=== rangerpbzzzz is now known as rangerpb | ||
=== nacc_ is now known as nacc | ||
dtp | i've worked around https://bugs.launchpad.net/cloud-init/+bug/1671927, but now the gateway is not being rendered to ifcfg-eth0. gateway is defined in network_data.json, but not content/0000, so more evidence i guess that it's preferring 0000 in this case | 17:22 |
---|---|---|
=== mgagne_ is now known as mgagne | ||
smoser | dtp, can you give the config drive that you ave? | 18:17 |
dtp | smoser some of the files are available at http://imgur.com/a/qEElh | 18:17 |
dtp | and others are attached to the bug. or is there another you need? | 18:18 |
smoser | dtp, you couldnt use pastebin ? for text :) | 18:18 |
dtp | i don't have net access to the VM / only have spice console | 18:19 |
smoser | dtp, so the issue... is that cloud-init did not write that /etc/network/interfaces | 18:19 |
smoser | as it says, nova "injected" it. | 18:19 |
smoser | essentially, your cloud platform broke your image by thinking it knows what it is doing. | 18:19 |
smoser | and cloud-init does not attempt to un-break it for that scenario. | 18:19 |
smoser | that make sense ? | 18:20 |
dtp | which /etc/network/interfaces? | 18:20 |
smoser | i'm 95% certain that nova mounted the image and write /etc/network/interfaces | 18:20 |
smoser | oh wait. | 18:20 |
smoser | no. | 18:20 |
smoser | ok. let me re-think | 18:20 |
dtp | this is fedora 25 / no /etc/network/interfaces, only network-scripts/... files | 18:20 |
smoser | right | 18:20 |
dtp | so where i'm coming from is this all worked on fedora 23 w/ older cloud-init | 18:21 |
dtp | trying to get fedora 25 up | 18:21 |
smoser | you're right. it sure would appear that cloud-init is preferring to parse the 0000 file | 18:22 |
smoser | theres no way you can get the disk there? | 18:22 |
dtp | which file do you need? | 18:23 |
smoser | you shoudl be able to bring up networking by hand | 18:23 |
smoser | if you just get me the config drive (dd if=/dev/vdb of=my.img) | 18:23 |
smoser | or something to that affect, thats the easiest. | 18:23 |
dtp | ok, i'm in a meeting but will after | 18:23 |
=== shardy is now known as shardy_afk | ||
smoser | dtp, just reading code, it sure seems like it should be looking at the network json in 0.7.8 | 19:10 |
dtp | some of the other guys in here on Friday seemed to think it was skipping that stuff and using meta data or something? | 19:11 |
smoser | well if you attached a config drive then it should be reading that. | 19:12 |
smoser | there is a config drive, right ? | 19:12 |
smoser | thats what you showed with 'cat' ? | 19:12 |
smoser | yeah | 19:12 |
dtp | yeah | 19:13 |
dtp | they were saying something else in the config drive caused it to ignore network_data.json | 19:13 |
dtp | meta_data i think | 19:13 |
dtp | but i didn't see much in there | 19:13 |
smoser | hm. | 19:15 |
smoser | can you show me ls /sr0/openstack/ ? | 19:15 |
dtp | smoser - http://imgur.com/a/YXfc5 - still working on bringing network up | 19:17 |
smoser | hm. | 19:19 |
smoser | that wasnt what i expected :) | 19:19 |
dtp | oh? | 19:19 |
smoser | that should work. i expected that you were going to honly have 2016-10-06 and 2017-02-22 | 19:20 |
dtp | one thing we noticed was that the debug logs in ConfigDrive where it says (if network_data use it, else use 0000) weren't being logged | 19:20 |
dtp | so it wasn't making it there before it went to write net conf | 19:21 |
dtp | i guess the trace is in that original imgur | 19:21 |
dtp | would explain how it skipped that part | 19:22 |
smoser | well, if you can get me a copy of the cdrom i can get further. without that, ... i'm not sure whats going on. and /var/log/cloud-init.log woudl be good too | 19:22 |
dtp | yeah, that's in the imgur already - http://imgur.com/a/qEElh | 19:22 |
dtp | at bottom | 19:22 |
smoser | well, the full thing.. | 19:22 |
dtp | ok | 19:22 |
dtp | i'll keep working to get networking up on it | 19:23 |
=== dtp is now known as dtp-afk | ||
rharper | smoser: re: network passthrough/netplan support; for zesty it's too late to switch to using netplan/systemd by default; so I'd like your opinion on how best to set a policy in the Distro object (or elsewhere that the Distro can use to select the right renderer) | 19:43 |
smoser | well, rharper for ubuntu we have (i think) 3 cases: | 19:47 |
smoser | a.) ifupdown without netplan: use ifupdown | 19:47 |
smoser | b.) netplan without ifupdown: use netplan | 19:47 |
smoser | c.) ifupdown and netplan: this is where we need a choice | 19:48 |
rharper | correct; it's the (c) I was interested in | 19:48 |
smoser | and i guess | 19:48 |
smoser | d.) no ifupdown, no netplan: scream | 19:48 |
rharper | yakkety and zesty; netplan (nplan) is seeded in the image (it's in main); so we're always in (c) for yak and zesty images | 19:49 |
smoser | so i guess generically, we have a list of available "backends" that cloud-init will come up with. | 19:51 |
smoser | in 'c', that'd be: available = ['ifupdown', 'netplan'] | 19:51 |
rharper | renders | 19:51 |
smoser | or something | 19:51 |
rharper | eni and netplan | 19:51 |
rharper | and sysconfig | 19:52 |
smoser | ok. yeah. | 19:52 |
smoser | i think that the distro should coome up with an ordered list of which are available. | 19:53 |
smoser | available = ['eni', 'netplan'] | 19:54 |
smoser | for 'c' above | 19:54 |
smoser | and for 'a', that'd just be: available = ['eni'] | 19:54 |
smoser | then, i guess we just have config (config/cloud.cfg i guess) that has a order of preference | 19:55 |
smoser | system_info/network/render_priority = ['eni', 'sysconfig', 'netplan'] | 19:56 |
smoser | and pick the first entry in render_priority that is also in available | 19:56 |
smoser | ? | 19:56 |
rharper | ok, so order of the available renders is the policy | 19:57 |
rharper | so, for say UC16 (or to be named 17.10 release), we 'd update the list to indicate the preference | 19:58 |
rharper | smoser: that seems reasonable; is config only reasonable? I know we do things like enumerate the DataSource classes to build a list; do we need to do that here for net/render.py to probe for possible renderers? | 19:59 |
smoser | well, i think something has to do that. we can put a generic implementation in the distro class. | 20:01 |
rharper | sure, but I'm questioning whether we're dynamically building that list from probing cloud-init .py files (or class loading); versus returning a list of known renderer names; | 20:04 |
smoser | it has to be dynamic | 20:08 |
smoser | such that in 'a', netplan is not available | 20:08 |
smoser | we can allow a force of netplan even if the discovery says netplan isnt there. | 20:08 |
rharper | well, there's whether cloud-init has a netplan renderer vs. whether the system has the tool to consume the rendered file | 20:09 |
rharper | so, older cloud-inits won't have a config list with 'netplan' in it without a netplan class; but they system may have nplan package and netplan binary available | 20:09 |
smoser | well, older cloud-inits wont read that list | 20:10 |
smoser | they're just goign to pick eni | 20:10 |
rharper | smoser: how's this look http://paste.ubuntu.com/24178612/ | 20:35 |
smoser | rharper, hangout ? | 20:39 |
rharper | y | 20:39 |
smoser | http://c.brickies.net/hangout | 20:39 |
=== dtp-afk is now known as dtp | ||
smoser | rharper, r class loading); versus returning a list of known renderer names; | 21:11 |
smoser | oos | 21:11 |
smoser | https://bugs.launchpad.net/cloud-init/+bug/1671927 | 21:11 |
smoser | http://paste.ubuntu.com/24178813/ | 21:13 |
dtp | smoser i'm giving up on getting networking up on this thing | 22:11 |
dtp | the command i need to run to add the default route is not working | 22:11 |
dtp | i'll add more of the cloud-init log to the imgur | 22:11 |
dtp | and let me know if there's anything else i can give | 22:11 |
=== rangerpb is now known as rangerpbzzzz | ||
dtp | smoser nevermind, i got the sr0 image | 22:30 |
dtp | how do you want it? | 22:30 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!