IRC logs for #baserock for Thursday, 2015-03-19

*** zoli__ has quit IRC05:27
*** zoli__ has joined #baserock05:27
*** zoli___ has joined #baserock05:43
*** zoli__ has quit IRC05:43
*** zoli___ has quit IRC05:51
*** zoli__ has joined #baserock05:51
*** zoli___ has joined #baserock06:00
*** zoli__ has quit IRC06:00
*** zoli___ has quit IRC06:01
*** zoli__ has joined #baserock06:01
*** zoli__ has quit IRC06:34
*** zoli__ has joined #baserock06:34
*** zoli__ has quit IRC06:35
paulsherwoodi think so07:01
* straycat thinks that would be slower, and might make it more difficult to express situations where 2 chunks build-depend on 2 different versions of the same software07:50
*** a1exhughe5 has joined #baserock08:01
*** paulw has quit IRC08:04
*** paulw has joined #baserock08:07
*** rdale has joined #baserock08:14
ratmice__also for gcc/flex/bison and other dev tools they don't necessarily need to be there at runtime08:17
*** jonathanmaw has joined #baserock08:26
*** gfinney has joined #baserock08:27
paulsherwoodi was meaning build-depends within a stratum. could still have build-depend strata08:30
paulsherwoodnot sure why this would be slower? less logic08:30
paulsherwood(build-depend strata not present at runtime)08:31
paulsherwoodthe only difference achieved by the current (imo obscure) behaviour is that some chunks which will be present at runtime are caused to be absent at build-time for some other chunks08:33
paulsherwoodstraycat: for the 2 chunks build depend on different versions, have 2 different strata? isn't that how we've dealt with gstreamer, for example?08:36
straycatif you implicitly build-depend on everything that's in your stratum then you'll have to, for each chunk in the stratum, unpack everything that's in your stratum into the staging area, so you'll end up unpacking a ton of stuff you don't build-depend on which will slow the build time.08:58
*** mdunford has joined #baserock09:04
*** bashrc_ has joined #baserock09:04
Kinnisonpaulsherwood: massively less efficient09:09
Kinnisonpaulsherwood: because it linearises the build of a stratum09:09
Kinnisonpaulsherwood: making parallel building of systems very hard09:09
Kinnisonpaulsherwood: Morph originally built strata in-order with chunk n implicitly b-depping on n-1..009:10
Kinnisonpaulsherwood: we introduced build-depends proper as a way to remove this linearisation09:10
* pedroalvarez agrees after building a 10-node distbuild network to speed up things09:11
* Kinnison doesn't understand why people are arguing so much regarding this build-dependency thing -- To me it's a "simple" case of transitive dependencies (pro: brevity of declaration, con: cognitive complexity on parsing) vs. explicit dependencies (pro: clarity of declaration, con: increased cognitive load to parse the definition)09:13
KinnisonBoth aspects have a cognitive load09:13
Kinnisonboth aspects have plusses09:13
Kinnisoneveryone should pick a side, and then the lead architect should pick one09:14
Kinnisonand it should be codified/documented09:14
*** mdizzle has joined #baserock09:14
*** gary_perkins has joined #baserock09:18
*** CTtpollard has joined #baserock09:21
straycatfwiw, i'm fine with build-depends as it is, it's just transitive dependencies, 's not like we're talking about abstract machines for higher-order term sharing or something09:21
KinnisonCool, so that's one vote for "Keep them transitive"09:24
*** lachlanmackenzie has joined #baserock09:27
SotKI agree with straycat09:28
KinnisonRuhroh, straycat is Nick Clegg?09:30
straycatmeow what?09:30
richard_mawThe only issue with transitive dependencies is the odd case where you explicitly don't want the indirect dependencies, but we currently only need that for things which are also in bootstrap mode, so we're probably fine keeping a special case there.09:36
paulsherwoodKinnison: ok, i understand the 'slower' point, now. i'll demur on 'transitive' dependencies if that's the consensus, but as i've said many times now - i believe it makes for unnecessary obscurity and thus harder for users to understand what's going on.09:37
Kinnisonpaulsherwood: I think having it not explicitly documented obscures things (and I personally would prefer explicit deps over implicit ones) but I also understand that it'll annoy other people who find the increase in b-deps sizes to be difficult to deal with09:38
* richard_maw thinks things would be obscured worse by unreadably large dependency lists09:38
Kinnisonpaulsherwood: hence I suggested everyone interested state their position (along with justification) and then trust to the lead architect to make an informed decision09:38
paulsherwoodis there a lead architect? :-)09:40
KinnisonAs far as I understand, yes, richard_maw09:40
petefoth  /me thinks that a 'list-depends' command could be added to morph with relatively litle effort, to output the dependencies of any chunk, stratum or system, so definition files don't get verbose, but full dependency information is only ever one step away09:41
Kinnisonpetefoth: an interesting idea, and not without precedent in other systems09:41
paulsherwoodso would it make sense to bring this to a vote? etherpad style? or some other mechanism?09:42
* petefoth suspects that most fo the code to do it already exists, bioth in morph and in ybd09:42
paulsherwoodpetefoth: you may be correct :)09:42
Kinnisonpaulsherwood: it's not a vote09:42
SotKpetefoth: you are correct09:42
Kinnisonpaulsherwood: it's everyone interested expressing their opinion and then accepting that whatever the lead-architect chooses is "right" for the project09:42
* Kinnison doesn't think software projects work well as democracies09:43
petefothBenevolent dictatorship, with consultation FTW!09:43
paulsherwoodsorry, i mis-spoke. where/how do you suggest folks communicate this opinion for richard_maw's consideration? is the existing ml and irc discussion already enough? (or too much :) )09:43
Kinnisonup to richard_maw09:43
* paulsherwood was advocating benevolent dictatorship when richard_maw was still at school, incidentally :-)09:44
SotKpetefoth: except chunk dependencies are "interesting" with our current definitions format unless you know at least which stratum the chunk is in09:45
petefothSotK: implementation detail! ;)09:45
franredis not the dependencies information available when we build the build-graph? I though also that that information is available via command line, but I can be wrong09:47
richard_mawIt's currently just all speculation as to whether making transitive dependencies explicit is more or less readable. I think we've already discussed all the relevant points. I don't believe we'll get any further without someone making a tool to mechanically alter the definitions to make them declare all their indirect dependencies.09:48
richard_mawSince we'd need a script to do the migration if we later decided to go down this route anyway, we could at least save some manual effort in the translation phase if we decide to go down this route, and if we don't at least we can say that we tried it and avoid having to expend time in further discussions.09:50
*** pacon has joined #baserock09:51
bashrc_in the wild, how often are transitive dependency issues encountered?09:51
SotKfranred: Yes, but we can only create the build-graph for a system at the moment (we could probably do it for a stratum with a minimal amount of fiddling, but chunk morphologies give no indication of what they depend on). The `morph list-artifacts` command lists all the build-dependencies of a system.09:52
Kinnisonrichard_maw: sounds good09:52
Kinnisonpaulsherwood: Since you're the primary driver of "explicit deps would be better" -- could you write a tool to generate the expanded stratum definitions?09:52
richard_mawbashrc_: would this be "issues where we actually don't want the implied indirect dependencies" or "issues with removing transitive dependencies causing things to stop working because we relied on it"?09:53
richard_mawI have data on the former, but not the latter09:53
paulsherwoodKinnison: actually, i don't think i can. the reality of what depends is *hidden* by the current algorithm in morph. for example m4 requiring gawk. short of tring all build combinations to see what breaks, the answer is just if foo depends on bar, it also depends on what bar depends on, all the way down09:54
bashrc_if it's not something which crops up very often then go with the existing build-depends but with an additional tool to check all dependencies09:54
Kinnisonpaulsherwood: Right, so it needs someone to define what currently happens in english first09:55
paulsherwoodKinnison: i think i just did? :-)09:55
Kinnisonpaulsherwood: I believe richard_maw might be the best person to write a one or two sentence statement on that09:55
richard_mawpaulsherwood: that's a different issue for exploring which indirect dependencies aren't actually required, rather than changing the representation09:55
Kinnisonpaulsherwood: no, since you didn't mention bootstrap mode, which richard_maw mentioned was different09:55
*** ssam2 has joined #baserock09:56
*** ChanServ sets mode: +v ssam209:56
paulsherwoodi don't think it's different in this respect, but i may be wrong09:56
KinnisonThat you're not *certain* implies that we need a clear statement first09:57
paulsherwoodrichard_maw: since there seems to be a reasonable level of support for sticking with what currently works, and 'fixing' it would mean 'breaking' things for real users, it's probably simplest to leave things as is09:58
* gfinney is finding this conversation quite interesting09:59
paulsherwoodheh. i'm somewhat sorry i started it :)09:59
KinnisonNaah, discussions are good, even if all it results in is extra documentation then it's a good thing10:00
KinnisonYou've identified a failing.  If it's not in how it works, then it's in how it's documented.10:00
paulsherwoodthank you for those small crumbs of comfort, Kinnison :-) in any case i'll now crawl back under my rock...10:01
Kinnisonplease don't drop this until at least there's a good statement somewhere permanent (ML at min, wiki may be better) about how it works10:02
ssam2basically we need to remove the documentation of the current morphology format from morph's readme, update it, and put it on the wiki10:03
Kinnisonssam2: +110:03
paulsherwoodgfinney: you seem interested, are you knowledgeable enough to attempt to document this yet?10:03
* paulsherwood takes that as a yes 10:03
* gfinney is going to find a rock to hid under10:04
ssam2gfinney: there is already fairly detailed (but fairly out of date) in Morph's README, so you wouldn't need to start from scratch at all10:04
petefothgfinney: take your laptop with you :)10:04
ssam2(the word 'documentation' was missing from my last sentence)10:04
* richard_maw things that would be a good place to put such documentation10:06
*** edcragg has joined #baserock10:11
*** gfinney has quit IRC10:13
*** gfinney has joined #baserock10:13
*** Krin has joined #baserock10:16
ssam2richard_maw: that implies that we'd either duplicate most of the content of definitions-V0 each time we changed the version of definitions, or that the documentation of the latest version of definitions would be spread across multiple pages10:16
ssam2I think having an up to date would be better, and could describe just the obsolete features of V010:16
*** wschaller has joined #baserock10:26
*** tiagogomes_ has joined #baserock10:31
KinnisonI am an awkward sod and think that best would be to have /definitions/ which documents the current standard, along with a changelog of what has altered since V(n-1)10:33
KinnisonAnd each time we increment N, we archive the statement of what n-1 was10:33
Kinnisonas a read-only document10:33
ssam2seems reasonable. is it possible to archive a document in our current wiki as a readonly document?10:33
KinnisonTake markdown, pandoc it to pdf, publish that10:34
Kinnisongit can keep the history of the markdown10:34
Kinnison\o/ git10:34
ssam2seems a bit longwinded10:36
* Kinnison shrugs -- your choice in the end, obviously :)_10:38
* Kinnison is just suggesting an approach10:38
* Kinnison dislikes the idea that V(n-1) remains editable wiki content10:38
ssam2me too.10:38
ssam2but I'd hoped ikiwiki would let us mark a page as readonly, rather than requiring someone to manually generate a PDF10:38
ssam2which doesn't stop anyone editing the wiki page anyway10:39
ssam2maybe I'm misunderstanding10:39
petefothhowaboot twp pages - current version, and previous versions whcih gets added to10:39
KinnisonUltimately it'd still be editable from git10:39
ssam2i suppose so10:39
Kinnisonikiwiki can't stop you editing files in the git repo10:39
Kinnisonwhich is why making it into a file not commonly edited (PDF) seemed like a plan10:39
paulsherwoodcan we just put READONLY - please do not edit at the top of applicable pages, and trust folks to do the right thing?10:41
paulsherwood(for now?)10:41
ssam2works for me. we have to monitor the wiki changelog in any case10:41
* Kinnison nods10:41
* paulsherwood is far too interested in #baserock today10:41
Kinnisonpaulsherwood: stand in the corner, facing the wall, and whatever you do, do NOT think about purple elephants10:42
* Kinnison hides10:42
straycat#baserock - better than pictures of cats10:43
straycatit's a good job i don't have op10:43
straycatand that this channel is +t10:43
ratmice__I think one issue with explicitly specifying all dependencies (rather than transitively) is that when something has optional dependencies they cannot be changed in one place, so when you want to remove one, you are back figuring out if it was an explicitly specified dependency required by something or an actual dependency...10:45
ratmice__it seems like it might be reasonable to me to have 2 forms of build-dep one which implies a transitive dependency and one which does not? (IIUC the conversation anyways)10:46
Kinnisonratmice__: An interesting proposal.  I'd be worried that there's too many possible forms at that point.  "transitive dep" "dep on foo only"  "dep on foo, and its b-deps but not further" etc.10:47
ssam2ratmice__: i'm trying to think of an example where that might actually be a problem10:48
ssam2let's say I want to remove X as a dependency of D-Bus. All I need to do is remove 'x' as a build-dep of 'dbus'10:48
ssam2oh no, you're right10:48
ssam2because then suddenly I have to work out what stuff that was in dbus' build dependencies that was actually only there because X needed it10:48
ratmice__ssam2: largely things like libsoundfile, assimp, etc which aggregate various file formats into an API i think10:49
ssam2and gst-plugins-*, that would be a nightmare :)10:49
*** gary_perkins has quit IRC11:06
*** gary_perkins has joined #baserock11:06
bashrc_so how is a baserock rootfs made?11:23
richard_mawat deployment time the cluster definition file lists type: rawdisk, which causes it to run rawdisk.write, which if given a file path and a disk size, creates a disk image, formats it with btrfs and copies the files made a build time into it11:25
richard_mawwe used to make disk images as system artifacts a while back, but this proved to be too inflexible11:25
*** zoli__ has joined #baserock11:53
bashrc_is there any documentation on the cluster format? what the possible deployment types are, etc11:55
paulsherwoodbashrc_: do the existing examples not give enough info?11:59
bashrc_so under deploy: tere are various types such as build-controller. what's the syntax of that? Is there a manpage?12:01
paulsherwoodbashrc_: morph deploy --help ?12:01
ssam2wow, 'btrfs filesystem usage /' is infinitely more useful than 'btrfs filesystem df /'12:02
ssam2the former makes it look like I have .4 GB free on this system when actually there is at least 17GB free12:03
bashrc_looks like the label under deploy can be arbitrary12:04
ssam2bashrc_: I see what you mean. it's a label, which can be passed on the commandline to `deploy` if you want to deploy only that system12:04
ssam2or set options for that system12:04
ssam2if the label is foo you can run `morph deploy foo foo.HOSTNAME=bar` to deploy only 'foo' and set its hostname12:05
ssam2wait, it'd be `morph deploy cluster.morph foo foo.HOSTNAME=bar'12:05
bashrc_oh I see12:06
ssam2interesting crash while trying to deploy an upgrade:
ssam2I think this is because this system has a blank line in /etc/fstab12:16
ssam2and the code wasn't expecting one12:16
ssam2this bug has been present the whole time ssh-rsync has existed, I can't believe it's taken two years to hit it12:18
ssam2wait, it's parsing /proc/mounts not /etc/fstab. not so simple after all12:19
ssam2right. So I was trying to upgrade the wrong machine, and that function was actually parsing an SSH error message.12:23
ssam2and because all of the checks in ssh-rsync.check only look for a specific message to determine failure, they didn't catch this.12:24
*** gary_perkins has quit IRC13:12
*** gary_perkins has joined #baserock13:13
*** gallit has joined #baserock13:13
ssam2i'm going to reboot for an upgrade, will be down for a few minutes13:15
*** gallit has joined #baserock13:15
ssam2it's back now13:17
*** wschaller_ has joined #baserock13:20
*** wschaller has quit IRC13:23
SotKI sent an RFC for partial deployment a week or so ago, does anyone have an opinion on what a partial deployment command as described therein should look like?13:39
bashrc_within the valid architectures is armv7b the same as armv7bhf ?13:52
*** gfinney has quit IRC13:52
persiabashrc_: No: the ABIs differ.  The ISA supports both ABIs.13:53
ssam2krin: I'll review your mox3 patch if you send it to instead of the mailing list13:56
Krinumm, ok ssam2.14:00
*** gfinney has joined #baserock14:01
*** gfinney has joined #baserock14:01
paulsherwoodand we're off!?14:02
paulsherwoodis pedroalvarez going on camera? :_)14:02
pedroalvarezdid the openssl came out?14:03
persiaCatching up on backscroll: to me the huge win for multiple +1s not meaning +2 is that one doesn't have to set an ACL on reviewers: anyone can review, but only some folk can accept the reviews.14:04
*** pacon has quit IRC14:05
persiaFor the workflow problem with the Submit button, we ought use some gate management software that verifies that a given change has met a certain set of criteria, and if so, automatically does the submission.14:05
pedroalvarezshould then we upgrade to this commit?
pedroalvarezok, g.b.o has lorried the branch14:14
paulsherwoodpedroalvarez: ^^14:14
paulsherwoodoh, you're on! :-)14:15
pedroalvarezno audio live audio though14:16
pedroalvarezlet's start with this14:16
jjardonpedroalvarez: I would go ahead an update to 1.0.2:
pedroalvarezhm.. I've been told that is better to stick to 1.0.1 for now14:21
straycatpaulsherwood, fwiw i do feel it would be nice to be able to tag a chunk with its direct dependencies in definitions, so that even if i "inherit" a chunk from some other stratum my current stratum build-depends on, or otherwise transitively depend on something within this stratum, i still have this chunk's direct dependencies visible.14:21
straycateven a comment would do, though morph edit would rip them out14:22
*** CTtpollard has quit IRC14:22
straycatall the ideas i have in this park are stupid though14:23
* straycat makes do with comments14:23
paulsherwoodstraycat: i wouldn't say that. we're in a complex area - it's hard to tease out all the threads14:24
* paulsherwood values straycat's input, even when we disagree14:24
pedroalvarezreview for openssl upgrade sent to gerrit14:25
* paulsherwood hits submit14:26
* pedroalvarez distbuilds14:27
straycatpaulsherwood, :) the tags would need to be verified against the actual build graph somehow and then there'd be duplication between build-depends and the direct-depends and... it'd be terrible14:28
* straycat goes back to doing something useful14:28
* jjardon is convinced about the need of transitive dependencies after reading ratmice__ comment14:28
*** CTtpollard has joined #baserock14:35
tlsaNetSurf is sticking with openssl 1.0.1 for now too.  1.0.2 is too new and openssl too critical.14:37
straycatpart of the problem i have with not easily being able to specify direct dependencies is caused by this flat namespace we've imposed i guess. if i were able to override chunks i've inherited from strata i depend on then i could easily specify the direct dependencies, happen to be thinking about this because i'm shuffling strata around atm14:37
tlsamost of the vulnerabilities was for 1.0.214:38
straycatlike with nixos you can say, i want this desktop environment, but i want to override chrome with a version of chrome that doesn't have flash in it... or something like that14:39
persiastraycat: One of the things I like about Baserock is that not all the direct dependencies are specified, giving one apparently more control over the content of an image.14:39
persiaWhen I built images from distros, there was always argument and debate over whether a given runtime dependency was really a dependency or not, especially things that were only required in certain use cases.14:40
persiaoverrides might help with that, but it is an area in which much thought is beneficial to finding a solution that meets many needs.14:41
* petefoth notes that with local build you get an idea how far through the build is ("step x of y") but there are no clues to that when ditdbuilding14:44
Kinnisonpetefoth: yeah, it's one of the reasons I'd like to see local and dist building code unified at some point14:45
edcragghi, i'm trying to get internet working within a completed native build bootstrap on a second NIC, booted directly with a kernel. i've tried adding the interface to /etc/network/interfaces, and setting routing manually. it apparently gets a response with an IP from udhcpc, but there is no IP showing up in ifconfig, and get 'network is unreachable' messages. any clues?14:47
persiaAnother reason I'd like to see it combined is that I don't understand why I have to build serially locally: even with large -j values, if I have tens of cores and tens of GB of RAM, I want to parallelise as much as possible.14:48
edcraggi also added nameservers to /etc/resolv.conf14:48
persiaedcragg: I would recommend initially ignoring config files.  If you know the network environment, use ifconfig to manually bring the interfaces up with the correct values, and route to manually set the routes.14:49
persiaOnce you have that working, try adjusting /etc/resolv.conf to set up the resolver.14:49
edcraggpersia: you mean use a static IP?14:50
persiaThis will give you a reference configuration that you can use to validate the automation.14:50
persiaedcragg: For initial testing, yes, that's how I'd start.14:50
persiaAlso, if you are using systemd-networkd, I think /etc/network/interfaces isn't where you want to configure the interfaces (but I don't recall the correct location off the top of my head)14:51
edcraggit's a bootstrap tarball, no init14:51
persia/etc/resolv.conf should be independent of any interface enablement, address assignment, or routing, but some automation will clobber it in some cases, so only validate that once the rest works, and be prepared to see change as you enable automation.14:51
persiaIf there is no init, then /etc/network/interfaces is definitely useless.14:52
edcraggit seems to work14:52
persiaThen you are running some sort of init.14:52
persiaOr at least, some automation for bringing up interfaces.14:52
persiaThe kenel won't bring up any intefaces by default with init=bash14:53
edcraggi was assuming there is some magic in the cross-bootstrap system14:54
persiaThere may be such magic, but if so, it is running under init.14:55
persiaEssentially, when the kernel boots, it hands control to PID 1, which we call init.14:56
persiaPID 1 may be something simple (e.g. a shell), or interestingly complex automation (e.g. systemd).14:56
persiaAnd network configuration happens in userspace, so that is the result of whatever PID 1 happens to do.14:56
ssam2the cross-bootstrap rootfs has no magic that I know off14:58
persiaOr is it just a tarball for use with chroot()?14:59
ssam2persia: it can be used either as a chroot or with init=/bin/sh15:05
edcraggor `init=/tools/bin/sh` even15:05
ssam2heh, yeah15:06
pdarheya, has anyone used kmod with baserock to load kernel modules?15:07
persiaRight then.  WIth init=/bin/sh, there's definitely no magic happening.15:10
persiaOr if there is, something is very deeply wrong with the implementation of /bin/sh15:10
pedroalvarezpdar: is this an out-of-tree kernel module?15:12
pdarpedroalvarez: I dont think so. But I guessed what out-of-tree kernel module meant.15:16
pdarwhat does that mean?15:16
pedroalvarezi don't really know what kmod is, though15:17
edcraggpersia: yep15:17
pedroalvarezpdar: an out of tree kernel module is a kernel module that is not included in the kernel tree (repository)15:17
pedroalvarezpdar: so that it has to be built in a different chunk15:18
pdarit handles things like modprobe and lsmod. Loading and unloading kernel modules15:19
pdarpedroalvarez: as I suspected. its not out-of-tree.15:20
pedroalvarezpdar: so, if I understand correcly, you want to enable a kernel module, that is already built in your kernel but not enabled15:21
pedroalvarezpdar: you can specify in the kernel chunk that you want to enable that module always, and then you don't need to use modprobe to enable anything15:22
radiofreepdar: if you do make modules_install properly, and depmod, systemd should take care of it15:23
radiofreepdar: here's an example
pdarpedroalvarez: ha, nope, it is enabled. I just wanted to know if anyone knew why running modprobe doesnt behave as i expected it to.15:24
radiofreepdar: you probably need to run depmod before modprobe will behave as you expect it to15:25
pedroalvarez`depmod -a` ?15:26
radiofreeyou don't need the -a15:26
radiofree`depmod` is fine (a is default if nothing is specified)15:26
*** gfinney has quit IRC15:27
pdargreat! depmod worked! thanks radiofree, pedroalvarez15:28
*** gfinney has joined #baserock15:32
*** petefoth has quit IRC15:51
*** jonathanmaw has quit IRC16:07
tiagogomes_I was wondering why I am not seeing patches on baserock-dev lately, is it because everything is done through Gerrit now?16:19
Kinnisonwhere possible, yes16:19
pedroalvarezWe were discussing yesterday the creation of another mailing list, so gerrit sends all the patches and comments there16:24
persiaIn another context, such a mailing list was created, and most of the discussion about it was on how to remove oneself from subscription.16:25
pedroalvarezI sometimes find useful the fact that one can google things and find the patch in the mail list16:26
persiaIn the archives, there were no cases of mail being sent to that list by humans.  There were rare cases of humans replying to mails to that list and sending to other lists, to highlight reviews that might have set precedent in ways with which the author disagreed, but I don't think it is onerous to reference a gerrit comment in a mail by hand, rather than subscribing to a list that one doesn't use, and replying to mail from that.16:26
persiaIf we insist on good commit messages, git can handle that.16:27
*** gallit has quit IRC16:31
bashrc_trying to cross boot to armv7bhf I get:  GCC is not passing the correct host/target flags to GMP's configure script16:33
ssam2any reason why we still don't set 'CLOUD_INIT: yes' in release.morph for any of the systems?16:34
ssam2would be nice to be able to deploy the released images straight into openstack16:35
pedroalvarezbashrc_: yeah, there is a bug in stage1-gcc16:35
pedroalvarezbashrc_: I never had time to get around it16:35
bashrc_so is that a bug in gcc itself?16:36
pedroalvarezwell, let's say is a bug in our workaround for a bug in gcc16:38
pedroalvarezour workaround is here:
tiagogomes_right, I should had been explicit in the case16:40
pedroalvareztiagogomes_: workaround works for a native build, but not for cross-bootstrap builds16:41
tiagogomes_mm, wondering why. Is the build of the cross-bootstrap tarball that is failing? Or is it after when the native-build script runs16:44
Kinnisonstage 1 is during the morph cross-bootstrap phase I think16:44
bashrc_tiagogomes: building to the tarball16:45
pedroalvareztiagogomes_: when building stage1-gcc, you can't assume "host=armv7a" if you are running the build on x8616:45
bashrc_it appears that the workaround is doing something resembling that16:46
tiagogomes_pedroalvarez right16:46
Kinnisonstage 1 needs host == build and stage 2 needs host == target16:47
KinnisonI think16:47
pedroalvarezI'd say that it also needs host == build host16:47
pedroalvarezno, I don't  have brain left to think about this16:48
Kinnisonssam2: might need to weigh in16:49
Kinnisonmy brain runs in circles thinking about the bootstrap16:49
ssam2i sent a mail a few weeks ago with a table of host, build, target16:49
pedroalvarezhe did :)16:50
paulsherwoodi saw that mail, meant to say thanks!16:50
Kinnisonssam2: my -dev backlog is over 1200 mails16:50
Kinnisonssam2: some of which are unresolved threads from the start of the year16:50
paulsherwoodmy *unread* backlog is 6912 mails :/16:51
tiagogomes_how do I clone a repository from gerrit using SSH? I am trying `git clone ssh://`. Wasn't my public key automatically added to gerrit?16:51
pedroalvareztiagogomes_: you have to use your username instad of "git"16:51
ssam2tiagogomes_: look in the Gerrit web UI for the exact clone URL16:51
ssam2you also need to use port 2941816:51
bashrc_I think there's a clone command which you can copy on each project site16:53
tiagogomes_it still doesn't work: `git clone ssh://`16:53
ssam2what is the error?16:54
SotKtiagogomes_: have you added your SSH public key?16:54
tiagogomes_I think that is the problem:
ssam2there'll be 30 seconds or so of Gerrit downtime coming up as I take a backup16:56
*** a1exhughe5 has quit IRC17:01
tiagogomes_heh, for some reason I didn't had an username in the gerrit account17:07
rdaleyou have to add one, even though to register your openid you need a username17:15
*** zoli__ has quit IRC17:22
straycattake it i'm the only person who seems to have intermittent connection problems with gbo?17:27
straycat(deploying a new trove and got another connection refused when morph tried to fetch cpython)17:28
pedroalvarezdidn't you had also problems with
paulsherwoodmaybe it's suffering a DoS?17:28
straycatpedroalvarez, sorry i meant, though i also have intermittent issues pulling over ssh from gbo17:29
straycatmaybe i have some dodgy networking at my end17:29
pedroalvarezANNOUNCEMENT: There will be some g.b.o downtime due OpenSSL upgrade17:32
pedroalvarezSorry about the short notice, but I couldn't estimate how long the build was going to take17:33
pedroalvarezI actually thought it was going to take less than 2 hours and it took 317:33
paulsherwoodpedroalvarez: how long?17:34
pedroalvarezJust a reboot, less than 3 minutes17:34
paulsherwoodcool, can't wait  :)17:34
pedroalvarez(assuming everything works :)17:34
straycatwill stay up?17:35
*** wschaller_ has quit IRC17:37
paulsherwoodweirdly, with master morph and master definitions, i appear to *not* be getting cached artifacts17:37
pedroalvarezstraycat: yes17:37
pedroalvarezok, g.b.o is going down now17:37
straycatpaulsherwood, maybe one of your fetches failed and morph started building, as mine just did?17:38
paulsherwoodah, that might be it17:39
paulsherwoodnope... im an idiot. no artifact.cache.server specified :)17:39
*** wschaller has joined #baserock17:40
paulsherwoodnow we should think about how to make it faster :-)17:43
pedroalvarezthe upgrade was pretty quick17:44
pedroalvarezthe build was really really slow for this17:44
pedroalvarezputting the new openssl on top of the system can be the fastest way to get a security fix on a system without building everything17:45
* pedroalvarez upgrades cache.baserock.org17:47
*** jonathanmaw has joined #baserock17:52
*** mdizzle has quit IRC17:56
*** zoli__ has joined #baserock18:01
*** wschaller_ has joined #baserock18:02
*** bashrc_ has quit IRC18:03
*** gary_perkins has quit IRC18:04
*** wschaller has quit IRC18:04
*** gary_perkins has joined #baserock18:04
*** Krin has quit IRC18:04
*** mdunford has quit IRC18:05
*** gary_perkins has quit IRC18:06
*** gary_perkins has joined #baserock18:06
edcraggdoes anyone recognise this build error (stage 2 gawk)?
*** jonathanmaw has quit IRC18:18
*** wschaller has joined #baserock18:19
*** gary_perkins has quit IRC18:20
*** gary_perkins has joined #baserock18:20
*** wschaller_ has quit IRC18:21
*** gary_perkins has quit IRC18:22
*** gary_perkins has joined #baserock18:22
pedroalvarezedcragg: make is trying to remove things from the read-only filesystem18:27
pedroalvarezOh, given that is in stage 2, is trying to remove things from your host!18:28
pedroalvarezedcragg: you may want to take a look at how gawk stage2 is being built?18:31
gary_perkinspedroalvarez: is it normal for my morph file to disappear (despite me chmodding it to 400!) from the definitions directory?18:32
pedroalvarezgary_perkins: I'm not sure about what you mean18:37
pedroalvarezWhen is this file being removed?18:38
gary_perkinsit keeps disappearing after my morph deploy, which is currently not working due to credenials not working18:38
*** gfinney has quit IRC18:40
edcraggpedroalvarez: thanks... i was thinking the read only thing was something to do with containerisation?18:40
gary_perkins/ws/master/ # morph deploy ct-trove-openstack.morph trove.OPENSTACK_PASSWORD=***************18:40
gary_perkinsERROR: OPENSTACK_PASSWORD was not given18:40
gary_perkinsERROR: openstack.check failed with code 1: ERROR: OPENSTACK_PASSWORD was not given18:40
gary_perkinsis the "trove." supposed to be the name of the cluster morph?18:43
pedroalvarezgary_perkins: the file going away is really odd18:43
gary_perkinspedroalvarez: it is. It just disappeared now, without a morph deploy18:44
pedroalvarezIs the name of the deployment, no the name defined on the morphology file on the top18:44
gary_perkinsthat root fs is 88% full18:44
gary_perkinsit's gone again18:44
gary_perkinsbtrfs bug?18:46
*** mwilliams_ct has quit IRC18:49
*** benbrown_ has quit IRC18:49
*** perryl has quit IRC18:49
*** pdar has quit IRC18:49
*** petefotheringham has quit IRC18:49
*** wdutch has quit IRC18:49
*** Zara has quit IRC18:49
*** jmacs has quit IRC18:49
*** paulsherwood has quit IRC18:49
*** bwh has quit IRC18:49
*** kejiahu has quit IRC18:49
*** SotK has quit IRC18:50
pedroalvarezgary_perkins:... Maybe?18:50
*** rdale has quit IRC18:50
*** pedroalvarez has quit IRC18:51
gary_perkinsI tried running 'sync' after copying etc. to try and make sure what I'm seeing is what is there. Doesn't really help18:51
*** pedroalvarez has joined #baserock18:52
*** ChanServ sets mode: +v pedroalvarez18:52
gary_perkinsI did a morph gc before, because the fs was 97% full18:52
gary_perkins<gary_perkins> I tried running 'sync' after copying etc. to try and make sure what I'm seeing is what is there. Doesn't really help18:52
*** gary_perkins has quit IRC18:54
*** gary_perkins has joined #baserock18:54
*** bwh has joined #baserock18:54
*** petefotheringham has joined #baserock18:54
pedroalvarezregarding the other error, can I see your cluster morpholgy?18:55
*** DavePage has joined #baserock18:55
pedroalvarezright, "trove." has to be "ct-trove-openstack." for this morphology18:57
*** wschaller has quit IRC18:57
*** pedroalvarez has left #baserock18:57
*** pedroalvarez has joined #baserock18:57
*** ChanServ sets mode: +v pedroalvarez18:57
gary_perkinsyes, I tried that :) Also, I need to change the tenancy again18:57
*** SotK has joined #baserock18:57
gary_perkinstrying again, but with a different password, seems to be working! :)18:59
gary_perkinsthanks pedroalvarez, I think it might work now :)19:00
gary_perkinsNope :(19:05
gary_perkinsError creating Btrfs system layoutERROR: Command failed: cp -a /srv/distbuild/tmp/deployments/tmpDBi2LH/tmp52W6pR/. /srv/distbuild/tmp/deployments/tmpDBi2LH/tmp_zUjJU/tmpTuLowG/systems/factory/orig/.19:05
gary_perkinscp: write error: No space left on device19:05
pedroalvarezI guess we can blame btrfs :)19:06
pedroalvarezcan you redeploy your devel system or use a different one?19:06
gary_perkinsnow df -h and btrfs filesystem df / report only 28% full!19:07
*** zoli__ has quit IRC19:07
gary_perkinsI think it would be best if /ws on the instance was a good size ext4 fs19:08
DavePagegary_perkins: Run out of inodes (or btrfs equivalent) rather than disk space?19:08
gary_perkinsbtrfs does strange things when almost full :(19:09
*** wschaller has joined #baserock19:09
*** lachlanmackenzie has quit IRC19:10
ssam2gary_perkins: try `btrfs filesystem usage /` instead19:11
ssam2it's a bit better19:11
pedroalvarezhe said that it also reported only 28% full19:11
gary_perkinsssam2: Yes, that's much better, and it reports ~75GB allocated out of 80GB19:12
gary_perkinsI'll try the other devel instance, cheers19:13
straycatpedroalvarez, i have a machine that's got libvirt but uses bridged networking rather than nat, i think i see your concern wrt to the kvm check i added19:14
pedroalvarezwhich I don't remember :)19:15
pedroalvareziirc my main concern was that having the network active wasn't a  blocker to create a vm.19:15
pedroalvarezIs a blocker if you want to boot it afterwards19:15
straycatoh, well it is a blocker19:16
straycatbut morph by default expects the virtual network 'default' to be started, which it doesn't need to be if you're using bridged networking19:16
pedroalvarezaha! that's another point19:16
straycatyay new trove19:17
pedroalvarezI think is only a blocker if your deplyment tries to boot it (AUTOSTART: Yes)19:17
pedroalvarezbut I may be wrong19:17
*** zoli__ has joined #baserock19:18
straycatbtw, why does the example specify --upgrade for an initial deployment?19:18
straycatis that meant to be there?19:18
straycati didn't use it and my new trove seems okay19:19
*** tiagogomes_ has quit IRC19:20
ssam2must be a mistake19:20
straycatokay, i'll fix it then19:21
pedroalvarezyeah, looks like a mistake19:21
pedroalvarezI sen't today the first patch to today and it was brilliant19:21
pedroalvarezI did it on live in the screencast :)19:22
straycatoh, is not the default upstream trove?19:24
pedroalvarezwhy wouldn't it?19:26
straycatthe list of remote troves is empty19:27
straycati didn't specify anything in UPSTREAM_TROVE because i assumed gbo would be default19:27
*** wschaller has quit IRC19:30
pedroalvarezah, I understand now19:34
pedroalvarezno, now is not the default, so you can deploy troves that don't do any mirroring19:35
pedroalvarezthis way we can deploy :)19:35
straycatguess i get to upgrade my trove then19:36
pedroalvarezan upgrade might not work, since the UPSTREAM_TROVE information is in a git repo19:37
straycatgood point, okay i guess i'll fix it in that repo19:38
* straycat adds lorry-controller client to wishlist19:44
straycatoh, lorry-controller automatically reread its configuration before i figured out what POST request to send >.>19:46
straycatalso whilst i'm being annoying isn't deploy --upgrade deprecated? if so should we update  ?19:49
straycatat the same time adding whatever command you're supposed to use to  ?19:51
straycati used deploy --upgrade the other day and it worked great, but still19:51
*** gary_perkins has quit IRC20:00
straycatperhaps a task for another day20:16
* straycat disappears20:16
*** ssam2 has quit IRC20:39
*** paulsherwood has joined #baserock21:16
*** edcragg has quit IRC21:59
*** zoli__ has quit IRC22:18

Generated by 2.14.0 by Marius Gedminas - find it at!