~rogpeppe/juju-core/azure : changes

~rogpeppe/juju-core/azure » Changes from revision 1464.1.1

From Revision 1464.1.1 to 1446

Rev	Summary	Authors	Date
1464.1.1	Add missing argument to gwacl call. Add missing argument to gwacl call.	Gavin Panella	11 years ago
1464	[r=jtv] Skip the service-label dance in Azure. [r=jtv] Skip the service-label dance in Azure. We're still working out how we're supposed to determine an instance's DNS name in Azure. We may have been chasing a red herring early on: on the one hand the documentation hinted that a service's name also formed the hostname part of its DNS name, but on the other, we were getting what looked like a randomly generated DNS name assigned by Azure. And a little dance was orchestrated around that: get the instance URL from the Deployment object, store it in the Service's label field, use the label field as the source of the DNS name. As we understand it now, the weird randomly assigned hostnames only happen in Staging deployments, and the instance URLs don't even resolve when we deploy to Production (as we must do in Juju). And so this branch does away with the little dance, uses the service name as the instance's DNS name, and rips out all code that I could find that was only there to support the dance. The service label is now a mere courtesy; it's required but we may as well set it to be identical to the service name.	jtv at canonical	11 years ago
1463	[r=jtv] Change how an Azure instance finds its ID. [r=jtv] Change how an Azure instance finds its ID. This can now come from the service name it digs out of the WALA config, rather than from the deployment ID. This means that the deployment ID no longer matters, and that in turn opens the door to multiple instances per hosted service.	jtv at canonical	11 years ago
1462	[r=rvb] Azure provider: implement more methods [r=rvb] Azure provider: implement more methods This branch does two things: - it replaces the panic() calls in the port-related methods with empty methods. The goal here is to be able to test the provider more easily, provided that you open/close the ports manually. - it updates the name of the image used because an image with support for Azure in cloud-init has just been made available. In case you wonder, the change s/ExtraSmall/Small is there to accomodate installing mysql on a node. ExtraSmall does not have enough memory. This is all for the sake of testing, all of the code that this branch changes will be re-worked later.	Raphael Badin	11 years ago
1461	[r=jtv] Constructor for cloudinit.MachineConfig. [r=jtv] Constructor for cloudinit.MachineConfig. This is preparation for a refactoring discussed with William Reade and outlined in bug 1199847. Much of the work of creating a cloudinit.MachineConfig is repetitive between providers, so we might as well extract it. This also makes it easier to move the construction from one function to another, as is likely to happen during the further refactoring. You'll see that "tools" is not a parameter. That's because the projected refactoring moves construction of the MachineConfig ahead of where the tools are selected. If these two steps end up fitting together, uniformly across providers, once the code starts to stabilize again then we can make "tools" a parameter after all.	jtv at canonical	11 years ago
1460	[r=wallyworld] New assignment policy is AssignCleanEmpty [r=wallyworld] New assignment policy is AssignCleanEmpty Add container support to AssignClean(Empty) policies and make the default policy AssignCleanEmpty. This allows manual deployment of charms into existing clean and empty containers/instances, with the fallback that a new container/instance is created if none are available. The system behaves the same as previously if a simple bootstrap and deploy is performed, but if add-machine is used to create some containers/instances, then those are no longer ignored. https://codereview.appspot.com/11019044/	Ian Booth	11 years ago
1459	[r=jtv] Remove unneeded internalStartInstance parameter. [r=jtv] Remove unneeded internalStartInstance parameter. This matters because it tightens up the signature of a function that is doing the advance reconnaissance for an ongoing refactoring in the StartInstance() and Bootstrap() methods.	jtv at canonical	11 years ago
1458	[r=jtv] Pointer receives for maas/azure tests. [r=jtv] Pointer receives for maas/azure tests. We've been told that gocheck may get confused when it sees test methods with non-pointer receivers, and fail to run some. So here's a branch to fix up the receivers for all pointers in the maas and azure providers. Those are the only providers I'm directly involved with; I didn't touch the others but I could if that's appreciated.	jtv at canonical	11 years ago
1457	[r=jtv] Unify providers' tools series check. [r=jtv] Unify providers' tools series check. This is preparation for a refactoring discussed with William Reade and outlined in bug 1199847. When starting an instance, the providers generally check that the proffered possibleTools are all for the required OS release series. (Well... except the dummy, which doesn't look like it cares, and the local provider where I added this because it seemed just as appropriate as for the "real" ones). In this branch I extract that check into a reusable function, with tests. With this in place it'll be easier to move these checks around, and as you can see in the maas/azure providers, it's no longer even necessary to pass a series down the call chain when starting an instance — it can be just a property of the selected tools. There was no documented way to generate fakes with different series, or indeed documented anything, so I hacked up my own very simple fakes. It looked as if it might have been possible to mess with what was there already, but the problem with such data-driven tests is that extending them in unexpected ways often leads to brittleness, poorly-understood tests, rigidity, and redundant or conflated coverage. I decided against trying to fit my tests into the code that was already in the test file.	jtv at canonical	11 years ago
1456	[r=jtv] Unify patching of AttemptStrategies. [r=jtv] Unify patching of AttemptStrategies. This came out of review for other work: when testing providers, it makes sense to patch up AttemptStrategy configurations so as to minimize the time that might be spent sleeping during tests. But the patching worked in different styles. This branch sweeps them together into a single function, with a single post-test cleanup style.	jtv at canonical	11 years ago
1455	[r=rvb] Azure provider: change deployment name. [r=rvb] Azure provider: change deployment name. The deployment should be named after the hosted service. The deployment name is the only information available on the instance itself (it's in the Windows Azure Agent configuration file). We need the value returned by environprovider.InstanceId() (extracted from the WAA config file) to match the instance id (as in, the juju instance id) and thus the deployment name should be the same as the hosted service name. This is something we had in mind right from the start but forgot when we actually got to implement the provider. I have tested that with this fix (and a couple of other fixes we have in flight right now), we can successfully bootstrap an environment and get a response from "juju status" (with only the bootstrap node).	Raphael Badin	11 years ago
1454	[r=jtv] Name "internal StartInstance()" methods uniformly. [r=jtv] Name "internal StartInstance()" methods uniformly. This is preparation for a refactoring discussed with William Reade and outlined in bug 1199847. All of the real-world providers (that is, all current providers except the dummy and local ones) have the same grouping of methods: an “internal StartInstance()” to start an instance, and the Bootstrap() and StartInstance() methods from the Environ interface which both call that internal method. Here, I'm giving each of the “internal StartInstance()” methods the same name to reflect the common structure. Regularizing this constellation between providers will allow us to factor out lots of commonality across the providers' code, from all three of those methods, and simplify their parameter-passing. The branch you see here does not regularize the parameter lists yet, but gets a small piece of the job behind us. The rest of the job will be easier without the confusion of different names for essentially the same method.	jtv at canonical	11 years ago
1453	[r=jtv] Consistently use shortAttempt in MAAS provider. [r=jtv] Consistently use shortAttempt in MAAS provider. We had a few utils.AttemptStrategy instances as local variables in the MAAS provider, albeit with the same timings as the provider's global shortAttempt. And we weren't shortcutting it in the tests, so longAttempt would time out very quickly in tests but shortAttempt could block for 5 seconds at a time. Not saying that it did, but it could. This branch fixes that, and as a drive-by, replaces the obfuscated test timings in the EC2 test fixtures with human-readable ones. I didn't bother figuring out exactly what the original timings were, and used timings we had in other cases. If we depended on the original timings, we were probably hiding a bug anyway. And the original timings were written as if to make it deliberately hard to figure out what they were meant to be. Floating-point fractionals with exponent notation as a way to write an integral nanoseconds, without any reminders of the meaning? “If normal engineers can maintain my code, I'm not elite enough.” ☺	jtv at canonical	11 years ago
1452	[r=jtv] Shortcut polling delays in Azure provider tests. [r=jtv] Shortcut polling delays in Azure provider tests. Provider tests normally override the provider's shortAttempt, as well as environs.LongAttempt, to avoid undesirable sleeping in the test suite. We hadn't done that for the Azure provider yet, so here you go.	jtv at canonical	11 years ago
1451	[r=frankban] Implemented service.EnsureMinUnits(). [r=frankban] Implemented service.EnsureMinUnits().	Francesco Banconi	11 years ago
1450	[r=rvb] Azure provider: Fix environprovider.Open() [r=rvb] Azure provider: Fix environprovider.Open() Right now, when Open() fails because NewEnviron() fails, it returns non-nil interface object (this is the ugly "nil-that-s-not-nil" problem http://golang.org/doc/faq#nil_error). environprovider.Open() must return an empty object in case of failure.	Raphael Badin	11 years ago
1449	[r=jtv] Use environment prefix for Azure hosted services. [r=jtv] Use environment prefix for Azure hosted services. The Azure provider uses a naming prefix to set one environment's instances apart from another that might be living on the same Azure account. Each instance lives in its own "hosted service" (Azure term) and each hosted service must have a unique name. In this branch, I include that prefix when creating a new hosted service for a new instance.	jtv at canonical	11 years ago
1448	[r=jtv] Open state/api ports on new instance. [r=jtv] Open state/api ports on new instance. Besides the usual ssh port, the bootstrap instance also needs the state (mongo) and API ports open. It seems a bit strange to open these on all new instances, but Raphaël points out that the EC2 provider does the same thing. It sounds like tech debt, so I'm adding some TODO comments.	jtv at canonical	11 years ago
1447	[r=jtv] Implement Azure's Environ.Provider(). [r=jtv] Implement Azure's Environ.Provider().	jtv at canonical	11 years ago
1446	[r=jtv] Make "/var/lib/juju" path a variable. [r=jtv] Make "/var/lib/juju" path a variable. That particular path, for cloudinit's benefit, is hard-coded all over the provider code. It may be an obstacle to testing, because tests can't substitute a fake (and obviously tests shouldn't mess with /var/lib/juju, or even rely on its existence). If we later find that one of these mentions is different from the others, it's quite possible that we have to specialize some of the uses into separate variables. For now, the main things are to be aware, and to have an injection point for tests.	jtv at canonical	11 years ago

Older »