[libvirt] QMP: Supporting off tree APIs
On Thu, 5 Jan 2012 13:48:43 +0000 Stefan Hajnoczi <stefanha@gmail.com> wrote:
On Wed, Jan 4, 2012 at 12:59 PM, Luiz Capitulino <lcapitulino@redhat.com> wrote:
On Tue, 13 Dec 2011 13:52:27 +0000 Stefan Hajnoczi <stefanha@linux.vnet.ibm.com> wrote:
Add the block_stream command, which starts copy backing file contents into the image file. Later patches add control over the background copy speed, cancelation, and querying running streaming operations.
Please also mention that you're adding an event, otherwise it comes as a surprise to the reviewer.
Ok.
When we talked about this implementation (ie. having events, cancellation etc) I thought you were going to do something very specific to the streaming api, like migration. However, you ended up adding async QMP support to the block layer.
I have the impression this could be generalized a bit more and be moved to QMP instead (and I strongly feel that's what we should do).
There's only one problem: we are waiting for the new QMP server to work on that. I hope to have it integrated by the end of this month, but it might be possible to add async support to the current server if waiting is not an option.
I'm not sure what you'd like here, could you be more specific? I have introduced the concept of a block job - a long-running operation on block devices - and the completion event when the job finishes. I guess you're thinking of a generic QMP job concept with a completion event?
Exactly. We should have QMP_JOB_COMPLETED instead of BLOCK_JOB_COMPLETED, for example.
Or something else?
What I'd like to avoid at this stage is changing the QMP API as seen by clients because we already have at least one client which relies on this API. I know that sucks and that this is my fault because I've been dragging out this patch series for too long. But in a situation like this I think it's better to live with an API that doesn't meet the current philosophy on nice APIs but works flawlessly with both new and existing clients. But it depends on the specifics, what changes do you suggest?
[...]
diff --git a/qapi-schema.json b/qapi-schema.json index fbbdbe0..65308d2 100644 --- a/qapi-schema.json +++ b/qapi-schema.json @@ -901,3 +901,56 @@ # Notes: Do not use this command. ## { 'command': 'cpu', 'data': {'index': 'int'} } + +## +# @block_stream:
Command names should be verbs and please use an hyphen.
These commands have been in the Image Streaming API spec from the beginning:
http://wiki.qemu.org/Features/LiveBlockMigration/ImageStreamingAPI
We cannot prettify the API at this stage. You, Anthony, and I discussed QAPI command naming on IRC maybe two months ago and this seemed to be the way to do it. These commands fit the existing block_* commands and are already in use by libvirt - if we change names now we break libvirt.
[...]
+# +# Since: 1.1 +## +{ 'command': 'block_stream', 'data': { 'device': 'str', '*base': 'str' } } diff --git a/qerror.c b/qerror.c index 14a1d59..605381a 100644 --- a/qerror.c +++ b/qerror.c @@ -174,6 +174,10 @@ static const QErrorStringTable qerror_table[] = { .desc = "No '%(bus)' bus found for device '%(device)'", }, { + .error_fmt = QERR_NOT_SUPPORTED, + .desc = "Not supported",
We have QERR_UNSUPPORTED already.
I know. We're stuck in a hard place here again because NotSupported has been in the Image Streaming API spec and hence implemented in libvirt for a while now. If we change this then an old client which only understands NotSupported will not know what to do with the Unsupported error.
(Unsupported was not in QEMU when the Image Streaming API was defined.)
Let me try to understand it: is libvirt relying on an off tree API and we are now required to have stable guarantees to off tree APIs? I can't even express how insane this looks to me, at least not without being too harsh.
On 01/05/2012 07:16 AM, Luiz Capitulino wrote:
I know. We're stuck in a hard place here again because NotSupported has been in the Image Streaming API spec and hence implemented in libvirt for a while now. If we change this then an old client which only understands NotSupported will not know what to do with the Unsupported error.
(Unsupported was not in QEMU when the Image Streaming API was defined.)
Let me try to understand it: is libvirt relying on an off tree API and we are now required to have stable guarantees to off tree APIs?
No. Libvirt recognizes the off-tree spelling, but does not rely on it - after all, the goal of libvirt is to provide the high level action, using whatever underlying mechanism(s) necessary to get to that action, even if it means using several different attempts until one actually works. If a user has the older libvirt, which only expects the off-tree spelling, then that user's setup will break if they upgrade qemu but not libvirt. But that's not a severe problem - they could have only been relying on the situation if they were using an off-tree build in the first place, so they should be aware that upgrading qemu is a potentially risky scenario, and that they may have to deal with the pieces. Newer libvirt can be easily taught to recognize both the off-tree and stable spellings of the error (checking the stable first, of course, since that will be more likely as the off-tree qemu builds filter out over time). At which point, using either the off-tree qemu or the stable qemu should both work with the newer libvirt. -- Eric Blake eblake@redhat.com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
On 01/05/2012 09:35 AM, Eric Blake wrote:
On 01/05/2012 07:16 AM, Luiz Capitulino wrote:
I know. We're stuck in a hard place here again because NotSupported has been in the Image Streaming API spec and hence implemented in libvirt for a while now. If we change this then an old client which only understands NotSupported will not know what to do with the Unsupported error.
(Unsupported was not in QEMU when the Image Streaming API was defined.)
Let me try to understand it: is libvirt relying on an off tree API and we are now required to have stable guarantees to off tree APIs?
No. Libvirt recognizes the off-tree spelling, but does not rely on it - after all, the goal of libvirt is to provide the high level action, using whatever underlying mechanism(s) necessary to get to that action, even if it means using several different attempts until one actually works.
If a user has the older libvirt, which only expects the off-tree spelling, then that user's setup will break if they upgrade qemu but not libvirt. But that's not a severe problem - they could have only been relying on the situation if they were using an off-tree build in the first place, so they should be aware that upgrading qemu is a potentially risky scenario, and that they may have to deal with the pieces.
Right, this is the difference between ABI compatibility and strict backwards compatibility. To achieve ABI compatibility, we need to not overload BLOCK_JOB_COMPLETED to mean something other than libvirt what expects it to mean. We MUST provide ABI compatibility and SHOULD provide backwards compatibility whenever possible. In this case, I'd suggest that in the very least, we should add BLOCK_JOB_COMPLETED to qapi-schema.json with gen=False set. That way it's codified in the schema to ensure we maintain ABI compatibility. That said, I'm inclined to say that we should just use the BLOCK_JOB_COMPLETED name because I don't think we gain a lot by using QMP_JOB_COMPLETED (not that we shouldn't introduce it, but using it here isn't going to make or break anything). With respect to libvirt relying on interfaces before they exist in QEMU, we need to be a bit flexible here. We want to get better at co-development to help make libvirt support QEMU features as the bleeding edge. Forcing libvirt to wait until a feature is fully baked in QEMU will ensure there's always a feature gap in libvirt which is in none of our best interests. Now that we have gen=False support in qapi-schema.json, we can agree to an API and add it to QEMU before we fully implement it. This gives libvirt something to work off of that they can rely upon. Regards, Anthony Liguori
Newer libvirt can be easily taught to recognize both the off-tree and stable spellings of the error (checking the stable first, of course, since that will be more likely as the off-tree qemu builds filter out over time). At which point, using either the off-tree qemu or the stable qemu should both work with the newer libvirt.
On Thu, 05 Jan 2012 09:56:44 -0600 Anthony Liguori <aliguori@us.ibm.com> wrote:
On 01/05/2012 09:35 AM, Eric Blake wrote:
On 01/05/2012 07:16 AM, Luiz Capitulino wrote:
I know. We're stuck in a hard place here again because NotSupported has been in the Image Streaming API spec and hence implemented in libvirt for a while now. If we change this then an old client which only understands NotSupported will not know what to do with the Unsupported error.
(Unsupported was not in QEMU when the Image Streaming API was defined.)
Let me try to understand it: is libvirt relying on an off tree API and we are now required to have stable guarantees to off tree APIs?
No. Libvirt recognizes the off-tree spelling, but does not rely on it - after all, the goal of libvirt is to provide the high level action, using whatever underlying mechanism(s) necessary to get to that action, even if it means using several different attempts until one actually works.
If a user has the older libvirt, which only expects the off-tree spelling, then that user's setup will break if they upgrade qemu but not libvirt. But that's not a severe problem - they could have only been relying on the situation if they were using an off-tree build in the first place, so they should be aware that upgrading qemu is a potentially risky scenario, and that they may have to deal with the pieces.
Right, this is the difference between ABI compatibility and strict backwards compatibility.
To achieve ABI compatibility, we need to not overload BLOCK_JOB_COMPLETED to mean something other than libvirt what expects it to mean.
We MUST provide ABI compatibility and SHOULD provide backwards compatibility whenever possible.
In this case, I'd suggest that in the very least, we should add BLOCK_JOB_COMPLETED to qapi-schema.json with gen=False set. That way it's codified in the schema to ensure we maintain ABI compatibility.
That said, I'm inclined to say that we should just use the BLOCK_JOB_COMPLETED name because I don't think we gain a lot by using QMP_JOB_COMPLETED (not that we shouldn't introduce it, but using it here isn't going to make or break anything).
What I'm proposing is not just a rename, but adding proper async support to QMP, instead of adding something that is specific to the block layer.
With respect to libvirt relying on interfaces before they exist in QEMU, we need to be a bit flexible here. We want to get better at co-development to help make libvirt support QEMU features as the bleeding edge.
Forcing libvirt to wait until a feature is fully baked in QEMU will ensure there's always a feature gap in libvirt which is in none of our best interests.
We can ask them to wait at least until the API is merged. Most good review and potential problems will only come when the patches are worked on and reviewed on the list.
Now that we have gen=False support in qapi-schema.json, we can agree to an API and add it to QEMU before we fully implement it. This gives libvirt something to work off of that they can rely upon.
Meaning that we can modify the API later? Okay, but in this case we have reasons to modify it before it's merged and I don't see why we shouldn't do it.
Regards,
Anthony Liguori
Newer libvirt can be easily taught to recognize both the off-tree and stable spellings of the error (checking the stable first, of course, since that will be more likely as the off-tree qemu builds filter out over time). At which point, using either the off-tree qemu or the stable qemu should both work with the newer libvirt.
On Thu, Jan 5, 2012 at 8:26 PM, Luiz Capitulino <lcapitulino@redhat.com> wrote:
On Thu, 05 Jan 2012 09:56:44 -0600 Anthony Liguori <aliguori@us.ibm.com> wrote:
On 01/05/2012 09:35 AM, Eric Blake wrote:
On 01/05/2012 07:16 AM, Luiz Capitulino wrote:
I know. We're stuck in a hard place here again because NotSupported has been in the Image Streaming API spec and hence implemented in libvirt for a while now. If we change this then an old client which only understands NotSupported will not know what to do with the Unsupported error.
(Unsupported was not in QEMU when the Image Streaming API was defined.)
Let me try to understand it: is libvirt relying on an off tree API and we are now required to have stable guarantees to off tree APIs?
No. Libvirt recognizes the off-tree spelling, but does not rely on it - after all, the goal of libvirt is to provide the high level action, using whatever underlying mechanism(s) necessary to get to that action, even if it means using several different attempts until one actually works.
If a user has the older libvirt, which only expects the off-tree spelling, then that user's setup will break if they upgrade qemu but not libvirt. But that's not a severe problem - they could have only been relying on the situation if they were using an off-tree build in the first place, so they should be aware that upgrading qemu is a potentially risky scenario, and that they may have to deal with the pieces.
Right, this is the difference between ABI compatibility and strict backwards compatibility.
To achieve ABI compatibility, we need to not overload BLOCK_JOB_COMPLETED to mean something other than libvirt what expects it to mean.
We MUST provide ABI compatibility and SHOULD provide backwards compatibility whenever possible.
In this case, I'd suggest that in the very least, we should add BLOCK_JOB_COMPLETED to qapi-schema.json with gen=False set. That way it's codified in the schema to ensure we maintain ABI compatibility.
That said, I'm inclined to say that we should just use the BLOCK_JOB_COMPLETED name because I don't think we gain a lot by using QMP_JOB_COMPLETED (not that we shouldn't introduce it, but using it here isn't going to make or break anything).
What I'm proposing is not just a rename, but adding proper async support to QMP, instead of adding something that is specific to the block layer.
Proper async support - if you mean the ability to have multiple QMP commands pending at a time - is harder than just fixing QEMU. Clients also need to start taking advantage of it. Clients that do not will be unable to continue when a QMP command takes a long time to complete. I think avoiding long-running QMP commands is a good idea. We have events which can be used to signal completion. It's easy to implement and does not require clients to change the way they think about QMP commands. Today I doubt many QMP clients have implemented multiple pending commands, although the wire protocol allows it.
With respect to libvirt relying on interfaces before they exist in QEMU, we need to be a bit flexible here. We want to get better at co-development to help make libvirt support QEMU features as the bleeding edge.
Forcing libvirt to wait until a feature is fully baked in QEMU will ensure there's always a feature gap in libvirt which is in none of our best interests.
We can ask them to wait at least until the API is merged. Most good review and potential problems will only come when the patches are worked on and reviewed on the list.
The API was agreed on between QEMU and libvirt developers on the mailing lists - you were included in that process. Back in August I sent patches which you saw ("[0/4] Image Streaming API"). I know the API is not what we'd design today when it comes to the cosmetics. We'd want to name things differently, use the Unsupported event which was introduced in the meantime, and maybe make the job completion concept generic. QMP and QAPI have evolved in the time that this feature has been reimplemented. I have tried to keep up with QMP but the API itself is from August. We can't keep redrawing the lines. In summary: * The API was designed and agreed several months ago. * You saw it back then and I've kept you up-to-date along the way. * It predates current QAPI conventions. * Merging it poses no problem, changing the API breaks existing libvirt. I do feel bad that the code has been out of qemu.git for so long and I certainly won't attempt this again in the future. But I really think the pros and cons say we should accept it as an August 2011 API just like many of the other HMP/QMP commands we carry. Regarding being more flexible about working together with libvirt, I do think it's important to work on APIs together. This avoids use developing something purely from the QEMU internal perspective which turns out to be unconsumable by our biggest QMP user :). Stefan
On Fri, 6 Jan 2012 11:06:12 +0000 Stefan Hajnoczi <stefanha@gmail.com> wrote:
On Thu, Jan 5, 2012 at 8:26 PM, Luiz Capitulino <lcapitulino@redhat.com> wrote:
On Thu, 05 Jan 2012 09:56:44 -0600 Anthony Liguori <aliguori@us.ibm.com> wrote:
On 01/05/2012 09:35 AM, Eric Blake wrote:
On 01/05/2012 07:16 AM, Luiz Capitulino wrote:
I know. We're stuck in a hard place here again because NotSupported has been in the Image Streaming API spec and hence implemented in libvirt for a while now. If we change this then an old client which only understands NotSupported will not know what to do with the Unsupported error.
(Unsupported was not in QEMU when the Image Streaming API was defined.)
Let me try to understand it: is libvirt relying on an off tree API and we are now required to have stable guarantees to off tree APIs?
No. Libvirt recognizes the off-tree spelling, but does not rely on it - after all, the goal of libvirt is to provide the high level action, using whatever underlying mechanism(s) necessary to get to that action, even if it means using several different attempts until one actually works.
If a user has the older libvirt, which only expects the off-tree spelling, then that user's setup will break if they upgrade qemu but not libvirt. But that's not a severe problem - they could have only been relying on the situation if they were using an off-tree build in the first place, so they should be aware that upgrading qemu is a potentially risky scenario, and that they may have to deal with the pieces.
Right, this is the difference between ABI compatibility and strict backwards compatibility.
To achieve ABI compatibility, we need to not overload BLOCK_JOB_COMPLETED to mean something other than libvirt what expects it to mean.
We MUST provide ABI compatibility and SHOULD provide backwards compatibility whenever possible.
In this case, I'd suggest that in the very least, we should add BLOCK_JOB_COMPLETED to qapi-schema.json with gen=False set. That way it's codified in the schema to ensure we maintain ABI compatibility.
That said, I'm inclined to say that we should just use the BLOCK_JOB_COMPLETED name because I don't think we gain a lot by using QMP_JOB_COMPLETED (not that we shouldn't introduce it, but using it here isn't going to make or break anything).
What I'm proposing is not just a rename, but adding proper async support to QMP, instead of adding something that is specific to the block layer.
Proper async support - if you mean the ability to have multiple QMP commands pending at a time - is harder than just fixing QEMU. Clients also need to start taking advantage of it. Clients that do not will be unable to continue when a QMP command takes a long time to complete.
They can be fixed if we offer proper async support. Today they can't.
I think avoiding long-running QMP commands is a good idea. We have events which can be used to signal completion. It's easy to implement and does not require clients to change the way they think about QMP commands.
I agree in principle, but in practice we risk having different subsystems and different commands introducing their own async support which is going to make our API (which is already far from perfect) impossible to use, not to mention the maintainability hell that will arise from it. Note that I'm not exactly advocating for heavyweight async support, I just want to avoid keeping messing with this area. Maybe, we could go real simple by having a standard event for asynchronous commands, say ASYNC_CMD_FINISHED or something and that event would contain only the command id and if the command succeeded or failed. The APIs for cancelling and querying would have to be provided by the command itself. I can start a new thread to discuss async support. I haven't done it yet because I don't have a concrete proposal and I also suspect that people are tired of discussing this over and over again.
Today I doubt many QMP clients have implemented multiple pending commands, although the wire protocol allows it.
That's true, but adding the id field in the command dict was silly, as we don't support multiple pending commands.
With respect to libvirt relying on interfaces before they exist in QEMU, we need to be a bit flexible here. We want to get better at co-development to help make libvirt support QEMU features as the bleeding edge.
Forcing libvirt to wait until a feature is fully baked in QEMU will ensure there's always a feature gap in libvirt which is in none of our best interests.
We can ask them to wait at least until the API is merged. Most good review and potential problems will only come when the patches are worked on and reviewed on the list.
The API was agreed on between QEMU and libvirt developers on the mailing lists - you were included in that process. Back in August I sent patches which you saw ("[0/4] Image Streaming API").
I know the API is not what we'd design today when it comes to the cosmetics. We'd want to name things differently, use the Unsupported event which was introduced in the meantime, and maybe make the job completion concept generic.
QMP and QAPI have evolved in the time that this feature has been reimplemented. I have tried to keep up with QMP but the API itself is from August. We can't keep redrawing the lines.
In summary: * The API was designed and agreed several months ago. * You saw it back then and I've kept you up-to-date along the way. * It predates current QAPI conventions. * Merging it poses no problem, changing the API breaks existing libvirt.
It does pose problems. The name changes I've proposed are not minor things, it's about conforming to the protocol which is quite important. Duplicating errors is something that just doesn't make sense either. And most importantly, you're adding async support to the block layer. This means that we'll have two different async APIs when we add one to QMP, or worse other subsystems will be motivated to have their own async APIs too.
I do feel bad that the code has been out of qemu.git for so long and I certainly won't attempt this again in the future. But I really think the pros and cons say we should accept it as an August 2011 API just like many of the other HMP/QMP commands we carry.
I disagree. This should be reviwed and changed as any other submission.
Regarding being more flexible about working together with libvirt, I do think it's important to work on APIs together. This avoids use developing something purely from the QEMU internal perspective which turns out to be unconsumable by our biggest QMP user :).
We do work together with them. I've never ignored their opinion and I'm probably the strongest opinionated when it comes to compatibility. I just can't see how accepting something that is now rotted is going to help either of us.
On 01/06/2012 06:45 AM, Luiz Capitulino wrote:
On Fri, 6 Jan 2012 11:06:12 +0000 Stefan Hajnoczi<stefanha@gmail.com> wrote:
Proper async support - if you mean the ability to have multiple QMP commands pending at a time - is harder than just fixing QEMU. Clients also need to start taking advantage of it. Clients that do not will be unable to continue when a QMP command takes a long time to complete.
They can be fixed if we offer proper async support. Today they can't.
I think avoiding long-running QMP commands is a good idea. We have events which can be used to signal completion. It's easy to implement and does not require clients to change the way they think about QMP commands.
I agree in principle, but in practice we risk having different subsystems and different commands introducing their own async support which is going to make our API (which is already far from perfect) impossible to use, not to mention the maintainability hell that will arise from it.
I absolutely agree with you but practically speaking, we don't have generic async support today. It's been my experience that holding up patch series for generic infrastructure that does exist 1) causes unnecessary angst in contributors 2) puts pressure on the infrastructure to get something in fast vs. get something in that's good. And honestly, it's (2) that I worry the most about. I don't want us to rush async support because we're eager to get block streaming merged. This is why I'm not holding any new devices back while we get QOM merged even if it creates more work for me and introduces new compatibility problems to solve. We also need to look at this interface as a public interface whether we technically committed it to or not. The fact is, an important user is relying upon so that makes it a supported interface. Even though I absolutely hate it, this is why we haven't changed the help output even after all of these years. Not breaking users should be one of our highest priorities. Now we could change this command to make it a better QMP interface and we could do that in a compatible fashion. However, since I think we'll get proper async support really soon and that will involve fundamentally changing this command (along with a bunch of other commands), I don't think there's a lot of value in making cosmetic changes right now. If we're going to break backwards compatibility, I'd rather do it once than twice. What I'd suggest is that we take the command in as-is and we mark it: Since: 1.1 Deprecated: 1.2 See Also: TBD The idea being that we'll introduce new generic async commands in 1.2 and deprecate this command. We can figure out the removal schedule then too. Since this command hasn't been around all that long, we can probably have a short removal schedule. We should also mark the other psuedo-async commands this way too FWIW. Regards, Anthony Liguori
Note that I'm not exactly advocating for heavyweight async support, I just want to avoid keeping messing with this area.
Maybe, we could go real simple by having a standard event for asynchronous commands, say ASYNC_CMD_FINISHED or something and that event would contain only the command id and if the command succeeded or failed. The APIs for cancelling and querying would have to be provided by the command itself.
I can start a new thread to discuss async support. I haven't done it yet because I don't have a concrete proposal and I also suspect that people are tired of discussing this over and over again.
Today I doubt many QMP clients have implemented multiple pending commands, although the wire protocol allows it.
That's true, but adding the id field in the command dict was silly, as we don't support multiple pending commands.
With respect to libvirt relying on interfaces before they exist in QEMU, we need to be a bit flexible here. We want to get better at co-development to help make libvirt support QEMU features as the bleeding edge.
Forcing libvirt to wait until a feature is fully baked in QEMU will ensure there's always a feature gap in libvirt which is in none of our best interests.
We can ask them to wait at least until the API is merged. Most good review and potential problems will only come when the patches are worked on and reviewed on the list.
The API was agreed on between QEMU and libvirt developers on the mailing lists - you were included in that process. Back in August I sent patches which you saw ("[0/4] Image Streaming API").
I know the API is not what we'd design today when it comes to the cosmetics. We'd want to name things differently, use the Unsupported event which was introduced in the meantime, and maybe make the job completion concept generic.
QMP and QAPI have evolved in the time that this feature has been reimplemented. I have tried to keep up with QMP but the API itself is from August. We can't keep redrawing the lines.
In summary: * The API was designed and agreed several months ago. * You saw it back then and I've kept you up-to-date along the way. * It predates current QAPI conventions. * Merging it poses no problem, changing the API breaks existing libvirt.
It does pose problems. The name changes I've proposed are not minor things, it's about conforming to the protocol which is quite important. Duplicating errors is something that just doesn't make sense either.
And most importantly, you're adding async support to the block layer. This means that we'll have two different async APIs when we add one to QMP, or worse other subsystems will be motivated to have their own async APIs too.
I do feel bad that the code has been out of qemu.git for so long and I certainly won't attempt this again in the future. But I really think the pros and cons say we should accept it as an August 2011 API just like many of the other HMP/QMP commands we carry.
I disagree. This should be reviwed and changed as any other submission.
Regarding being more flexible about working together with libvirt, I do think it's important to work on APIs together. This avoids use developing something purely from the QEMU internal perspective which turns out to be unconsumable by our biggest QMP user :).
We do work together with them. I've never ignored their opinion and I'm probably the strongest opinionated when it comes to compatibility.
I just can't see how accepting something that is now rotted is going to help either of us.
On Fri, 06 Jan 2012 09:08:19 -0600 Anthony Liguori <anthony@codemonkey.ws> wrote:
On 01/06/2012 06:45 AM, Luiz Capitulino wrote:
On Fri, 6 Jan 2012 11:06:12 +0000 Stefan Hajnoczi<stefanha@gmail.com> wrote:
Proper async support - if you mean the ability to have multiple QMP commands pending at a time - is harder than just fixing QEMU. Clients also need to start taking advantage of it. Clients that do not will be unable to continue when a QMP command takes a long time to complete.
They can be fixed if we offer proper async support. Today they can't.
I think avoiding long-running QMP commands is a good idea. We have events which can be used to signal completion. It's easy to implement and does not require clients to change the way they think about QMP commands.
I agree in principle, but in practice we risk having different subsystems and different commands introducing their own async support which is going to make our API (which is already far from perfect) impossible to use, not to mention the maintainability hell that will arise from it.
I absolutely agree with you but practically speaking, we don't have generic async support today.
It's been my experience that holding up patch series for generic infrastructure that does exist 1) causes unnecessary angst in contributors 2) puts pressure on the infrastructure to get something in fast vs. get something in that's good.
And honestly, it's (2) that I worry the most about. I don't want us to rush async support because we're eager to get block streaming merged. This is why I'm not holding any new devices back while we get QOM merged even if it creates more work for me and introduces new compatibility problems to solve.
I agree.
We also need to look at this interface as a public interface whether we technically committed it to or not. The fact is, an important user is relying upon so that makes it a supported interface. Even though I absolutely hate it, this is why we haven't changed the help output even after all of these years. Not breaking users should be one of our highest priorities.
One thing I don't understand: how is libvirt relying on it if it doesn't exist in qemu.git yet?
Now we could change this command to make it a better QMP interface and we could do that in a compatible fashion.
However, since I think we'll get proper async support really soon and that will involve fundamentally changing this command (along with a bunch of other commands), I don't think there's a lot of value in making cosmetic changes right now. If we're going to break backwards compatibility, I'd rather do it once than twice.
It goes beyond cosmetic changes. For example, will we allow other async block commands to use this interface? And if we're doing this for block, why not accept something similar for other subsystems if someone happen to submit it? Let me take a non-cosmetic change request example. The BLOCK_JOB_COMPLETED event has a 'error' field. However, it's impossible to know which error happened because the 'error' field contains only the human error description. Another problem: the event is called BLOCK_JOB_COMPLETED, but it's tied to the streaming API. If we allow other commands to use it, they will likely have to add fields there, making the event worse than it already is. There's more, because I skipped this review in v3 as I jumped to the "proper async support" discussion...
What I'd suggest is that we take the command in as-is and we mark it:
Since: 1.1 Deprecated: 1.2 See Also: TBD
The idea being that we'll introduce new generic async commands in 1.2 and deprecate this command. We can figure out the removal schedule then too. Since this command hasn't been around all that long, we can probably have a short removal schedule.
That makes its inclusion even discussable :) A few (very honest) questions: 1. Is it really worth it to have the command for one or two releases? 2. Will we allow other block commands to use this async API? 3. Are we going to accept other ad-hoc async APIs until we have a proper one?
We should also mark the other psuedo-async commands this way too FWIW.
Regards,
Anthony Liguori
Note that I'm not exactly advocating for heavyweight async support, I just want to avoid keeping messing with this area.
Maybe, we could go real simple by having a standard event for asynchronous commands, say ASYNC_CMD_FINISHED or something and that event would contain only the command id and if the command succeeded or failed. The APIs for cancelling and querying would have to be provided by the command itself.
I can start a new thread to discuss async support. I haven't done it yet because I don't have a concrete proposal and I also suspect that people are tired of discussing this over and over again.
Today I doubt many QMP clients have implemented multiple pending commands, although the wire protocol allows it.
That's true, but adding the id field in the command dict was silly, as we don't support multiple pending commands.
With respect to libvirt relying on interfaces before they exist in QEMU, we need to be a bit flexible here. We want to get better at co-development to help make libvirt support QEMU features as the bleeding edge.
Forcing libvirt to wait until a feature is fully baked in QEMU will ensure there's always a feature gap in libvirt which is in none of our best interests.
We can ask them to wait at least until the API is merged. Most good review and potential problems will only come when the patches are worked on and reviewed on the list.
The API was agreed on between QEMU and libvirt developers on the mailing lists - you were included in that process. Back in August I sent patches which you saw ("[0/4] Image Streaming API").
I know the API is not what we'd design today when it comes to the cosmetics. We'd want to name things differently, use the Unsupported event which was introduced in the meantime, and maybe make the job completion concept generic.
QMP and QAPI have evolved in the time that this feature has been reimplemented. I have tried to keep up with QMP but the API itself is from August. We can't keep redrawing the lines.
In summary: * The API was designed and agreed several months ago. * You saw it back then and I've kept you up-to-date along the way. * It predates current QAPI conventions. * Merging it poses no problem, changing the API breaks existing libvirt.
It does pose problems. The name changes I've proposed are not minor things, it's about conforming to the protocol which is quite important. Duplicating errors is something that just doesn't make sense either.
And most importantly, you're adding async support to the block layer. This means that we'll have two different async APIs when we add one to QMP, or worse other subsystems will be motivated to have their own async APIs too.
I do feel bad that the code has been out of qemu.git for so long and I certainly won't attempt this again in the future. But I really think the pros and cons say we should accept it as an August 2011 API just like many of the other HMP/QMP commands we carry.
I disagree. This should be reviwed and changed as any other submission.
Regarding being more flexible about working together with libvirt, I do think it's important to work on APIs together. This avoids use developing something purely from the QEMU internal perspective which turns out to be unconsumable by our biggest QMP user :).
We do work together with them. I've never ignored their opinion and I'm probably the strongest opinionated when it comes to compatibility.
I just can't see how accepting something that is now rotted is going to help either of us.
On 01/06/2012 01:42 PM, Luiz Capitulino wrote:
On Fri, 06 Jan 2012 09:08:19 -0600
We also need to look at this interface as a public interface whether we technically committed it to or not. The fact is, an important user is relying upon so that makes it a supported interface. Even though I absolutely hate it, this is why we haven't changed the help output even after all of these years. Not breaking users should be one of our highest priorities.
One thing I don't understand: how is libvirt relying on it if it doesn't exist in qemu.git yet?
Because there was a discussion on qemu-devel and we agreed on an interface that both sides would implement to. I expect this to happen more often in the future too.
Now we could change this command to make it a better QMP interface and we could do that in a compatible fashion.
However, since I think we'll get proper async support really soon and that will involve fundamentally changing this command (along with a bunch of other commands), I don't think there's a lot of value in making cosmetic changes right now. If we're going to break backwards compatibility, I'd rather do it once than twice.
It goes beyond cosmetic changes. For example, will we allow other async block commands to use this interface? And if we're doing this for block, why not accept something similar for other subsystems if someone happen to submit it?
Yes, I'm in agreement with you on direction. But adding proper async support is not a simple task, and blocking this interface based on that is a bad idea for reasons previously mentioned.
Let me take a non-cosmetic change request example. The BLOCK_JOB_COMPLETED event has a 'error' field. However, it's impossible to know which error happened because the 'error' field contains only the human error description.
Ok. That's worth thinking about.
Another problem: the event is called BLOCK_JOB_COMPLETED, but it's tied to the streaming API. If we allow other commands to use it, they will likely have to add fields there, making the event worse than it already is.
But aren't we going to introduce a proper async interface? This is what has me confused.
There's more, because I skipped this review in v3 as I jumped to the "proper async support" discussion...
Well let's do proper async support. As I mentioned, I'd rather take this command in as-is, introduce proper async support, and then deprecate a bunch of stuff at once.
What I'd suggest is that we take the command in as-is and we mark it:
Since: 1.1 Deprecated: 1.2 See Also: TBD
The idea being that we'll introduce new generic async commands in 1.2 and deprecate this command. We can figure out the removal schedule then too. Since this command hasn't been around all that long, we can probably have a short removal schedule.
That makes its inclusion even discussable :) A few (very honest) questions:
1. Is it really worth it to have the command for one or two releases?
Yes. The most important consideration IMHO is parallelizing development. We want the block layer to evolve to the greatest extent possible independent of our other subsystems. If we don't have the right infrastructure in QMP to support a block feature, that shouldn't hold up progress in the block layer to the greatest extent possible.
2. Will we allow other block commands to use this async API?
Depends on how long it takes to do "proper async support".
3. Are we going to accept other ad-hoc async APIs until we have a proper one?
Yes. So let's get serious about getting a proper interface in :-) Regards, Anthony Liguori
On Tue, 10 Jan 2012 13:18:41 -0600 Anthony Liguori <anthony@codemonkey.ws> wrote:
On 01/06/2012 01:42 PM, Luiz Capitulino wrote:
On Fri, 06 Jan 2012 09:08:19 -0600
We also need to look at this interface as a public interface whether we technically committed it to or not. The fact is, an important user is relying upon so that makes it a supported interface. Even though I absolutely hate it, this is why we haven't changed the help output even after all of these years. Not breaking users should be one of our highest priorities.
One thing I don't understand: how is libvirt relying on it if it doesn't exist in qemu.git yet?
Because there was a discussion on qemu-devel and we agreed on an interface that both sides would implement to.
I expect this to happen more often in the future too.
For future commands we either, implement it right away or ask libvirt to wait for the command to be merged, or at least pass initial review.
Now we could change this command to make it a better QMP interface and we could do that in a compatible fashion.
However, since I think we'll get proper async support really soon and that will involve fundamentally changing this command (along with a bunch of other commands), I don't think there's a lot of value in making cosmetic changes right now. If we're going to break backwards compatibility, I'd rather do it once than twice.
It goes beyond cosmetic changes. For example, will we allow other async block commands to use this interface? And if we're doing this for block, why not accept something similar for other subsystems if someone happen to submit it?
Yes, I'm in agreement with you on direction. But adding proper async support is not a simple task, and blocking this interface based on that is a bad idea for reasons previously mentioned.
Let me take a non-cosmetic change request example. The BLOCK_JOB_COMPLETED event has a 'error' field. However, it's impossible to know which error happened because the 'error' field contains only the human error description.
Ok. That's worth thinking about.
Another problem: the event is called BLOCK_JOB_COMPLETED, but it's tied to the streaming API. If we allow other commands to use it, they will likely have to add fields there, making the event worse than it already is.
But aren't we going to introduce a proper async interface? This is what has me confused.
Yes, I was thinking about new block commands using this API before we get proper async support...
There's more, because I skipped this review in v3 as I jumped to the "proper async support" discussion...
Well let's do proper async support. As I mentioned, I'd rather take this command in as-is, introduce proper async support, and then deprecate a bunch of stuff at once.
Ok, I'm willing to do this as Stefan has agreed on deprecating this as soon as we get proper async support.
What I'd suggest is that we take the command in as-is and we mark it:
Since: 1.1 Deprecated: 1.2 See Also: TBD
The idea being that we'll introduce new generic async commands in 1.2 and deprecate this command. We can figure out the removal schedule then too. Since this command hasn't been around all that long, we can probably have a short removal schedule.
That makes its inclusion even discussable :) A few (very honest) questions:
1. Is it really worth it to have the command for one or two releases?
Yes. The most important consideration IMHO is parallelizing development. We want the block layer to evolve to the greatest extent possible independent of our other subsystems. If we don't have the right infrastructure in QMP to support a block feature, that shouldn't hold up progress in the block layer to the greatest extent possible.
2. Will we allow other block commands to use this async API?
Depends on how long it takes to do "proper async support".
3. Are we going to accept other ad-hoc async APIs until we have a proper one?
Yes. So let's get serious about getting a proper interface in :-)
ACK
Regards,
Anthony Liguori
On 01/10/2012 02:55 PM, Luiz Capitulino wrote:
On Tue, 10 Jan 2012 13:18:41 -0600 Anthony Liguori<anthony@codemonkey.ws> wrote:
On 01/06/2012 01:42 PM, Luiz Capitulino wrote:
On Fri, 06 Jan 2012 09:08:19 -0600
We also need to look at this interface as a public interface whether we technically committed it to or not. The fact is, an important user is relying upon so that makes it a supported interface. Even though I absolutely hate it, this is why we haven't changed the help output even after all of these years. Not breaking users should be one of our highest priorities.
One thing I don't understand: how is libvirt relying on it if it doesn't exist in qemu.git yet?
Because there was a discussion on qemu-devel and we agreed on an interface that both sides would implement to.
I expect this to happen more often in the future too.
For future commands we either, implement it right away or ask libvirt to wait for the command to be merged, or at least pass initial review.
Or commit the schema entry to qapi-schema.json with gen=False. But when this command was first proposed, qapi-schema.json didn't exist in the tree :-)
But aren't we going to introduce a proper async interface? This is what has me confused.
Yes, I was thinking about new block commands using this API before we get proper async support...
Well let's avoid that problem by doing proper async support before we get new block commands ;-)
There's more, because I skipped this review in v3 as I jumped to the "proper async support" discussion...
Well let's do proper async support. As I mentioned, I'd rather take this command in as-is, introduce proper async support, and then deprecate a bunch of stuff at once.
Ok, I'm willing to do this as Stefan has agreed on deprecating this as soon as we get proper async support.
Excellent. BTW, you mentioned that you were going to send an RFC for proper async support? Regards, Anthony Liguori
What I'd suggest is that we take the command in as-is and we mark it:
Since: 1.1 Deprecated: 1.2 See Also: TBD
The idea being that we'll introduce new generic async commands in 1.2 and deprecate this command. We can figure out the removal schedule then too. Since this command hasn't been around all that long, we can probably have a short removal schedule.
That makes its inclusion even discussable :) A few (very honest) questions:
1. Is it really worth it to have the command for one or two releases?
Yes. The most important consideration IMHO is parallelizing development. We want the block layer to evolve to the greatest extent possible independent of our other subsystems. If we don't have the right infrastructure in QMP to support a block feature, that shouldn't hold up progress in the block layer to the greatest extent possible.
2. Will we allow other block commands to use this async API?
Depends on how long it takes to do "proper async support".
3. Are we going to accept other ad-hoc async APIs until we have a proper one?
Yes. So let's get serious about getting a proper interface in :-)
ACK
Regards,
Anthony Liguori
On Tue, 10 Jan 2012 15:02:48 -0600 Anthony Liguori <aliguori@us.ibm.com> wrote:
On 01/10/2012 02:55 PM, Luiz Capitulino wrote:
On Tue, 10 Jan 2012 13:18:41 -0600 Anthony Liguori<anthony@codemonkey.ws> wrote:
On 01/06/2012 01:42 PM, Luiz Capitulino wrote:
On Fri, 06 Jan 2012 09:08:19 -0600
We also need to look at this interface as a public interface whether we technically committed it to or not. The fact is, an important user is relying upon so that makes it a supported interface. Even though I absolutely hate it, this is why we haven't changed the help output even after all of these years. Not breaking users should be one of our highest priorities.
One thing I don't understand: how is libvirt relying on it if it doesn't exist in qemu.git yet?
Because there was a discussion on qemu-devel and we agreed on an interface that both sides would implement to.
I expect this to happen more often in the future too.
For future commands we either, implement it right away or ask libvirt to wait for the command to be merged, or at least pass initial review.
Or commit the schema entry to qapi-schema.json with gen=False.
But when this command was first proposed, qapi-schema.json didn't exist in the tree :-)
But aren't we going to introduce a proper async interface? This is what has me confused.
Yes, I was thinking about new block commands using this API before we get proper async support...
Well let's avoid that problem by doing proper async support before we get new block commands ;-)
There's more, because I skipped this review in v3 as I jumped to the "proper async support" discussion...
Well let's do proper async support. As I mentioned, I'd rather take this command in as-is, introduce proper async support, and then deprecate a bunch of stuff at once.
Ok, I'm willing to do this as Stefan has agreed on deprecating this as soon as we get proper async support.
Excellent.
BTW, you mentioned that you were going to send an RFC for proper async support?
It's just a few proposals for the high-level API (ie. no patches), I can send it tomorrow.
Regards,
Anthony Liguori
What I'd suggest is that we take the command in as-is and we mark it:
Since: 1.1 Deprecated: 1.2 See Also: TBD
The idea being that we'll introduce new generic async commands in 1.2 and deprecate this command. We can figure out the removal schedule then too. Since this command hasn't been around all that long, we can probably have a short removal schedule.
That makes its inclusion even discussable :) A few (very honest) questions:
1. Is it really worth it to have the command for one or two releases?
Yes. The most important consideration IMHO is parallelizing development. We want the block layer to evolve to the greatest extent possible independent of our other subsystems. If we don't have the right infrastructure in QMP to support a block feature, that shouldn't hold up progress in the block layer to the greatest extent possible.
2. Will we allow other block commands to use this async API?
Depends on how long it takes to do "proper async support".
3. Are we going to accept other ad-hoc async APIs until we have a proper one?
Yes. So let's get serious about getting a proper interface in :-)
ACK
Regards,
Anthony Liguori
participants (5)
-
Anthony Liguori -
Anthony Liguori -
Eric Blake -
Luiz Capitulino -
Stefan Hajnoczi