2020/01/16/#cloud-init.txt

[00:20] <amansi26> Hi. I need to know, I installed a cloud-init v19.1 on a rhel8. I can see IP at /etc/sysconfig/network-script/ifcfg-env2 but when I do ip a ,IP is not getting assigned.
[00:20] <amansi26> What can be the possible flow I can check?
[00:22] <powersj> amansi26, https://cloudinit.readthedocs.io/en/latest/topics/faq.html#where-are-the-logs
[00:22] <powersj> I would review the logs in the above link
[00:29] <amansi26> powersj: I check the log, I can see the IP under " applying net config names for" and the datasource used is configdrive
[09:37] <shubas> Hi all ,is it possible to apply network config from datasource nocloud-net(for vm with pre configured temp network/dhcp) in either stage of cloudinit? asking after searching the docs and multiple testing with no success.
[09:45] <meena> shubas: can you define no success? by, say, posting your cloud-config, and your  log output?
[09:49] <shubas> meta-data and user-data are aplied but network-config arn't ,using the same files in iso all configs are aplied.
[09:52] <shubas> using default centos config with : datasource_list: [ NoCloud ]
[09:52] <shubas> datasource:
[09:52] <shubas>   NoCloud:
[09:52] <shubas>     seedfrom: http://192.168.5.5/
[09:53] <meena> okay, then lets dig into the logs, and find out why network-config isn't applied.
[10:20] <shubas> here is the log: https://send.firefox.com/download/fe67a07c819f84ac/#xI5VGQGxaK6pYqTiWJ947w
[11:07] <meena> shubas: it says: 2020-01-16 09:57:49,299 - DataSourceNoCloud.py[DEBUG]: Seed from http://192.168.5.5/ not supported by DataSourceNoCloud [seed=None][dsmode=net]
[11:12] <meena> oh, wait, that was init-local
[11:14] <meena> 2020-01-16 09:57:58,999 - stages.py[DEBUG]: applying net config names for {'version': 1, 'config': [{'subnets': [{'type': 'dhcp'}], 'type': 'physical', 'name': 'eth0', 'mac_address': 'aa:ce:dc:b2:5c:06'}]}
[11:17] <meena> so, shubas this reads fairly okay / successful to me. What about the config that you get as a result is different from what you're expecting?
[11:22] <shubas> this is the fallback config ,it does not fetch the network-config file from seed
[11:22] <meena> shubas: but it says it's doing that…
[11:23] <meena> shubas: what's curl http://192.168.5.5/meta-data look like?
[11:23] <meena> (or whatever the correct url is)
[11:23] <shubas> also checked the access log for http://192.168.5.5/ , only fetching meta-data and user-data files
[11:24] <shubas> the curl is ok ,i get the content
[11:30] <shubas> is there a way to manually fetch and apply network config from seed through cloud-init?
[11:31] <shubas> might provide more insight on what's happening
[11:37] <meena> shubas: what's the config look like?
[11:38] <shubas> cloud.cfg?
[11:44] <meena> shubas: no, i meant the thing that you get from curl
[11:49] <shubas> meta-data -
[11:49] <shubas> instance-id: someid123
[11:49] <shubas> user-data -
[11:49] <shubas> #cloud-config
[11:49] <shubas> hostname: somehostname
[11:49] <shubas> network-config -
[11:49] <shubas> network:
[11:49] <shubas> version: 1
[11:49] <shubas> config:
[11:49] <shubas>    - type: physical
[11:49] <shubas>      name: eth0
[11:49] <shubas>      subnets:
[11:49] <shubas>         - type: static
[11:49] <shubas>           address: someIP
[11:49] <shubas>           netmask: someMASK
[11:49] <shubas>           gateway: someGW
[11:49] <shubas>           dns_nameservers:
[11:49] <shubas>             - 8.8.8.8
[11:49] <shubas>             - 8.8.4.4
[11:55] <meena> not sure what, but something there isn't quite right
[11:56] <meena> would be cool if it said what, rather than just applying dhcp
[12:06] <shubas> at this point my thought is that the network config mathod is determined in the intial stage before the network seed is checked and so ignored in later stages
[12:47] <otubo> talking about configuration, meena shubas can you confirm network_state is populated by conf file?
[12:48] <otubo> I'm seeing too much code lately, need a visual confirmation :-)
[13:28] <shubas> @otubo can you clearify what do you mean by network state and where should it be populated?
[13:31] <otubo> shubas: sorry! I was able to figure out by my self. The answer is yes :)
[13:32] <otubo> shubas: network_state is the variable filled with network configuration both read from cloud.cfg and from already existing network configuration present on the guest
[13:45] <otubo> also, removing /var/lib/cloud/instance would be enough to emulate the first boot? Or should I need to do something else?
[14:02] <shubas> "cloud-init clean" will do that
[14:05] <meena> cloud-init clean --logs --reboot \o/
[14:11] <otubo> thanks people!
[15:30] <rharper> otubo: yeah, reading the bug now
[15:36] <otubo> rharper: _render_networkmanager_conf() is writing 99-cloud-init.conf *after* NM starts, and that's causing the problem
[15:37] <rharper> otubo: so, the design for the network config  by cloud-init is that at local time, we crawl imds for network config, and write out this config to the os dirs, *before* the networking service starts;  so in your case, we've written sysconfig *and* resolv.conf values before NM starts
[15:37] <rharper> otubo: so why does network-manager start before cloud-init-local.service  ?
[15:38] <rharper> Before=NetworkManager.service
[15:38] <rharper> Before=network-pre.target
[15:38] <rharper> we run local before NM and network-pre.target;  this is when we write out all of our network config, so the 99 file *is* written before NM starts
[15:38] <rharper> or something is very wrong with units
[15:40] <otubo> So what you're saying is that on first boot, resolv.conf comes preconfigured for some reason, cloud-init writes dns=none and NM starts. And even though this happens NM wipes out the file on shutdown
[15:40] <rharper> otubo: and " it does not change the resolv.conf file and this file is clean (reverted to clean state by NetworkManager during the first shutdown)."   seems to be where our conflict lies;  you said that NM does this because it started before the 99-cloud-init.conf which tells NM *not* to do that
[15:41] <rharper> otubo: I don't know the contents of your resolv.conf before boot;  typically it is a symbolic link to the systemd-resolved local caching resolver; but in older images it may be an empty file
[15:42] <rharper> I know on SUSE we've had to adjust whether cloud-init writes anyting at all (we used to always write resolv.conf even if we didn't have dns values, which was fine for Ubuntu which had resolvconf managing the file)
[15:43] <otubo> rharper: ok, good thoughts. I need to confirm the exact boot sequence and if NM is processing correctly the options on the config file.
[15:43] <rharper> the correct sequence should be:  1. cloud-init local runs first, before NM, reads net config, writes out all sysconfig/NM config files 2. NM starts, and it should see the 99-cloud-init.conf which says "dns = none" which should prevent it cleaning it up on shutdown  3. cloud-init net runs after NM has started but before all things are online ...
[15:44] <rharper> is the target OS systemd based?  if so, I've found journalctl -o short-monotonic -b -u cloud-init-local.service -u NetworkManager.service -u cloud-init.service   to be useful to see when they started
[15:45] <rharper> otubo: your log seems to indicate that there may be a race here;  but if you have a complete boot log from both scenarios, we can see if there's a different bug here;
[15:46] <rharper> https://bugs.launchpad.net/cloud-init/+bug/1843334
[15:46] <ubot5> Ubuntu bug 1843334 in cloud-init "Change location of DHCP leases in CloudStack provider as it doesn't work for RHEL8" [Medium,Fix released]
[15:46] <rharper> otubo: that landed very recently, maybe that's not present in your cloud-init yet ?
[15:47] <rharper> which could explain the race
[15:53] <otubo> rharper: it looks like the boot sequence is correct, cloud-init indeed starts before NM https://pastebin.com/y65yFajd
[15:54] <rharper> otubo: yes, but I wonder if there's a race between when cloud-init-local finishes and NM starts ...
[15:54] <otubo> rharper: but I think NM ignores dns=none configuration, along the logs I can see one single entry of NM updating DNS
[15:54] <rharper> or it's not yet written
[15:55] <rharper> the ordering ensures that a unit is started before or after, but not necessariy complete
[15:55] <otubo> oh I see
[15:56] <otubo> rharper: rhel tree doesn't contain this fix for the bug 1843334 you pointed. I'll cherry pick and test.
[15:56] <ubot5> bug 1843334 in cloud-init "Change location of DHCP leases in CloudStack provider as it doesn't work for RHEL8" [Medium,Fix released] https://launchpad.net/bugs/1843334
[15:57] <rharper> that may be enough to ensure there's time for local to write it's config
[16:00] <rharper> otubo: so I think the ordering is strong enough;  however, in the case where cloud-init local is doing a dhcp for metadata and the dhcp/network response is slow; I'm wondering if we'd find that NM would start before cloud-init writes that file;  if so;  we *might* want to issue a 'systemctl try-restart NetworkManager.service'  which would restart NM IIF it was already started.
[16:01] <otubo> rharper: Well, but i believe this is a good idea even if that fix solves my problem.
[16:02] <otubo> rharper: I could write that feature in a near future
[16:02] <rharper> yeah
[16:02] <rharper> look at cloudinit/net/netplan.py  which uses the _postcmds config; it allows a Distro class to specify commands to run after rendering a network config
[16:03] <rharper> the freebsd renderer also makes use of this as another example
[16:08] <otubo> rharper: yeah, didn't work.
[16:09] <rharper> so
[16:10] <rharper> local also does this: ExecStart=/bin/touch /run/cloud-init/network-config-ready
[16:11] <rharper> hrm, we want logic like:  if cloud-init ran, NM should wait untll this file is touched before starting;  but in the case that cloud-init doesn't run (it's not activated) then NM should run whenever it normally would;
[16:12] <otubo> rharper: what about we reload configuration when cloud-init.final finishes? Like `pkill -HUP NetworkManager'
[16:13] <rharper> hrm, well, we could PreExec the reload in cloud-init.service as a drop in
[16:14] <rharper> until code changes to use the try-restart land and release
[16:25] <otubo> rharper: ok, I'm gonna try to include something with `systemctl try-restart NetworkManager.service' on _postcmds tomorrow morning. Thanks for the help :-)
[16:28] <rharper> sure
[16:32] <rharper> https://paste.ubuntu.com/p/tGyGWJBq5t/
[16:32] <rharper> otubo: that is a systemd unit drop it config which would trigger the try-restart right before cloud-init.service
[17:52] <blackboxsw_> Odd_Bloke: just pushed https://github.com/cloud-init/ubuntu-sru/pull/79 for SRU start
[17:52] <blackboxsw_> -> errand
=== blackboxsw_ is now known as blackboxsw
[19:17] <Odd_Bloke> rharper: util.subp will raise a ValueError if target is anything but None.  The docstring says this is for compatibility with curtin's subp.  Do you think that's a goal we still want to retain?
[19:18] <Odd_Bloke> (I'm asking because nothing actually calls get_dpkg_architecture with a target that isn't None, so I was going to remove that parameter, then dug deeper and found this.)
[19:38] <blackboxsw> Odd_Bloke: I just added another note about what clouds SRU verification already covers https://github.com/canonical/cloud-init/pull/167
[19:41] <blackboxsw> SRU -proposed bits are up and accessible in xenial bionic and eoan-proposed. I'm writing up the notification email now
[20:34] <Odd_Bloke> blackboxsw: https://github.com/canonical/cloud-init/pull/167 reviewed, we're getting there!
[20:54] <blackboxsw> pushed changes Odd_Bloke thanks
[20:54] <blackboxsw> and accepted all suggestions
[21:17] <Odd_Bloke> blackboxsw: LGTM now.  I have another couple of minor nits (a missing : and the directive names), which I'll apply then merge.
[21:19] <Odd_Bloke> blackboxsw: Oh, no I'm not, I don't have permissions.
[21:23] <blackboxsw> ?
[21:24] <Odd_Bloke> blackboxsw: I can't commit to your branch, so you'll need to apply this last round of nits. :)
[21:24] <blackboxsw> Odd_Bloke: I just clicked allow edits
[21:24] <blackboxsw> Odd_Bloke: does it work now I forgot to click that
[21:26] <blackboxsw> I think in general Odd_Bloke if we are working through a bunch of nits it might be easier for you to review and me to grok the set of changes if you push over my branch, then I can diff to origin and walk through the full visual diff. walking through a bunch of accept suggestion links makes me feel I'm going to miss something.
[21:27] <Odd_Bloke> OK, we can bear that in mind next time.
[21:28] <Odd_Bloke> That's now how I prefer to receive changes, which is why I didn't try to do that. :)
[21:28] <Odd_Bloke> (And also, I want you to be happy with the changes being made before I make them, unless they really are trivial!)
[21:29] <blackboxsw> makes sense. I was just feeling a bit guilty for the number of review rounds you had to do Odd_Bloke
[21:30] <blackboxsw> Odd_Bloke, so do I need to accept your last round of review comments or are you able to push to my branch?
[21:30] <Odd_Bloke> Oh, I don't mind that at all.  I've been feeling guilty about how many review rounds you had to do. :p
[21:31] <Odd_Bloke> blackboxsw: It's telling me I still don't have permissions, so why don't you accept them.
[21:31] <Odd_Bloke> We can figure that out next time.
[21:31] <blackboxsw> good we all have feelings of guilt to work out this year it seems :)
[21:31] <blackboxsw> will do
[21:33] <Odd_Bloke> rharper: blackboxsw: FYI, I've asked waveform (the Pi guy on the Foundations team) to file a bug for a mount issue they're seeing with cloud-init (I assume in Ubuntu Core), as they were talking about how to work around it.
[21:36] <Odd_Bloke> blackboxsw: You have a long line in there still, doc8 fails.
[21:37] <Odd_Bloke> doc/rtd/topics/debugging.rst:171: D001 Line too long
[21:39] <blackboxsw> d'oh
[21:39] <Odd_Bloke> The dangers of using the GH suggestions. :)
[21:39] <rharper> Odd_Bloke: oh, where was the discussion re: mount issues ?
[21:39] <rharper> bug is best though
[21:39] <Odd_Bloke> rharper: In the Foundations channel that I never /part'd. :p
[21:40] <rharper> it's more likely run with systemd but cloud-init plays a role in writing those fstab entries out
[21:40] <rharper> gotcha
[21:40] <blackboxsw> Odd_Bloke: force pushed the sru doc branch
[21:40] <blackboxsw> should be good2go
[21:41] <blackboxsw> community notice: I also just pushed a origin/stable-19.4 branch which was our last version of cloud-init to claim support for py2.7
=== blackboxsw changed the topic of #cloud-init to:  cloud-init pull-requests https://git.io/JeVed | Meeting minutes: https://goo.gl/mrHdaj | Next status meeting January 21 17:15 UTC | 19.4 (Dec 17) | 20.1 DROP py2.7 (Feb 18: origin/stable-19.4) | https://bugs.launchpad.net/cloud-init/+filebug
=== blackboxsw changed the topic of #cloud-init to:  cloud-init pull-requests https://git.io/JeVed | Meeting minutes: https://goo.gl/mrHdaj | Next status meeting January 21 17:15 UTC | 19.4 (Dec 17) | 20.1 DROP py2.7 (Feb 18) | https://bugs.launchpad.net/cloud-init/+filebug
[21:42] <Odd_Bloke> rharper: It's something to do with the cloud-init seed being on a drive that's already being mounted for some other reason; when we go to mount it, it's already mounted so *sad trombone*
[21:43] <Odd_Bloke> But a bug is incoming.
[21:43] <rharper> I suspect this is core20 fun
[21:43] <Odd_Bloke> Yeah, I don't see why a Pi would have cloud-init (in a weird configuration) on it if it weren't.
[21:43] <rharper> the writable partition includes the seed so it can be modified outside of the read-only image;
[21:44] <rharper> oh, non-core pi image has cloud-init to auto generate keys, import them and do that outside of the base image;
=== blackboxsw changed the topic of #cloud-init to:  cloud-init pull-requests https://git.io/JeVed | Meeting minutes: https://goo.gl/mrHdaj | Next status meeting January 21 17:15 UTC | 19.4 (Dec 17) drops Py2.7 : origin/stable-19.4 | 20.1 (Feb 18) | https://bugs.launchpad.net/cloud-init/+filebug
[21:44] <Odd_Bloke> Aha, OK, well we'll find out!
[21:44] <rharper> core just makes things more complicated w.r.t not having a writable filesystem
[21:44] <rharper> it was a nice improvement to allow cloud-init to read the boot directory of the pi's to look for cloud-config; so they can do a dd of the image and then write a cloud.cfg in the boot dir
[21:45] <rharper> yes, we shall see =)
[21:46] <Odd_Bloke> blackboxsw: Did you overwrite the last set of changes you applied from the GH UI?
[21:46] <Odd_Bloke> Looks like the directives have reverted.
[21:46] <blackboxsw> geez man, you're probably correct.
[21:46] <blackboxsw> I'll re-apply
[21:50] <blackboxsw> Odd_Bloke: tried to reapply, erased the code-block suggestions plus and opening cloud-init:
[21:52] <Odd_Bloke> blackboxsw: Approved. \o/
[21:54] <blackboxsw> thanks!
[22:01] <johnsonshi> Hi, whenever I run tox -e py27, I get an error about test-requirements.txt
[22:01] <johnsonshi> Requirement already satisfied: setuptools in ./.tox/py27/lib/python2.7/site-packages (from -r /source/test-requirements.txt (line 13)) (45.0.0)
[22:01] <johnsonshi> ERROR: Package 'setuptools' requires a different Python: 2.7.12 not in '>=3.5'
[22:01] <johnsonshi> ERROR: could not install deps [-r/source/test-requirements.txt]; v = InvocationError('/source/.tox/py27/bin/pip install -r/source/test-requirements.txt (see /source/.tox/py27/log/py27-1.log)', 1)
[22:02] <johnsonshi> I have not modified test-requirements.txt at all, and my tox tests were running fine until a few days ago
[22:02] <johnsonshi> Any possible reasons why this error is being thrown?
[22:19] <Odd_Bloke> johnsonshi: I believe setuptools have dropped support for Python 2 in the latest release, which is what's causing that error.
[22:20] <Odd_Bloke> johnsonshi: We've followed suit, however, and 19.4 was the last cloud-init release that supported Python 2.
[22:20] <Odd_Bloke> So you probably don't need to run `tox -e py27` any longer, and instead should be running `tox -e py3`.
[22:24] <johnsonshi> Odd_Bloke: Thanks for clearing this up!
[22:41] <Odd_Bloke> johnsonshi: Happy to help!