. ;) -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
Hello, I'm not sure I have all my facts straight -and by all means, please help straighten them out-, but I would like to address Adam's original proposals. This is more of an issue for Workstation and the spins destined for desktop use, so any possible solution needs to rely on Anaconda/Initial Setup and each DE. In essence, we need a script that runs with elevated privileges and disables/enables the makecache timer and Packagit. This could then be wrapped in zenity or whatever similar works in every DE. I think this would work regardless of NetworkManager's presence, metered connection detection etc.. Would it be unfeasible to craft an Initial Setup add-on that runs after the EULA screen, displays a notification explaining background data usage and allows the user to turn it off (temporarily)? Since having updates turned off indefinitely is a Bad Thing™, this add-on could spawn a systemd service that checks if the makecache timer and PackageKit have been disabled and if so, throw a nag screen (which could be disabled, but not too easily ;) ) every n minutes/hours/days. After the first run, every DE could have each own (control) panel item/button/widget to enable toggling automatic updates on and off. All of the above would require a) the Initial Setup add-on, b) a systemd service and/or timer and c) a DE-specific package to alter the state of things, which would have to be among the default packages in each DE group. Having looked at the code of yumex, it doesn't look that difficult to adapt the toggle to work with any underlying package management system, at least the mainstream ones. Perhaps each DE could enhance it even further, e.g. GNOME could integrate gnome-online-accounts, so users could selectively disable specific sync activities or any background data-consuming jobs altogether. And now that I re-read what I have written, is this work that can be done in Fedora or would we have to work with upstream from the get-go? ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, 2016-10-31 at 11:55 +, Tom Hughes wrote: > On 31/10/16 11:41, Richard Hughes wrote: > > On 30 October 2016 at 01:26, Adam Williamson> > wrote: > > > 1) Both dnf and GNOME Software / PackageKit default to performing > > > fairly data-hungry transactions in the background, out of the box, > > > without telling you about it. GNOME's is particularly bad, as it will > > > happily download available updates in the background, which can be > > > gigabytes worth of data. > > > > If you're on an "unmetered" connection type... > > The problem, as I believe has been repeatedly explained, is that you > don't really have any idea whether the connection is metered. Yes you > may be trying to guess by considering things like 3G connections as > metered but that is an extremely poor proxy for whether a connection is > metered. > > It's not something that can be tied to connection types anyway. A wifi > interface might be unmetered if I'm at home but metered if I'm in a cafe > or hotel. An ethernet interface might be metered at certain times of day > and not at others. > > The idea that you can programatically determine, using a simple > heuristic, if a connection will be metered is just nonsense. One of the bugs has a note I was not aware of before: there's actually a DHCP value that can be sent by the server to denote a metered connection. If that's actually widely respected, and set by phone wifi AP applications and the like, it could be pretty accurate. Not perfect, but better than heuristics. I've no idea how widely it's used in the real world, though. -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 31 October 2016 at 14:56, Bastien Nocerawrote: > I'd really like the kernel to do QoS on the user's own connections. We can > know > whether downloads are interactive or not, so there is metadata available > to make this better, and not cripple interactive downloads while background > downloads are ongoing. We actually do this with PackageKit. It used to work when we were using yum, but I don't think the libdnf code does any kind of throttling. Certainly PackageKit knows which transactions are background and non-interactive, so it certainly makes sense to dribble those down slowly. I think this kind of issue is really fixed the hard way, i.e. fixing bugs and adding unimplemented features rather than just adding complex UI workarounds. Richard. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
- Original Message - > > On 30 October 2016 at 01:26, Adam Williamson> > wrote: > > > 1) Both dnf and GNOME Software / PackageKit default to performing > > > fairly data-hungry transactions in the background, out of the box, > > > without telling you about it. GNOME's is particularly bad, as it will > > > happily download available updates in the background, which can be > > > gigabytes worth of data. > > > > If you're on an "unmetered" connection type... > > This can be problematic even on an unmetered connection. An anecdotal > experience: A few months back I was on a hotel wifi, I vitally needed some > information quick, and the wifi simply didn't work - all web pages timed > out. I was very disgruntled about a crappy hotel wifi (that used to work the > day before), when in 5-10 minutes, I saw "Your updates were downloaded and > are ready to install" popup. Then I realized... tried the web browser and > web pages loaded normally. The wifi connection was so slow that while > PackageKit was downloading updates in the background, I couldn't access the > web at all. > > My poor experience stemmed from: > a) not being informed that updates were being downloaded in the background - > so I assumed the problem was elsewhere > b) not being able to pause/abort background downloads - even if I had > realized/figured out PackageKit was hogging the network, there'd have been > no way to stop the downloads (certainly no user accessible one, and even > when I tried to kill the process some time in the past, it just kept > respawning) > > You can disregard this as a "slow hotel wifi problem only", but I live in a > block of flats, the air is jammed with 20-30 wifi networks all around me, > and I experience a similar situation (though not that severe) from time to > time even at my home, a few meters from the AP - one full speed download can > completely kill any other (my own) network traffic. Again, this would not be > a problem if I a) knew about it b) could stop it. I'd really like the kernel to do QoS on the user's own connections. We can know whether downloads are interactive or not, so there is metadata available to make this better, and not cripple interactive downloads while background downloads are ongoing. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, 2016-10-31 at 10:32 -0400, Kamil Paral wrote: > > > > On 30 October 2016 at 01:26, Adam Williamson> t.org> > > wrote: > > > > > > 1) Both dnf and GNOME Software / PackageKit default to performing > > > fairly data-hungry transactions in the background, out of the > > > box, > > > without telling you about it. GNOME's is particularly bad, as it > > > will > > > happily download available updates in the background, which can > > > be > > > gigabytes worth of data. > > > > If you're on an "unmetered" connection type... > > This can be problematic even on an unmetered connection. An anecdotal > experience: A few months back I was on a hotel wifi, I vitally needed > some information quick, and the wifi simply didn't work - all web > pages timed out. I was very disgruntled about a crappy hotel wifi > (that used to work the day before), when in 5-10 minutes, I saw "Your > updates were downloaded and are ready to install" popup. Then I > realized... tried the web browser and web pages loaded normally. The > wifi connection was so slow that while PackageKit was downloading > updates in the background, I couldn't access the web at all. > > My poor experience stemmed from: > a) not being informed that updates were being downloaded in the > background - so I assumed the problem was elsewhere > b) not being able to pause/abort background downloads - even if I had > realized/figured out PackageKit was hogging the network, there'd have > been no way to stop the downloads (certainly no user accessible one, > and even when I tried to kill the process some time in the past, it > just kept respawning) > > You can disregard this as a "slow hotel wifi problem only", but I > live in a block of flats, the air is jammed with 20-30 wifi networks > all around me, and I experience a similar situation (though not that > severe) from time to time even at my home, a few meters from the AP - > one full speed download can completely kill any other (my own) > network traffic. Again, this would not be a problem if I a) knew > about it b) could stop it. Some sort of notification stating that it is happening with a [pause/cancel/never do this] button so you could control this in the moment. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
> On 30 October 2016 at 01:26, Adam Williamson> wrote: > > 1) Both dnf and GNOME Software / PackageKit default to performing > > fairly data-hungry transactions in the background, out of the box, > > without telling you about it. GNOME's is particularly bad, as it will > > happily download available updates in the background, which can be > > gigabytes worth of data. > > If you're on an "unmetered" connection type... This can be problematic even on an unmetered connection. An anecdotal experience: A few months back I was on a hotel wifi, I vitally needed some information quick, and the wifi simply didn't work - all web pages timed out. I was very disgruntled about a crappy hotel wifi (that used to work the day before), when in 5-10 minutes, I saw "Your updates were downloaded and are ready to install" popup. Then I realized... tried the web browser and web pages loaded normally. The wifi connection was so slow that while PackageKit was downloading updates in the background, I couldn't access the web at all. My poor experience stemmed from: a) not being informed that updates were being downloaded in the background - so I assumed the problem was elsewhere b) not being able to pause/abort background downloads - even if I had realized/figured out PackageKit was hogging the network, there'd have been no way to stop the downloads (certainly no user accessible one, and even when I tried to kill the process some time in the past, it just kept respawning) You can disregard this as a "slow hotel wifi problem only", but I live in a block of flats, the air is jammed with 20-30 wifi networks all around me, and I experience a similar situation (though not that severe) from time to time even at my home, a few meters from the AP - one full speed download can completely kill any other (my own) network traffic. Again, this would not be a problem if I a) knew about it b) could stop it. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, Oct 31, 2016 at 01:38:17PM +, Richard Hughes wrote: > I guess use a dconf override to change the default for all users, e.g. > https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/7/html/Desktop_Migration_and_Administration_Guide/custom-default-values-system-settings.html So in this case, a file named `/etc/dconf/db/local.d/download-updates-false` containing: [org/gnome/software] download-updates=false and then `dconf update`? Hmmm, wait -- need to create a profile file first, I guess — the default is just "user's personal database is consulted and there are no system settings"... -- Matthew MillerFedora Project Leader ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 31 October 2016 at 13:33, Matthew Millerwrote: > This is per-user, right? Right. > How do you do it for the whole system? I guess use a dconf override to change the default for all users, e.g. https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/7/html/Desktop_Migration_and_Administration_Guide/custom-default-values-system-settings.html Richard. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 31/10/16 13:29, Richard Hughes wrote: On 31 October 2016 at 12:55, Bastien Nocerawrote: So how do I tag a connection as unmetered with systemd-networkd? You can't. You get a box of bits if you start replacing parts of the Workstation experience Right, in this case you'd just have to do "gsettings set org.gnome.software download-updates false" -- we can't possibly cope with every kind of configuration possible when you start removing key bits of infrastructure that GNOME expects to be there. Well I long ago removed gnome-software completely from my machines anyway so it doesn't really bother me but really the original point in respect of people using NM and Gnome Software is that trying to track whether the connection is metered at the connection level doesn't really make any sense. On top of which I wonder how many people actually know that you can mark a connection as metered in NetworkManager - it's not like it asks when you connect to a new WiFi network if it is metered or not... The DNF cache is a separate issue of course as I'm not sure that makes any attempt to check for metered connections. Tom -- Tom Hughes (t...@compton.nu) http://compton.nu/ ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Mon, Oct 31, 2016 at 01:29:45PM +, Richard Hughes wrote: > > You can't. You get a box of bits if you start replacing parts of the > > Workstation experience > Right, in this case you'd just have to do "gsettings set > org.gnome.software download-updates false" -- we can't possibly cope > with every kind of configuration possible when you start removing key > bits of infrastructure that GNOME expects to be there. This is per-user, right? How do you do it for the whole system? -- Matthew MillerFedora Project Leader ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 31 October 2016 at 12:55, Bastien Nocerawrote: >> So how do I tag a connection as unmetered with systemd-networkd? > You can't. You get a box of bits if you start replacing parts of the > Workstation experience Right, in this case you'd just have to do "gsettings set org.gnome.software download-updates false" -- we can't possibly cope with every kind of configuration possible when you start removing key bits of infrastructure that GNOME expects to be there. Richard ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
- Original Message - > On 31/10/16 12:18, Bastien Nocera wrote: > > > They're metered unless you either tag them as unmetered, or hints are > > provided > > to NetworkManager by what you're connected to. For example, Android > > tethering > > is automatically tagged as metered as Android provides a hint in its DHCP > > configuration. > > What if you're not even using NetworkManager? > > Only my laptop, which needs the easy UI of Network Manager to connect to > different networks, uses NM while all my non-mobile machines just use > systemd-networkd to manage the network. > > So how do I tag a connection as unmetered with systemd-networkd? You can't. You get a box of bits if you start replacing parts of the Workstation experience, eg. the network status icon also won't work as expected. That's par for the course. > More to the point how do I tag it as unmetered at certain times of day... Right now, you'd script it. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 31/10/16 12:18, Bastien Nocera wrote: They're metered unless you either tag them as unmetered, or hints are provided to NetworkManager by what you're connected to. For example, Android tethering is automatically tagged as metered as Android provides a hint in its DHCP configuration. What if you're not even using NetworkManager? Only my laptop, which needs the easy UI of Network Manager to connect to different networks, uses NM while all my non-mobile machines just use systemd-networkd to manage the network. So how do I tag a connection as unmetered with systemd-networkd? More to the point how do I tag it as unmetered at certain times of day... Tom -- Tom Hughes (t...@compton.nu) http://compton.nu/ ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
- Original Message - > On 31 Oct 2016, at 11:41, Richard Hugheswrote: > > > > On 30 October 2016 at 01:26, Adam Williamson > > wrote: > >> 1) Both dnf and GNOME Software / PackageKit default to performing > >> fairly data-hungry transactions in the background, out of the box, > >> without telling you about it. GNOME's is particularly bad, as it will > >> happily download available updates in the background, which can be > >> gigabytes worth of data. > > > > If you're on an "unmetered" connection type... > > > > How does a connection become "unmetered"? It can't just be on interface type, > as I have metered connections on all interface types, so presumably you use > some form of web service to distinguish "metered" from "unmetered" based on > a list of known IP blocks? > > Or do you simply assume that all connections are metered until the user says > otherwise? They're metered unless you either tag them as unmetered, or hints are provided to NetworkManager by what you're connected to. For example, Android tethering is automatically tagged as metered as Android provides a hint in its DHCP configuration. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 31 Oct 2016, at 11:41, Richard Hugheswrote: > > On 30 October 2016 at 01:26, Adam Williamson > wrote: >> 1) Both dnf and GNOME Software / PackageKit default to performing >> fairly data-hungry transactions in the background, out of the box, >> without telling you about it. GNOME's is particularly bad, as it will >> happily download available updates in the background, which can be >> gigabytes worth of data. > > If you're on an "unmetered" connection type... > How does a connection become "unmetered"? It can't just be on interface type, as I have metered connections on all interface types, so presumably you use some form of web service to distinguish "metered" from "unmetered" based on a list of known IP blocks? Or do you simply assume that all connections are metered until the user says otherwise? -- Simon Farnsworth ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 31/10/16 11:41, Richard Hughes wrote: On 30 October 2016 at 01:26, Adam Williamsonwrote: 1) Both dnf and GNOME Software / PackageKit default to performing fairly data-hungry transactions in the background, out of the box, without telling you about it. GNOME's is particularly bad, as it will happily download available updates in the background, which can be gigabytes worth of data. If you're on an "unmetered" connection type... The problem, as I believe has been repeatedly explained, is that you don't really have any idea whether the connection is metered. Yes you may be trying to guess by considering things like 3G connections as metered but that is an extremely poor proxy for whether a connection is metered. It's not something that can be tied to connection types anyway. A wifi interface might be unmetered if I'm at home but metered if I'm in a cafe or hotel. An ethernet interface might be metered at certain times of day and not at others. The idea that you can programatically determine, using a simple heuristic, if a connection will be metered is just nonsense. Tom -- Tom Hughes (t...@compton.nu) http://compton.nu/ ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 30 October 2016 at 01:26, Adam Williamsonwrote: > 1) Both dnf and GNOME Software / PackageKit default to performing > fairly data-hungry transactions in the background, out of the box, > without telling you about it. GNOME's is particularly bad, as it will > happily download available updates in the background, which can be > gigabytes worth of data. If you're on an "unmetered" connection type... > 'hey, do you want to turn on background update downloads'? This needs to go to the GNOME upstream designers as a proposal, rather than to a downstream list. We're actually working on supporting a mode like this for Endless (i.e. we download the metadata, but not the packages themselves) but I'm not sure we've talked about any UI to configure it. Richard. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Sun, 2016-10-30 at 22:50 -0600, Chris Murphy wrote: > Since offline updates are the default, and packagekit downloads > everything currently needing updating, if the user doesn't ever do a > Restart & Install to proceed with offline updates, i.e. they only > ever > use dnf for updates and never disable packagekit updates, I'm pretty > sure the expected result is it just accumulates all of these > unapplied > updates. A 7 GB cache cannot be the expected result. PackageKit can only have one update prepared at a time, so it should only cache one prepared update at a time and should clear the cache before preparing the next one. Otherwise it's just a cache leak. Michael ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
You can add this one https://bugzilla.gnome.org/show_bug.cgi?id=768632 ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Sun, Oct 30, 2016 at 10:02:43AM -0500, Michael Catanzaro wrote: > On Sun, 2016-10-30 at 09:15 +0100, Theodore Papadopoulo wrote: > > I > > just found that the packagekit cache on my machine is about 7Gb !!! > > This has got to be a bug, please report it. There are several bug reports about the bad behaviour: https://bugs.freedesktop.org/show_bug.cgi?id=80053 https://bugzilla.redhat.com/show_bug.cgi?id=1306992 https://bugzilla.redhat.com/show_bug.cgi?id=1321319 Kind regards Till ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Sun, Oct 30, 2016 at 9:02 AM, Michael Catanzarowrote: > On Sun, 2016-10-30 at 09:15 +0100, Theodore Papadopoulo wrote: >> I >> just found that the packagekit cache on my machine is about 7Gb !!! > > This has got to be a bug, please report it. Since offline updates are the default, and packagekit downloads everything currently needing updating, if the user doesn't ever do a Restart & Install to proceed with offline updates, i.e. they only ever use dnf for updates and never disable packagekit updates, I'm pretty sure the expected result is it just accumulates all of these unapplied updates. -- Chris Murphy ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Sun, Oct 30, 2016 at 10:02:43AM -0500, Michael Catanzaro wrote: > On Sun, 2016-10-30 at 09:15 +0100, Theodore Papadopoulo wrote: > > I > > just found that the packagekit cache on my machine is about 7Gb !!! > > This has got to be a bug, please report it. My packagekit cache regularly fills up / on my systems. I almost never use packagekit to perform updates or installs, so maybe that's why. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Sun, 2016-10-30 at 09:15 +0100, Theodore Papadopoulo wrote: > I > just found that the packagekit cache on my machine is about 7Gb !!! This has got to be a bug, please report it. ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 10/30/2016 02:26 AM, Adam Williamson wrote: > There's two things I think are somewhat unfortunate here: > > 1) Both dnf and GNOME Software / PackageKit default to performing > fairly data-hungry transactions in the background, out of the box, > without telling you about it. GNOME's is particularly bad, as it will > happily download available updates in the background, which can be > gigabytes worth of data. DNF only updates its metadata caches (on a > systemd timer), but even that could be behaviour that users in certain > circumstances really really do not want. > > 2) There is no particularly obvious or visible mechanism for a 'typical > user' (or, if you prefer, many of the target 'personas' for our > flavors) to configure this behaviour...and you have to figure out two > completely *different* configuration mechanisms in order to shut off > both. Add to this that these caches seem to be never cleaned, so that they grow up very large up to the point they prevent updating the system. I just found that the packagekit cache on my machine is about 7Gb !!! Probably because I do use packageit at all (dnf cache is much smaller at ~700 Mb). A common cache, would be very welcome if possible. Less time, less download volume, less used space. Theo. smime.p7s Description: S/MIME Cryptographic Signature ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 10/30/2016 02:26 AM, Adam Williamson wrote: > * We could have some kind of configuration interface appear on install > / first boot. This would require integration with anaconda and/or > initial-setup and gnome-initial-setup. > > * We could invert the defaults and have the apps ask the user if they > want to enable data-hungry background operations on first interaction: > the first time you use dnf (without -y) it could say 'hey, do you want > to turn on background cache refreshes?' Similarly, the first time you > run GNOME Software or click on an update notification, it could say > 'hey, do you want to turn on background update downloads'? > > * We could at least make both of them respect one single config setting > for 'do data hungry background operations' (this is kinda one small > part of the bigger issue that is 'dnf and PK-based stuff don't share > any configuration aside from repo definitions'). The first two points are necessary but not sufficient at least for people with varying internet connexions (laptops). Ideally, you'd want to configure it per internet interface. Specifically, this is typically the thng I do not want while being connected thought a smartphone (even more when abroad). So not only a first time configuration, but also an easy way to disable it Your third point seems crucial. Theo. smime.p7s Description: S/MIME Cryptographic Signature ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On 10/30/2016 03:26 AM, Adam Williamson wrote: Hi folks! I kinda hate kicking off discussions like this without having a solid solution to propose or being able to promise to work on one, but this really seems important. Unfortunately I can't claim I'm gonna have time to do any concrete work on it, though I'd really *like* to. But I thought it would be worthwhile to kick around still; perhaps someone else will be inspired. I just read Hedayat's review of Fedora 25 Beta: https://hedayatvk.wordpress.com/2016/10/30/fedora-25-beta and this really jumped out at me: "And, if you care about your internet usage, make sure that you disable both dnf makecache timer, and stop PackageKit from downloading updates automatically. I don’t allow a new Fedora installation to access internet before doing these, as it might just eat a considerable amount of data." There's two things I think are somewhat unfortunate here: 1) Both dnf and GNOME Software / PackageKit default to performing fairly data-hungry transactions in the background, out of the box, without telling you about it. GNOME's is particularly bad, as it will happily download available updates in the background, which can be gigabytes worth of data. DNF only updates its metadata caches (on a systemd timer), but even that could be behaviour that users in certain circumstances really really do not want. 2) There is no particularly obvious or visible mechanism for a 'typical user' (or, if you prefer, many of the target 'personas' for our flavors) to configure this behaviour...and you have to figure out two completely *different* configuration mechanisms in order to shut off both. I think this is kind of poor behaviour on our part and we should make it better. Do I have a specific concrete proposal? No. But I do have some vague ideas. * We could have some kind of configuration interface appear on install / first boot. This would require integration with anaconda and/or initial-setup and gnome-initial-setup. * We could invert the defaults and have the apps ask the user if they want to enable data-hungry background operations on first interaction: the first time you use dnf (without -y) it could say 'hey, do you want to turn on background cache refreshes?' Similarly, the first time you run GNOME Software or click on an update notification, it could say 'hey, do you want to turn on background update downloads'? * We could at least make both of them respect one single config setting for 'do data hungry background operations' (this is kinda one small part of the bigger issue that is 'dnf and PK-based stuff don't share any configuration aside from repo definitions'). Anyone have thoughts on this? Any DNF or Software devs want to say I'm totally wrong and an idiot? Anyone inspired to do something more concrete than a mailing list post? :) Fedora updates so often that attempts to pre-download anything updates related are pointless. Chances are you a) waste gobs of bandwidth downloading that changing data over and over again without ever using it b) when you actually *do* use it, it's typically already outdated and the update you're looking for isn't there, so you need to force manual refresh anyway Just turn the automatic downloads off already. It's okay to have an option to enable such a thing if it actually makes sense for somebody in some context but in Fedora context, it makes no sense whatsoever. IMNSHO. On a related note, why on earth is the main Fedora repo set to expire every two weeks? (and its -source and -debuginfo every week??) It's not supposed to change *ever* for a released distro version now is it? - Panu - ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
Re: DNF and PackageKit background data usage
On Sat, Oct 29, 2016 at 5:26 PM, Adam Williamsonwrote: > Hi folks! > > I kinda hate kicking off discussions like this without having a solid > solution to propose or being able to promise to work on one, but this > really seems important. Unfortunately I can't claim I'm gonna have time > to do any concrete work on it, though I'd really *like* to. But I > thought it would be worthwhile to kick around still; perhaps someone > else will be inspired. > > I just read Hedayat's review of Fedora 25 Beta: > https://hedayatvk.wordpress.com/2016/10/30/fedora-25-beta > and this really jumped out at me: > > "And, if you care about your internet usage, make sure that you disable > both dnf makecache timer, and stop PackageKit from downloading updates > automatically. I don’t allow a new Fedora installation to access > internet before doing these, as it might just eat a considerable amount > of data." > > There's two things I think are somewhat unfortunate here: > I'll add: 3) dnf and PackageKit still seem to *separately* download metadata. Could they be made to share their metadata cache and download only one copy? Bonus points if they could share their cache of downloaded rpm files, too. 4) Why is the metadata so big in the first place? As far as I can tell, it's a small number (3?) of xml files that don't change all that fast. As a simple but rather silly solution, imagine if the repodata XML files were actually in a git repository and the local machine maintained a clone of the git repo. Then it would download deltas and would probably update very quickly indeed. --Andy ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org
DNF and PackageKit background data usage
Hi folks! I kinda hate kicking off discussions like this without having a solid solution to propose or being able to promise to work on one, but this really seems important. Unfortunately I can't claim I'm gonna have time to do any concrete work on it, though I'd really *like* to. But I thought it would be worthwhile to kick around still; perhaps someone else will be inspired. I just read Hedayat's review of Fedora 25 Beta: https://hedayatvk.wordpress.com/2016/10/30/fedora-25-beta and this really jumped out at me: "And, if you care about your internet usage, make sure that you disable both dnf makecache timer, and stop PackageKit from downloading updates automatically. I don’t allow a new Fedora installation to access internet before doing these, as it might just eat a considerable amount of data." There's two things I think are somewhat unfortunate here: 1) Both dnf and GNOME Software / PackageKit default to performing fairly data-hungry transactions in the background, out of the box, without telling you about it. GNOME's is particularly bad, as it will happily download available updates in the background, which can be gigabytes worth of data. DNF only updates its metadata caches (on a systemd timer), but even that could be behaviour that users in certain circumstances really really do not want. 2) There is no particularly obvious or visible mechanism for a 'typical user' (or, if you prefer, many of the target 'personas' for our flavors) to configure this behaviour...and you have to figure out two completely *different* configuration mechanisms in order to shut off both. I think this is kind of poor behaviour on our part and we should make it better. Do I have a specific concrete proposal? No. But I do have some vague ideas. * We could have some kind of configuration interface appear on install / first boot. This would require integration with anaconda and/or initial-setup and gnome-initial-setup. * We could invert the defaults and have the apps ask the user if they want to enable data-hungry background operations on first interaction: the first time you use dnf (without -y) it could say 'hey, do you want to turn on background cache refreshes?' Similarly, the first time you run GNOME Software or click on an update notification, it could say 'hey, do you want to turn on background update downloads'? * We could at least make both of them respect one single config setting for 'do data hungry background operations' (this is kinda one small part of the bigger issue that is 'dnf and PK-based stuff don't share any configuration aside from repo definitions'). Anyone have thoughts on this? Any DNF or Software devs want to say I'm totally wrong and an idiot? Anyone inspired to do something more concrete than a mailing list post? :) -- Adam Williamson Fedora QA Community Monkey IRC: adamw | Twitter: AdamW_Fedora | XMPP: adamw AT happyassassin . net http://www.happyassassin.net ___ devel mailing list -- devel@lists.fedoraproject.org To unsubscribe send an email to devel-le...@lists.fedoraproject.org