| === cpaelzer_ is now known as cpaelzer | ||
| Odd_Bloke | smoser: So I'm looking at /sys, and I can't see how to find the logical sector size of a block device. | 09:36 |
|---|---|---|
| ndonegan | Is it possible to make cloud-init run an abitrary script after it has gone through it's other sources? | 10:26 |
| larsks | Odd_Bloke: I don't know about /sys, but this works, I think: blockdev --getbsz /dev/sda | 13:11 |
| Odd_Bloke | larsks: Yep, blockdev would do it; we're trying to avoid forking if we can. | 13:16 |
| larsks | It's just calling the BLKBSZGET ioctl. I guess you could do that natively in python w/ the 'fcntl' module... | 13:17 |
| smoser | Odd_Bloke, i put iot in my comments. | 13:19 |
| smoser | oh carp. looks like i didn't hit 'submit' on some comments. | 13:20 |
| smoser | but anyway, its /sys/class/block/<devname>/qeueu/logical_block_size | 13:21 |
| smoser | and physical_block_size | 13:21 |
| smoser | ndonegan, yes. just send script as user-data . if it starts with '#!' it gets executed. | 13:22 |
| smoser | sucks that i lost that comment. | 13:22 |
| smoser | Odd_Bloke, i had also rquessted that we round reasonably well (to 1MB or 4MB) increments so we land on fs alignment. and also start at 1MB in. | 13:22 |
| ndonegan | smoser: Having issues with a certain meta data provider, hence why I want to be able to run a script after which verifies stuff happened. | 13:23 |
| Odd_Bloke | smoser: If {logical,physical}_block_size are actually sector sizes, why is there also hw_sector_size? | 13:23 |
| smoser | well, i dont know. in my tests, logical and physical match up with what i fed through qemu-system-x86 | 13:24 |
| smoser | and qemu calls them 'logical_block_size' and 'physical_block_size' | 13:25 |
| smoser | so i'm justified in my assumption (as those names line up). but i admittedly did make an assumption :) | 13:26 |
| smoser | ndonegan, i dont really follow. | 13:26 |
| ndonegan | smoser: Having great fun with metadata proxy where it occasionally sends back crap. | 13:27 |
| ndonegan | As an example, the request it should be sending to nova-metadata-api gets sent as the resonse to the client. | 13:28 |
| ndonegan | So looking for some way to try and very that some valid data was sent back | 13:28 |
| ndonegan | *verify | 13:29 |
| Odd_Bloke | ndonegan: You're looking to change the image in order to do this, or configure cloud-init to do it from outside the instance? | 13:30 |
| ndonegan | Odd_Bloke: Exploring any option, including edits to the image, to see if there's some way to verify that what comes back from the metadata service is correct. | 13:31 |
| smoser | ndonegan, as in openstack ? | 13:31 |
| smoser | yeah. | 13:31 |
| ndonegan | It would actually be better if the proxy just failed entirely and timed out. That way I can have the "None" data source as a fallback, and let it write a raw userdata | 13:32 |
| ndonegan | However, it sends back just enough before screwing up that the Ec2 source doesn't fail outright. | 13:32 |
| ndonegan | smoser: Openstack with Contrail. | 13:32 |
| smoser | right. | 13:32 |
| ndonegan | And it's Contrail that's the issue. | 13:32 |
| smoser | 2 things you could try | 13:32 |
| smoser | a.) cirros | 13:32 |
| smoser | b.) if you can enable config drive in launching... then you can probably avoid cloud-init hitting the md so you can get in and debug from there. | 13:33 |
| smoser | ie, launch with 'config_drive=1'. | 13:34 |
| smoser | i suggest cirros because it will time out faster and you can log in with password : cirros/cubswin:) | 13:34 |
| Odd_Bloke | smoser: So, looking at the kernel source, hw_sector_size is the same value as... logical_block_size. | 13:44 |
| ndonegan | We already have a base image we use everywhere (mini Centos setup for our environment). | 13:45 |
| ndonegan | ConfigDrive might be a good option. | 13:45 |
| smoser | Odd_Bloke, i would not have thoguth that :) | 13:45 |
| mwak | smoser: I updated the PR to ensure we're on Scaleway when using the datasource, let me know if it's ok for you. I'll add unit test asap | 13:53 |
| smoser | mwak, and sign cla ? | 13:54 |
| smoser | mwak, the tewst there is fine, and does help some. | 13:55 |
| smoser | but.. we'd like to avoid hitting a network service if we can | 13:55 |
| smoser | as any connection to a network service might hang indefinitely | 13:56 |
| Odd_Bloke | smoser: OK, the BLKSSZGET ioctl calls bdev_logical_block_size, which calls queue_logical_block_size which is also called for the logical_block_size sysfs entry. | 13:56 |
| smoser | ie, in many environments a 'wget http://169.254.169.254/' will just block due to firewall rules and the like. | 13:56 |
| Odd_Bloke | smoser: So it looks like we can use /sys/.../logical_block_size. :) | 13:56 |
| smoser | Odd_Bloke, i'd like to fall back if that file isnt present. you plan on that , right ? | 13:57 |
| Odd_Bloke | smoser: Yeah, that was the plan. | 13:57 |
| smoser | ok. the other thing i didnt' say, that got lost in my browser. | 13:58 |
| smoser | if we can start first partition at 1MB and only use 1MB units in sizes, that'd be good. | 13:59 |
| smoser | i'm tempted to use 4MB units. | 13:59 |
| smoser | https://bugs.launchpad.net/maas/+bug/1502259 | 13:59 |
| smoser | lvm uses 4MB extents by default. | 14:00 |
| natorious | morning | 14:01 |
| mwak | smoser: yep sign the cla too | 14:03 |
| smoser | hey | 14:04 |
| smoser | mwak, so does that make sense above ^ | 14:05 |
| smoser | is there any way that we can determine "am i on scaleway" without hitting a network resource ? | 14:05 |
| smoser | hey. it is time for cloud-inti 2.0 meeting | 14:05 |
| mwak | smoser: not really | 14:05 |
| jgrimm | o/ | 14:05 |
| * Odd_Bloke is in a partner meeting, so is MIA. | 14:05 | |
| natorious | channel topic links need to be updated w/ new gerrit paths | 14:06 |
| natorious | ie - https://review.openstack.org/#/q/project:openstack/cloud-init+status:open,n,z | 14:06 |
| smoser | ah. yeah, i'll do that natorious | 14:06 |
| smoser | ok. so looking at that fine url that natiorisou added. | 14:07 |
| natorious | lol | 14:08 |
| smoser | a couple that i shoudl grab quickly | 14:08 |
| smoser | as they address the namespace change. | 14:08 |
| smoser | natorious, had you made any more progress on yours? | 14:08 |
| natorious | was just looking over harlowja's comments | 14:09 |
| smoser | k | 14:09 |
| natorious | ideally it would be nice if we had a utils.execute or similar shared subprocess call for exec | 14:10 |
| smoser | natorious, yeah. there is a fairly good one in cluod-init | 14:11 |
| smoser | utils.subp | 14:11 |
| smoser | in 0.7 | 14:12 |
| smoser | that i'd suggest we pull forward | 14:12 |
| natorious | k | 14:14 |
| natorious | assuming it is compatible with windows and all | 14:14 |
| smoser | ah. yeah, it would need some windows work likely. | 14:15 |
| smoser | it wrapps subprocess. | 14:15 |
| smoser | subprocess.popen | 14:15 |
| smoser | but i'd guess we could address anything non windows specific there | 14:15 |
| niluje | @smoser | ie, in many environments a 'wget http://169.254.169.254/' will just block due to firewall rules and the like. >> do you think we should add a timeout for the test? | 14:15 |
| niluje | that can easily be done with requests | 14:16 |
| smoser | niluje, for tests ? | 14:16 |
| niluje | to check the server's is on scaleway | 14:16 |
| niluje | (I'm working with mwak there) | 14:16 |
| smoser | yes, for sure. | 14:17 |
| smoser | timeouts suck | 14:17 |
| niluje | we thought about it for a while, and doing something else than a network test didn't seem straightforward | 14:17 |
| smoser | because you're either safe (meaning unnecessary long wait) or aggressive (meaning your endpoint might not return in time) | 14:17 |
| niluje | yes, I agree | 14:18 |
| smoser | niluje, are you in a place wehere you could feed some dmi data to the guest ? | 14:18 |
| smoser | it seems like generally good practice to identify your platform in such a way | 14:18 |
| smoser | wrt timeouts... i dont want to enable a datasource by default that could block boot as it waits. I'd *like* to change EC2 to not do that. | 14:19 |
| smoser | but EC2 was "first" so i'm ok giving them a legacy grandfather-in. | 14:19 |
| niluje | I totally understand your point of view :) | 14:20 |
| === jgrimm is now known as jgrimm-brb | ||
| niluje | unfortunately, at the moment, we can't really rely on something else than network resource and if we do, we won't be sure the test will still work in a few months/years | 14:21 |
| niluje | we prefered to add this not-so-good-test rather than a better test that might not work in a while | 14:22 |
| niluje | however if that's a blocker, we'll think about another solution | 14:22 |
| === jgrimm-brb is now known as jgrimm | ||
| natorious | smoser: any ideas on windows configdrive detection? | 14:25 |
| smoser | natorious, i might look at how cloudbase has done it. | 14:26 |
| smoser | i think they read the device with bsd tar | 14:26 |
| smoser | ! | 14:26 |
| smoser | you cant make that stuff up :) | 14:26 |
| smoser | but at least at one point that was what alexpilloti settled on. | 14:27 |
| smoser | (honestly, i wuldnt mind having a vfat and iso9660 filesystem reader in pure python) | 14:27 |
| natorious | oh wow, lol | 14:27 |
| natorious | yeah, looked at ctypes methods a bit | 14:27 |
| natorious | should be universal but alot less readable | 14:28 |
| smoser | i suspect also faster than 'mount ... read ... unmont' for the amount of data we're pulling. | 14:28 |
| natorious | uni to windows that is | 14:28 |
| natorious | k, I'll see how their doing it and go from there | 14:30 |
| === cliles_ is now known as cliles | ||
| === shardy is now known as shardy_afk | ||
| === pratikma_ is now known as pratikmallya | ||
| === gondoi is now known as zz_gondoi | ||
| === shardy_afk is now known as shardy | ||
| === jgrimm is now known as jgrimm-afk | ||
| === jgrimm-afk is now known as jgrimm | ||
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!