=== cpaelzer_ is now known as cpaelzer [04:57] good morning [06:12] Good morning [07:55] Hello! [07:55] Is it possible to cluster 2 or more nodes (computers) as a SSI (Single System Image)? [07:56] zzarr: this lists multiple Linux based solutions https://en.wikipedia.org/wiki/Single_system_image#Examples but most of them seem to not have received updates in a while. [07:57] the only relevant info I find is http://www.popcornlinux.org/ [08:00] reading the wikipedia article which states "The interest in SSI clusters is based on the perception that they may be simpler to use and administer than more specialized clusters. Different SSI systems may provide a more or less complete illusion of a single system." makes me suspect that maybe there are now fewer people interested in creating this illusion than there used to be in the past. [08:01] (at least in linux lands) [08:01] have you considered other HA models? [08:02] zzarr: ^ [08:02] I just wanted to build a cluster as an experiment for own use [08:03] I was thinking about buying 2 https://www.pine64.org/?product=rockpro64-4gb-single-board-computer [08:03] and cluster them [08:04] what would this cluster then do? [08:06] while the CPU's on the boards are strong but nothing special, they have something called a NPU (neural network processor) which would be nice to have 2 of so to speak [08:06] other whys it's just to learn and have fun :) [08:07] my (surely subjective) perception is that in the end you always want to create / provide a highly available application / service. so, for example, you want a HA web or database server. to do this, you don't need classic server clustering, but just service / application level clustering, and this can actually make more sense. [08:08] nice idea [08:10] regarding the NPU i'd assume something similar to distcc to exist, i.e. a small service layer running on all nodes which then allows you to address an abstraction of these as a single entity. [08:10] I think so [08:14] i can't seem to find much on clustering NPUs, maybe the concept is not applicable to this kind of processor, or it has just not yet been done (or not in open source software) [08:14] do you know if it is possible to make 2 kernels communicate over PCIe? [08:15] well, I could write my own software once I know how to use the NPU [08:15] * tomreyn does not know [08:16] i assume you're loking for a fast link and DMA [08:16] what is DMA? [08:16] direct memory access [08:16] sure, a fast link yes [08:17] that is correct [08:17] an external process writing to memory normally controlled by the kernel [08:18] you usually dont want DMA because it means abandoning security, but for low latency processes and clustering it can the the only option [08:18] I see [08:20] there are several standard interfaces which can do DMA. i think pci and thunderbolt at least. bluetooth provides a mode which allows for it. [08:20] maybe that's also a ##hardware and ##linux question [08:21] I was thinking of buying some FPGA as a switch between the two boards (and possibly expandable), but the frequency if PCIe is 8GHz [08:21] so it might not be possible [08:22] I'll ask in those channels [08:23] but it might be possible to just hook PCIe -> PCIe (Tx -> Rx) [08:23] rbasak: you should be the first to be around, would you have some time for https://code.launchpad.net/~paelzer/ubuntu/+source/qemu/+git/qemu/+merge/354695 as the automated tests will conclude in a few hours and I'd want to be ready then [08:25] zzarr: https://serverfault.com/questions/336649/pci-e-connection-between-two-servers - infiniband actually can do RDMA (remote direct memory access), too, IIRC [08:40] cool, I'll have a look [08:44] very interesting, thanks tomreyn [08:45] you're welcome, but i'm really not knowledgable in this area, it'd be better to talk to someone who is. [09:46] rbasak: thanks for the check I have the edit for the SRU tempalte open already :-) [09:47] askring rharper is a good idea I'll do so privately [09:47] and on the dep3, I'll extend my template to differ between the two [11:18] Hi! I'm experiencing some network issues with my server, seems to appear when i clone the mac address. (It's only used by the server). [11:19] The problem is that connections drops once in a while ahen the mac address is cloned, my network contoller is 82566DM-2. [11:19] Running the e1000e driver. [11:20] I'm cloning the mac address in systemd-networkd, should this be done in udev? How to do that in 18.04? [11:36] smoser: could you add SRU information to bug 1781039 please? [11:36] bug 1781039 in cloud-init (Ubuntu Trusty) "GCE cloudinit and ubuntu keys from metadata to ubuntu authorized_keys" [Medium,Confirmed] https://launchpad.net/bugs/1781039 [11:36] smoser: in particular Regression Potential. Looks like the entire cloudinit/sources/DataSourceGCE.py file has been rewritten or wholesale backported? [11:56] halvors1: "cloning"? you mean spoofing? cloning would imply multiple NICs with teh same MAC, which would explain issues. [12:40] Hello. Attempting to install Ubuntu Server Lts 18.04.1 in a multipath FibreChannel environment using LPFC (Emulex) adapters. I am unable to enable any sort of multipath support, as the documentation seems outdated (for 16.04, which has a different kernel command line on the install media). [12:41] Trying an LVM2 installation without any multipath, thinking it could be enabled later, results in duplicate LVM2 PV ID reports and a failed boot. Any attempt to install to the same disk(s) also fails, as it cannot umount/disengage from the flawed LVM2 configuration that results. [12:41] Thoughts? [12:43] I mean besides using a rescue CD to dd /dev/zero to the disk to try again, which is what I'm doing. [12:51] Hi, i've setup a samba share, everytimes i create a new directory the default permission are drwxr-sr-x , how can i force to disable the 's' permission in samba conf? [12:51] blackflow: Yeah mean spoofing, the reason is that only a specifid mac is allowed by the upstream switch. [12:53] stukdev: see "inherit permissions", "force directory mode" and "directory mask". [12:54] rbasak: i just set the force directory mode 0775 but don't work [12:54] stukdev: do you have "inherit permissions" set? [12:55] rbasak: i'll ask shaner to do that. sorry for not having asked that before uploading. this was me just sponsoring someone elses review. [12:55] err... work. [12:55] rbasak: this is what i'm seraching, becouse in webmin i don't find it.. [12:55] Chainsaw: are you trying to put root fs and boot from mpth? [12:55] mpath* [13:02] rbasak: ok, i setup inherit permission = yes, force directory mode = 0775, directory mask = 0755, but new fdirecoty always create 's' permission [13:03] stukdev: have you actually read the definition of "inherit permission"? I didn't suggest that you set it to "yes". [13:05] rbasak: yes, but using no, its the same, so i try using yes [13:05] I'm not sure then, sorry. [13:30] jelly: Correct. [13:30] jelly: These are boot-from-SAN systems with no local storage of any kind. (HP ProLiant BL460c Gen 8 on a 3PAR 8440) [13:54] how do I google for a reason why a log.jcifs* get created in /var/log/samba ? [13:59] DammitJim: samba creates logs in there with the client machine suffix [14:00] thanks sdeziel [14:00] weird! but these files are empty! [14:01] DammitJim: yeah, some of mine are empty as well. [14:01] searching for: samba log.jcifs* access denied doesn't turn up anything on google [14:01] that's why I was trying to get some ideas on what to search for [14:02] DammitJim: jcifs is probably the name of the client machine so I'd remove this part from the google search as it's really specific to your environment [14:03] it says: log.jcifs255_105_ac [14:03] I'm verifying 255.105 are the last 2 octects of the IP of the client machine [14:05] let's say I removed a file that belongs to a package, how can I get it back? apt install --reinstall doesn't work, it doesn't restore the file [14:25] how did you remove the file? [14:25] it wasn't me, someone made a mistake, I don't know the details [14:25] bug report [14:25] it's easier to just restore it instead of investigating how he managed to remove the file [14:26] what's the name of the file? [14:26] https://bugs.launchpad.net/ubuntu/+source/postfix/+bug/1791403 [14:26] Launchpad bug 1791403 in postfix (Ubuntu) "postfix 3.3.0 unable to create missing queue directories" [Undecided,Incomplete] [14:26] /etc/postfix/postfix-files [14:26] I told him to extract the file from the package manually [14:27] maybe because it's in /etc --reinstall didn't work as expected [14:27] i.e., a config file [14:34] kstenerud: to which ppa did you say you pushed a new postfix upload? [14:35] the mp description now only lists https://launchpad.net/~kstenerud/+archive/ubuntu/cosmic-postfix-mysql-config-1791139 [14:35] and that one isn't using the ~ppaN suffix [14:36] I *think* you can delete the packages in there and upload new ones using -1ubuntu2~ppa1, but the ppa might reject it as being lower, even after the deletion [14:36] I've had conflicting experiences with that. My mantra is "PPAs remember" [14:38] OK, this is totally off topoic, but to lighten things up a bit. Today [14:38] OK folks, today is a HISTORIC day, today in history: https://www.youtube.com/watch?v=hcJAWKdawuM [14:42] ahasenack: OK I'll push a ppa version [14:47] ahasenack: How do I delete the bad upload to the PPA and replace it with a new one? [14:48] kstenerud: view package details -> then you should see a delete option in the top right next to a copy packages one [14:49] hmm it still won't let me upload [14:49] Package has already been uploaded to ppa on ppa.launchpad.net [14:49] Nothing more to do for ../postfix_3.3.0-1ubuntu2~ppa1_source.changes [14:50] that's from dput [14:50] not a server response, right? [14:50] everytime dput uploads a package, it creates a *.upload(ed?) file, and that is what it looks for [14:50] you can dput -f [14:51] or remove that upload file [14:51] ah ok [14:51] you can tell by how quickly dput complained that it wasn't a server response :) [14:51] afaik beside this particular condition, dput always succeeds [14:52] it's async. If it actually worked or not, an email will tell you [14:53] jelly: You've successfully understood the question. Are you able to help please? [15:02] My ubuntu instance on EC2 suddenly became unresponsive. I tried to restart it from the EC2 console. For 10 minutes, nothing happened. I then tried to stop it and EC2 offered to "force-kill" it. That took another 10 or so minutes. What logs can I check or other tests can I perform to see what caused it to go down and then to be so hard to restart? [15:02] I would try to check it's console [15:03] also, if its root is an ebs volume, maybe mount it from another working instance to inspect it [16:09] I'm curious - in RHEL/CentOS, "systemctl restart cryptsetup.target" will reparse /etc/crypttab and make additional volumes available. This appears not to be the case in Ubuntu 18.04. Is there a preferred way to reparse crypttab in Ubuntu, without a reboot or manual set-up? [16:19] Ah, EL didn't want systemctl daemon-reload first, but Ubuntu does, after which restarting cryptsetup.target works. [16:34] mason: you don't need daemon-reload for changes in crypptab as that's not systemd file. I guess you're seeing the generator being retriggered with daemon reload [16:34] (as cryptsetup is handled by a generator) [16:35] uhh... crypttab. [16:35] blackflow: That sounds right, yeah. [16:35] The odd difference is that evidently EL runs the generator without daemon-reload [16:39] afaik RH does some changes to their systemd that's not upstreamed, like that 0day issue, RH specifically allows usernames starting with a number. that's something I remembered. yeah, that's my pet peeve with it.... so much for one, standardized interface across all distros. [16:42] My pet peeve is largely "here's an ever-growing project that stifles opportunities for people to invent novel standards-based components to drop into a modular system". [16:45] stifles how? [16:52] is /var/log/daemon.log gone from ubuntu? [16:56] kstenerud: iirc that was defined through rsyslog, so yes. you can always define it yourself. [17:10] blackflow: rsyslog is still there by default [17:11] blackflow: My 'Net cut out due to maintenance, but in short, systemd disincentivizes new development in the spaces it occupies, since it's got fairly complex interdependencies on its components. [17:12] The previous model was far more modular, with standards-based modules being drop-in-compatible with each other. [17:13] the previous model was a shell script :) [17:15] No, think of rsyslog, MTAs, MDAs, all sorts of server software, DNS, etc. [17:15] In short, everything we do that defines the field. === InfoTest1 is now known as InfoTest [17:39] ahasenack: I think logwatch needs a bit of an overhaul. It's referring to a bunch of logfiles that haven't been present since precise [17:39] kstenerud: upstream or in our package? [17:40] note that if it's upstream, that's because it is meant to be installable everywhere, even on ancient systems probably. General OSS cruft :) [17:40] I'd have to check upstream. Not sure of debian has things in dist.conf or not [17:40] The basic idea is to leave default.conf alone, and for distributions to add config files to dist.conf, and users to modify /etc/logwatch.d [17:41] but ubuntu dist.conf is pointing to old things that are no longer true [17:41] kstenerud: ah i see [17:41] kstenerud: yeah, it probably needs love :) [17:41] which is the root cause of most of the logwatch bugs in lp [17:42] In particular, we have no daemon.log or messages in /var/log [17:42] kstenerud: to old things? i thought most of the bugs we were getting were missing new things :) [17:42] Some are from getting new unexpected log entries, and others are from things that stopped working in 14.04 [17:43] I think it coincided with the move to rsyslogd [17:43] ah could be [17:44] rsyslogd has a very restricted set of logfiles (auth.log, kern.log, mail.log, and syslog) [17:44] * in ubuntu config [17:45] ah that makes sense [17:45] yeah, by default [17:45] logwatch, on the other hand, is still looking for daemon.log and messages, which aren't configured in rsyslogd [17:46] right [17:48] So I think I can fix this in dist.conf. The only thing is that once I point things to where they're supposed to go, logwatch is going to choke on a lot of the new messages it hasn't been updated to handle (I did some preliminary tests and got quite a few unexpected message warnings) [17:49] kstenerud: yeah, it probably needs two sets of updates [17:49] But at least it will be processing what it's supposed to be processing [17:49] kstenerud: that ends up being the hard decision -- how much should we be fixing, vs. upstream should be fixing [17:49] Well, unless dist.conf is being driven by debian, it's our problem. The unexpected messages will be something that should be fixed upstream ideally [17:51] kstenerud: right, that's what i meant -- you're right to fix dist.conf for our distro configuration; but the unexpected messages should be fixed upstream, if they are new formats, etc. [17:51] kstenerud: iirc, logwatch upstream was pretty responsive [17:59] mason: I think you're making the common mistake of assuming that systemd the project defines what a distro should or shouldn't include. one can still have _just_ the pid1 and process manager, and use all the other components from before. [18:00] mason: including crypttab generator which is strictly speaking not needed, but is a good thing to have so it benefits from dependency triggers if systemd knows about it [18:01] * mason sips the Kool-Aid. [18:02] mason: patches welcome ;) [18:03] blackflow: I can think of places I'd rather spend time. [18:04] Patches would require excessive time and attention in this case. [18:28] sarnold: lovepopsickle in #ubuntu points out no torbrowser-launcher AA profile exists, though it is listed here: https://wiki.ubuntu.com/SecurityTeam/KnowledgeBase/AppArmorProfiles [18:28] sarnold: could you reply/follow-up? (or i can redirect them to a better channel) [19:09] anyone know how to address this issue of getting php-dev installed on ubuntu 16.04 server? https://gist.github.com/arooni/f535d32c8fb8c0a74c73e6c4425c44da [19:19] arooni: you are using ondrej's ppa [19:19] arooni: if you want to use it, use it for everythign and seek support there, or just us ubuntu's packages [19:19] i think i must have been ; https://askubuntu.com/questions/813788/cant-install-libpcre3-dev i followed that and i think hopefully downgraded to the official repos [19:20] arooni: if you don't intend to use ondrej's ppa, you should ppa-purge it [19:20] nacc: thats a good call; so is there any advantage to using it at all? [19:25] arooni: ondrej's ppa lets you use non-ubuntu version of php, e.g. 7.1/7.2 on 16.04 [19:31] and is that recommended? or not so much [19:36] I mean, it's recommended if necessary ;) That is to say, one might need PHP 7.1 or higher and also have to be running 16.04. But with 7.2 officially in the 18.04 repos, the use-case is currently not high. As the LTS goes on that will probably change. But if you don't expressly need a version of PHP other than the one in main, it's better to stick with the officially-packaged version. [19:39] thanks for that explanation [19:39] i remember seeing perf benefits to php 7.1 and 7.2 over 7 [19:39] but i also think i remember that 7 is a big step up from 5.x [19:40] Yeah and there were some missing features in 7.0 that were in 5.x, that were re-added finally with 7.1 (which is why I'm still using ondrej's PPA on the Phabricator server at work, which I haven't yet updated to 18.04). [19:41] what are you missing in 7.0 that you have in 7.1? [19:41] br [19:42] arooni: Not 100% sure, just stuff Phabricator needed, heh. Looking at https://secure.phabricator.com/T9640 I guess it was primarily "asynchronous signal handling". [20:24] coreycb: I updated this bug with some details https://bugs.launchpad.net/cloud-archive/+bug/1758094 [20:24] Launchpad bug 1758094 in Ubuntu Cloud Archive "ironic-conductor fails to start in Pike" [Undecided,New] [20:24] we are still hitting this issue in CI https://review.openstack.org/#/c/602059/ [20:25] could you please look at it when you are back [20:25] jamespage: fyi ^ [20:27] tobias-urdin: looking [22:07] i think my packages are broken; cant seem to install pmp-memcahced [22:07] https://gist.github.com/arooni/f1efd9f1db704732eea1043187846170 [22:07] arooni: i can help you once i'm off the phone [22:08] no rush ; just a noob here trying his best lol ;) [22:08] arooni: your php-common is i think from ondrej still [22:08] arooni: `apt-cache policy php-memcached php-common` ? [22:08] ahh ha; i did run ppa-purge as recommended earlier [22:08] but i think it didnt remove all the old ppa packages [22:09] ok [22:09] it should have [22:09] https://gist.github.com/74d4b80b76cf8df5690710bcf5e34749 [22:13] arooni: right, your php-common is still wrong :/ [22:14] arooni: you can satisfy it manually with (on the apt-get line php-common=...) [22:14] so do i just need to uninstall it [22:14] what about uninstalling php-common and reinstalling it [22:14] shouldnt it pull from official repos? [22:14] also would owrk [22:15] Reinstallation of php-common is not possible, it cannot be downloaded. [22:15] i really butchered this didnt i [22:23] uninstalled all my php packages and reinstalled them [22:23] seems to have fixed it [22:28] arooni: :) [22:30] is apxs something that runs only on apache or on nginx too [22:50] tobias-urdin: update for oslo.utils now in pike-proposed [22:50] hopefully that will resolve your issue