| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2022-07-26 | |||
| 16:46:33 | bauzas | https://docs.openstack.org/arch-design/design-storage/design-storage-concepts.html | |
| 16:46:48 | bauzas | Ephemeral storage - If you only deploy OpenStack Compute service (nova), by default your users do not have access to any form of persistent storage. The disks associated with VMs are ephemeral, meaning that from the user’s point of view they disappear when a virtual machine is terminated. | |
| 16:47:07 | bauzas | we name them "ephemeral" because we delete the disk when the instance is removed | |
| 16:47:21 | gibi | and we delete volumes as well if so configured :D | |
| 16:47:26 | sean-k-mooney | whihc is not correct | |
| 16:47:40 | sean-k-mooney | deleting somting at the end of its lifetime does not make it epmeral | |
| 16:48:05 | bauzas | technically, you can have ephemeral disks on shared storage (NFS) | |
| 16:48:07 | sean-k-mooney | if we delete a volume we also expect the data to be removed | |
| 16:48:14 | sean-k-mooney | or ceph | |
| 16:48:20 | sean-k-mooney | if you use the rbd image_backend | |
| 16:48:37 | bauzas | sean-k-mooney: I don't disagree with you, I'm just explaining our docs | |
| 16:49:17 | sean-k-mooney | yep | |
| 16:50:05 | bauzas | not only our upstream docs btw... https://access.redhat.com/documentation/en-us/red_hat_openstack_platform/16.2/html/creating_and_managing_instances/con_types-of-instance-storage_osp :) | |
| 16:50:12 | sean-k-mooney | so i would prefer something like local_disk_encryption, instance_disk_encypriton or similar | |
| 16:50:29 | sean-k-mooney | bauzas: right but there are many other things wrong with our downstream docs | |
| 16:50:32 | gibi | +1 on local disk encryption | |
| 16:51:55 | bauzas | I'd prefer local disk encryption over instance disk encryption | |
| 16:51:58 | sean-k-mooney | im not going to block this progressing on the name | |
| 16:52:22 | sean-k-mooney | i just find the usage of ephmeral to rever to storage tied to the vms lifecycle to be mildly insulting | |
| 16:52:26 | bauzas | because an instance disk can either be "ephemeral" (from local libvirt disk), or something else | |
| 16:53:02 | bauzas | technically, this ephemeral storage is actually our virt driver local storage | |
| 16:53:21 | bauzas | depending on the virt driver you use | |
| 16:53:30 | sean-k-mooney | its not nessiarly local | |
| 16:53:33 | bauzas | tell ironic about it :) | |
| 16:53:44 | sean-k-mooney | well for libvirt or vmware | |
| 16:53:47 | bauzas | yeah, see, I made that mistake | |
| 16:53:51 | sean-k-mooney | the storage can be clusterd or remote | |
| 16:53:55 | bauzas | shared storage is ephemeral | |
| 16:54:05 | bauzas | so, not local | |
| 16:54:08 | sean-k-mooney | this is why i dont like using that term | |
| 16:54:18 | melwitt | hey o/ thanks for discussing the naming bikeshed :) just wanted to get some confidence that ppl would prefer a name change and lessen the possibility of someone coming to the review later and saying "why did you change this, it should be changed back to ephemeral" | |
| 16:54:27 | sean-k-mooney | ephemeral is ambiguious | |
| 16:54:38 | bauzas | as is "evacuate" :D | |
| 16:54:53 | bauzas | our Gods of Naming didn't help | |
| 16:55:09 | sean-k-mooney | maybe we need dansmith to write another blog post | |
| 16:55:20 | sean-k-mooney | melwitt: ack so i guess the real question is | |
| 16:55:20 | bauzas | problem solved. | |
| 16:55:28 | bauzas | #action dansmith to write a blogpost | |
| 16:55:31 | sean-k-mooney | do we think channging the name will make thing clearer | |
| 16:55:35 | dansmith | heh | |
| 16:55:45 | bauzas | oh, snap, he saw it | |
| 16:55:46 | opendevmeet | Removing item from minutes: #action dansmith to write a blogpost | |
| 16:55:46 | bauzas | #undo | |
| 16:55:46 | melwitt | :) | |
| 16:55:52 | dansmith | In general, I'm not for renaming things like this | |
| 16:56:05 | bauzas | I'm not against renaming ity | |
| 16:56:17 | bauzas | I just wanted to make sure we all agree on what this is | |
| 16:56:30 | bauzas | ephemeral is a bad name, but that's a name we already use | |
| 16:56:34 | dansmith | because you'll end up with all old docs being inaccurate for new stuff, and people who already understand this will also have to change | |
| 16:56:46 | melwitt | that was one of my concerns | |
| 16:56:47 | bauzas | if we pick something else, this has to be better understandable about what it is | |
| 16:57:19 | bauzas | yeah, if we need to write some doc explaining "ephemeral" == "this new thing" this is bad | |
| 16:57:39 | bauzas | hence the challenge | |
| 16:58:00 | gibi | so this is considered a non fixable terminology mistake of the past? | |
| 16:58:08 | bauzas | like tenant ? :) | |
| 16:58:15 | gibi | we are fixing tenant | |
| 16:58:22 | sean-k-mooney | well to me that doc that is erfernce is not a nova doc | |
| 16:58:25 | bauzas | I know, I'm opening a can of worms | |
| 16:58:26 | sean-k-mooney | so we could jsut fix it | |
| 16:58:45 | bauzas | and pretend it never existed, heh ? :) | |
| 16:59:09 | bauzas | we're running out of time, but for the sake of the conversation, let's continue | |
| 16:59:13 | sean-k-mooney | well from my point of view the only thing that nova ever said was ephemreal is the falvor.ephemeral storage disks | |
| 16:59:22 | bauzas | I'll just formally end the meeting at the top of the hour | |
| 16:59:56 | sean-k-mooney | melwitt: are we encypting the flavor.epmermal disks by the way | |
| 17:00:00 | sean-k-mooney | or just root and swap | |
| 17:00:11 | bauzas | problem is | |
| 17:00:17 | bauzas | root is also "ephemeral" | |
| 17:00:17 | sean-k-mooney | i thikn we will be encypting all 3 types | |
| 17:00:28 | sean-k-mooney | bauzas: it depend on the difinition | |
| 17:00:28 | bauzas | (depending on the conf options) | |
| 17:00:31 | sean-k-mooney | form our api its not | |
| 17:00:39 | melwitt | sean-k-mooney: what is "flavor.ephemeral"? it is encrypting the root disk and any other attached local disks | |
| 17:01:09 | sean-k-mooney | in our flavor we have 3 types of storage | |
| 17:01:11 | bauzas | correct, the point is that *by default, we don't do any difference between root disk and other local (or non-local on shared) disk | |
| 17:01:21 | sean-k-mooney | root, swap and ephemeral | |
| 17:01:39 | sean-k-mooney | https://docs.openstack.org/nova/latest/user/flavors.html | |
| 17:01:50 | melwitt | ok, this is encrypting root and ephemeral, and not swap | |
| 17:01:56 | bauzas | right | |
| 17:02:00 | bauzas | about the new feature | |
| 17:02:18 | sean-k-mooney | so we proably should be encyrpting swap too but we can maybe add that next cycle | |
| 17:02:58 | bauzas | swap is out of scope AFAICT | |
| 17:03:10 | sean-k-mooney | im not sure why it would be | |
| 17:03:35 | sean-k-mooney | we declared it out of scope for this cycle i guess | |
| 17:03:47 | sean-k-mooney | but i would hope it woudl get done before we condire this fully complete | |
| 17:03:48 | bauzas | because swap isn't using QEMU file-based storage ? | |
| 17:03:54 | sean-k-mooney | it is | |
| 17:04:00 | sean-k-mooney | depending on your backend | |
| 17:04:11 | bauzas | f*** | |
| 17:04:12 | sean-k-mooney | it will use a qcow files or a rbd volume | |
| 17:04:16 | bauzas | I'm not expert on swap | |
| 17:04:41 | bauzas | then, all disks (root, swap and others) go into a same bucket | |
| 17:04:57 | bauzas | which is by default the virt driver storage backend | |
| 17:05:07 | melwitt | basically this is encrypting things that are under the 'ephemerals' and 'image' keys in block_device_info: https://review.opendev.org/c/openstack/nova/+/826529/7/nova/virt/driver.py#107 | |
| 17:05:40 | melwitt | 'swap' has its own key in block_device_info | |
| 17:06:45 | sean-k-mooney | so ephemerals should be the storage form flavor.ephemeral_gb | |
| 17:06:49 | bauzas | do we know if https://docs.openstack.org/nova/latest/configuration/config.html?highlight=ephemeral#DEFAULT.default_ephemeral_format is also used for root and swap ? | |
| 17:07:01 | sean-k-mooney | image is presumable the storage form flavor.root_gb | |
| 17:07:05 | melwitt | I don't know the reason swap is not included and I just checked the specs again and don't find it mentioned why | |
| 17:07:24 | sean-k-mooney | bauzas: no i belive that is for ephemeral_gb only | |
| 17:08:29 | sean-k-mooney | bauzas if you dont specy how you want flavor.ephemeral_gb to be devied up on the server create api request | |