| flaccid | smoser: i took your advice; now i am doing ec2-bundle-image; just about to test one | 04:39 |
|---|---|---|
| flaccid | erichammond, smoser : i noticed that bundling an image and uploading takes longer, i guess this is because its a full uncompressed img, in this case 10GB, right? | 04:39 |
| erichammond | flaccid: bundling and uploading an image takes longer than what? | 05:25 |
| flaccid | just saying that ec2-bundle-image takes longer than ec2-bundle-vol | 05:25 |
| flaccid | because the image has a physical size | 05:26 |
| flaccid | thus why lots of parts compared to the compressed userland on bundle vol | 05:26 |
| erichammond | flaccid: Ah. I use bundle-vol. It avoids potential security issues with deleted files and lets you start up faster. | 05:27 |
| flaccid | yeah, i found the opposite with the security files thing | 05:27 |
| erichammond | flaccid: However, I think S3 based AMIs are obsolescent and should only be used for rare cases. EBS boot should be the default for most folks. | 05:27 |
| flaccid | so you think ebs boot even if you are not going to bundle again i.e. true appliance? | 05:28 |
| erichammond | flaccid: With EBS boot you snapshot instead of bundling. You can register the snapshot as a new AMI if you want. | 05:28 |
| flaccid | yeah | 05:29 |
| flaccid | i'm talking about re-snaphotting in this case for persistence | 05:29 |
| flaccid | i think you can ebs vol boot as well | 05:29 |
| flaccid | 625seconds to bundle from an image | 05:36 |
| flaccid | that is considerably more | 05:36 |
| flaccid | 36 parts | 05:36 |
| flaccid | upload 72 seconds | 05:37 |
| flaccid | good rule of thumb to upload image parts and manifest into a dedicated 'sub-bucket' ? | 06:17 |
| flaccid | the bucket can get large with lots of images | 06:20 |
| flaccid | e.g. i just did rightscale-rightimages/rightimage_debian_squeeze_i386_server_v5.4.6_20100628.1.manifest.xml | 06:20 |
| flaccid | rightscale-rightimages will get very big with child objects | 06:20 |
| flaccid | looks like it takes about an extra 3mins for startup with a bundled image | 06:31 |
| flaccid | yeah hmm 6 mins to go from pending to booting | 07:31 |
| flaccid | i guess its unpacking a 10GB | 07:31 |
| erichammond | flaccid: There's a fairly low limit on buckets, so I put bundles into relatively few buckets. Of course, I recommending using EBS boot instead and with EBS boot, you don't have to worry about buckets. | 08:46 |
| flaccid | how does ebs boot work - do you boot directly from an ebs volume that is restored from an ebs snapshot ? | 08:47 |
| erichammond | flaccid: Yes. http://aws.amazon.com/about-aws/whats-new/2009/12/03/amazon-ec2-instances-now-can-boot-from-amazon-ebs/ | 08:48 |
| flaccid | yeah so you just pay a bit more for the ebs usage right? | 08:48 |
| erichammond | ye | 08:49 |
| erichammond | s | 08:49 |
| flaccid | so you make persistent by taking an ebs snapshot and restoring/booting from that? | 08:49 |
| erichammond | The root EBS volume is persistent itself. You can snapshot to improve reliability. | 08:50 |
| erichammond | You can also register a snapshot as an AMI to use as the base for new instances. | 08:50 |
| flaccid | yeah, the equiv of a rebundle to s3 right? | 08:50 |
| erichammond | Depends on your use, but generally. | 08:50 |
| flaccid | i assume the snapshot size is irrelevant to the size of your volume | 08:51 |
| flaccid | which means the snapshot is only the size of the data in the fs | 08:51 |
| erichammond | Just like normal EBS volume snapshots. Depends on the data onthe disk and the blocks modified since the last snapshot. | 08:52 |
| flaccid | yep the delta | 08:52 |
| flaccid | i'm just thinking about the process when using a bootstrapped ami mounted to loopback.. | 08:52 |
| erichammond | You create the original AMI by copying the file system over to an EBS volume and creating a snapshot. | 08:53 |
| flaccid | yep i'm just thinking of avoiding the copying | 08:53 |
| flaccid | don't think its possible | 08:54 |
| flaccid | rsync -avz | 08:54 |
| erichammond | If you don't copy, then you have more modified blocks on the EBS volume including any deleted files. | 08:54 |
| erichammond | It is possible, but I recommend the rsync. No need to use -z | 08:55 |
| flaccid | ok thanks | 08:55 |
| flaccid | basically everything is strapped to a .img | 08:55 |
| flaccid | then s3 or ebs are options after that.. | 08:55 |
| erichammond | You don't need to use loopback. You can debootstrap to a subdirectory | 08:56 |
| erichammond | assuming you're still building Debian. | 08:56 |
| flaccid | yeah but my build process uses an image so you can also archive that | 08:57 |
| flaccid | and optionally remount and check prior to bundling/regstration | 08:57 |
| erichammond | ok | 08:58 |
| flaccid | its handy for the possibilities of generic images between ec2 and eucalyptus | 08:59 |
| erichammond | flaccid: Yep, that's what Canonical is doing. I start my EBS boot AMIs by downloading their pre-built images. | 09:12 |
| flaccid | sweet as | 09:13 |
| === dendro-afk is now known as dendrobates | ||
| === dendrobates is now known as dendro-afk | ||
| === dendro-afk is now known as dendrobates | ||
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!