2012-12-11 4:36, Saggi Mizrahi:
----- Original Message -----
From: "Adam Litke" agl@us.ibm.com To: "Saggi Mizrahi" smizrahi@redhat.com Cc: "Shu Ming" shuming@linux.vnet.ibm.com, "engine-devel" engine-devel@ovirt.org, "VDSM Project Development" vdsm-devel@lists.fedorahosted.org Sent: Monday, December 10, 2012 1:39:51 PM Subject: Re: [vdsm] RFC: New Storage API
On Thu, Dec 06, 2012 at 11:52:01AM -0500, Saggi Mizrahi wrote:
----- Original Message -----
From: "Shu Ming" shuming@linux.vnet.ibm.com To: "Saggi Mizrahi" smizrahi@redhat.com Cc: "VDSM Project Development" vdsm-devel@lists.fedorahosted.org, "engine-devel" engine-devel@ovirt.org Sent: Thursday, December 6, 2012 11:02:02 AM Subject: Re: [vdsm] RFC: New Storage API
Saggi,
Thanks for sharing your thought and I get some comments below.
Saggi Mizrahi:
I've been throwing a lot of bits out about the new storage API and I think it's time to talk a bit. I will purposefully try and keep implementation details away and concentrate about how the API looks and how you use it.
First major change is in terminology, there is no long a storage domain but a storage repository. This change is done because so many things are already called domain in the system and this will make things less confusing for new-commers with a libvirt background.
One other changes is that repositories no longer have a UUID. The UUID was only used in the pool members manifest and is no longer needed.
connectStorageRepository(repoId, repoFormat, connectionParameters={}): repoId - is a transient name that will be used to refer to the connected domain, it is not persisted and doesn't have to be the same across the cluster. repoFormat - Similar to what used to be type (eg. localfs-1.0, nfs-3.4, clvm-1.2). connectionParameters - This is format specific and will used to tell VDSM how to connect to the repo.
Where does repoID come from? I think repoID doesn't exist before connectStorageRepository() return. Isn't repoID a return value of connectStorageRepository()?
No, repoIDs are no longer part of the domain, they are just a transient handle. The user can put whatever it wants there as long as it isn't already taken by another currently connected domain.
disconnectStorageRepository(self, repoId)
In the new API there are only images, some images are mutable and some are not. mutable images are also called VirtualDisks immutable images are also called Snapshots
There are no explicit templates, you can create as many images as you want from any snapshot.
There are 4 major image operations:
createVirtualDisk(targetRepoId, size, baseSnapshotId=None, userData={}, options={}):
targetRepoId - ID of a connected repo where the disk will be created size - The size of the image you wish to create baseSnapshotId - the ID of the snapshot you want the base the new virtual disk on userData - optional data that will be attached to the new VD, could be anything that the user desires. options - options to modify VDSMs default behavior
returns the id of the new VD
I think we will also need a function to check if a a VirtualDisk is based on a specific snapshot. Like: isSnapshotOf(virtualDiskId, baseSnapshotID):
No, the design is that volume dependencies are an implementation detail. There is no reason for you to know that an image is physically a snapshot of another. Logical snapshots, template information, and any other information can be set by the user by using the userData field available for every image.
Statements like this make me start to worry about your userData concept. It's a sign of a bad API if the user needs to invent a custom metadata scheme for itself. This reminds me of the abomination that is the 'custom' property in the vm definition today.
In one sentence: If VDSM doesn't care about it, VDSM doesn't manage it.
userData being a "void*" is quite common and I don't understand why you would thing it's a sign of a bad API. Further more, giving the user choice about how to represent it's own metadata and what fields it want to keep seems reasonable to me. Especially given the fact that VDSM never reads it.
The reason we are pulling away from the current system of VDSM understanding the extra data is that it makes that data tied to VDSMs on disk format. VDSM on disk format has to be very stable because of clusters with multiple VDSM versions. Further more, since this is actually manager data it has to be tied to the manager backward compatibility lifetime as well. Having it be opaque to VDSM ties it to only one, simpler, support lifetime instead of two.
Making userData being opaque gives flexibilities to the management applications. To me, opaque userDaa can have two types at least. The first is the userData for runtime only. The second is the userData expected to be persisted into the metadata disk. For the first type, the management applications can store their own data structures like temporary task states, VDSM query caches &etc. After the VDSM host is fenced, the userData goes away. For the second type, different management applications can have different formats of userData and persisting it to the VDSM metadata disk will corrupt each other in very high possibility. You may argue that, all the management applications should reach a agreement before they use it as the second type. But in practice, it is pretty hard to maintain.
I guess you are implying that it will make it problematic for multiple users to read userData left by another user because the formats might not be compatible. The solution is that all parties interested in using VDSM storage agree on format, and common fields, and supportability, and all the other things that choosing a supporting *something* entails. This is, however, out of the scope of VDSM. When the time comes I think how the userData blob is actually parsed and what fields it keeps should be discussed on ovirt-devel or engine-devel.
The crux of the issue is that VDSM manages only what it cares about and the user can't modify directly. This is done because everything we expose we commit to. If you want any information persisted like:
- Human readable name (in whatever encoding)
- Is this a template or a snapshot
- What user owns this image
You can just put it in the userData. VDSM is not going to impose what encoding you use. It's not going to decide if you represent your users as IDs or names or ldap queries or Public Keys. It's not going to decide if you have explicit templates or not. It's not going to decide if you care what is the logical image chain. It's not going to decide anything that is out of it's scope. No format is future proof, no selection of fields will be good for any situation. I'd much rather it be someone else's problem when any of them need to be changed. They have currently been VDSMs problem and it has been hell to maintain.
createSnapshot(targetRepoId, baseVirtualDiskId, userData={}, options={}): targetRepoId - The ID of a connected repo where the new sanpshot will be created and the original image exists as well. size - The size of the image you wish to create baseVirtualDisk - the ID of a mutable image (Virtual Disk) you want to snapshot userData - optional data that will be attached to the new Snapshot, could be anything that the user desires. options - options to modify VDSMs default behavior
returns the id of the new Snapshot
copyImage(targetRepoId, imageId, baseImageId=None, userData={}, options={}) targetRepoId - The ID of a connected repo where the new image will be created imageId - The image you wish to copy baseImageId - if specified, the new image will contain only the diff between image and Id. If None the new image will contain all the bits of image Id. This can be used to copy partial parts of images for export. userData - optional data that will be attached to the new image, could be anything that the user desires. options - options to modify VDSMs default behavior
Does this function mean that we can copy the image from one repository to another repository? Does it cover the semantics of storage migration, storage backup, storage incremental backup?
Yes, the main purpose is copying to another repo. and you can even do incremental backups. Also the following flow
- Run a VM using imageA
- write to disk
- Stop VM
- copy imageA to repoB
- Run a VM using imageA again
- Write to disk
- Stop VM
- Copy imageA again basing it of imageA_copy1 on repoB creating a
diff on repo diff without snapshotting the original image.
return the Id of the new image. In case of copying an immutable image the ID will be identical to the original image as they contain the same data. However the user should not assume that and always use the value returned from the method.
removeImage(repositoryId, imageId, options={}): repositoryId - The ID of a connected repo where the image to delete resides imageId - The id of the image you wish to delete.
getImageStatus(repositoryId, imageId) repositoryId - The ID of a connected repo where the image to check resides imageId - The id of the image you wish to check.
All operations return once the operations has been committed to disk NOT when the operation actually completes. This is done so that:
- operation come to a stable state as quickly as possible.
- In case where there is an SDM, only small portion of the
operation actually needs to be performed on the SDM host.
- No matter how many times the operation fails and on how many
hosts, you can always resume the operation and choose when to do it.
- You can stop an operation at any time and remove the
resulting object making a distinction between "stop because the host is overloaded" to "I don't want that image"
This means that after calling any operation that creates a new image the user must then call getImageStatus() to check what is the status of the image. The status of the image can be either optimized, degraded, or broken. "Optimized" means that the image is available and you can run VMs of it. "Degraded" means that the image is available and will run VMs but it might be a better way VDSM can represent the underlying data.
What does the "represent" mean here?
Anything, but mostly image formate RAW\QCOW2 when performance strategy has been selected.
"Broken" means that the image can't be used at the moment, probably because not all the data has been set up on the volume.
Apart from that VDSM will also return the last persisted status information which will conatin hostID - the last host to try and optimize of fix the image
Any host can optimize the image? No need to be SDM?
On anything but lvm based block domains there will not even be an SDM. On SDM based domains we will try as hard as we can to have as many operations executable on any host.
stage - X/Y (eg. 1/10) the last persisted stage of the fix. percent_complete - -1 or 0-100, the last persisted completion percentage of the aforementioned stage. -1 means that no progress is available for that operation. last_error - This will only be filled if the operation failed because of something other then IO or a VDSM crash for obvious reasons. It will usually be set if the task was manually stopped
The user can either be satisfied with that information or as the host specified in host ID if it is still working on that image by checking it's running tasks.
So we need a function to know what tasks are running on the image
getImageStatus()
checkStorageRepository(self, repositoryId, options={}): A method to go over a storage repository and scan for any existing problems. This includes degraded\broken images and deleted images that have no yet been physically deleted\merged. It returns a list of Fix objects. Fix objects come in 4 types: clean - cleans data, run them to get more space. optimize - run them to optimize a degraded image merge - Merges two images together. Doing this sometimes makes more images ready optimizing or cleaning. The reason it is different from optimize is that unmerged images are considered optimized. mend - mends a broken image
The user can read these types and prioritize fixes. Fixes also contain opaque FIX data and they should be sent as received to fixStorageRepository(self, repositoryId, fix, options={}):
That will start a fix operation.
All major operations automatically start the appropriate "Fix" to bring the created object to an optimize\degraded state (the one that is quicker) unless one of the options is AutoFix=False. This is only useful for repos that might not be able to create volumes on all hosts (SDM) but would like to have the actual IO distributed in the cluster.
Other common options is the strategy option: It has currently 2 possible values space and performance - In case VDSM has 2 ways of completing the same operation it will tell it to value one over the other. For example, whether to copy all the data or just create a qcow based of a snapshot. The default is space.
You might have also noticed that it is never explicitly specified where to look for existing images. This is done purposefully, VDSM will always look in all connected repositories for existing objects. For very large setups this might be problematic. To mitigate the problem you have these options: participatingRepositories=[repoId, ...] which tell VDSM to narrow the search to just these repositories and imageHints={imgId: repoId} which will force VDSM to look for those image ID just in those repositories and fail if it doesn't find them there. _______________________________________________ vdsm-devel mailing list vdsm-devel@lists.fedorahosted.org https://lists.fedorahosted.org/mailman/listinfo/vdsm-devel
--
舒明 Shu Ming Open Virtualization Engineerning; CSTL, IBM Corp. Tel: 86-10-82451626 Tieline: 9051626 E-mail: shuming@cn.ibm.com or shuming@linux.vnet.ibm.com Address: 3/F Ring Building, ZhongGuanCun Software Park, Haidian District, Beijing 100193, PRC
vdsm-devel mailing list vdsm-devel@lists.fedorahosted.org https://lists.fedorahosted.org/mailman/listinfo/vdsm-devel
-- Adam Litke agl@us.ibm.com IBM Linux Technology Center