TOSCA says some things about how imports happen, but not that match. I think we are aligned with the spec here.
By the way, I'm quite proud of the implementation in ARIA: we handle imports in a threadpool, so that if you are importing from many http repositories at the same time, each download will happen simultaneously. Performance is as optimal as it can be. (We are also caching the results according to the conditional http request standard.) On Mon, Oct 23, 2017 at 12:06 PM, Thomas Nadeau <[email protected]> wrote: > > Just for my own understanding, is the algorithm you describe used > based on what TOSCA specifies, > or is this our implementation detail? I ask because you can imagine > process done > one at a time, i.e.: per-file, but the processed results can of course, be > different. > > —Tom > > > > > On Oct 23, 2017, at 12:46 PM, Tal Liron <[email protected]> wrote: > > > > Yes, you can do that. Think of imports as being essentially like a smart > > copy-paste from one YAML file to another. In the end, ARIA treats it as > > just one big service template. > > > > In other words, the requirements-and-capabilities matching occurs only > > after all imports were done and the final service template is valid. > > > > On Mon, Oct 23, 2017 at 10:45 AM, Steve Baillargeon < > > [email protected]> wrote: > > > >> Thank you Tal > >> > >> Can ARIA handle the case when ST1 imports ST2 with a relationship (link) > >> between source node template AA defined in ST1 and target node template > BB > >> defined in ST2? > >> For instance, a SoftwareComponent node is defined in ST1 with a specific > >> requirement to run on a Compute node that is defined in ST2. > >> > >> -Steve > >> > >> -----Original Message----- > >> From: Tal Liron [mailto:[email protected]] > >> Sent: Thursday, October 19, 2017 2:18 PM > >> To: [email protected] > >> Subject: Re: Note template names > >> > >> Some good questions here. Let me try to break it down, > >> > >> On Thu, Oct 19, 2017 at 12:39 PM, Steve Baillargeon < > >> [email protected]> wrote: > >> > >>> Node template names must be unique within a given service template. > >>> > >> > >> Yes. Actually, there is an "interesting" issue with this: in YAML, a > dict > >> could have multiple keys with the same name. What happens in that case > is > >> that subsequent keys (node template names in this case) override > previous > >> ones of the same name. So, you can syntactically have two node > templates of > >> the same name, but only one will count. > >> > >> We tried to find ways in ARIA to emit an error in this case, but > actually > >> it's very hard to do because it's a YAML spec issue that allows it. > >> > >> > >>> What about the case when ST1 imports ST2 as follows. > >>> ST1 has 2 node templates: template A derived from node type X and > >>> template B derived from node type Y. > >>> ST2 has 1 node template: template A derived from node type Z . > >>> > >> > >> Actually, the "imports" keyword isn't 100% clear in the TOSCA spec. What > >> we do in ARIA is merge the dicts, and again identical keys (node > template > >> names) will override without error. So, the final behavior will be > >> undefined. > >> > >> However, I think that in this particular case ARIA can do something > more, > >> because we handle merging ourselves (it's not an issue of the YAML > spec). I > >> have opened a new JIRA about this: > >> https://issues.apache.org/jira/browse/ARIA-390 > >> > >
