@wuhuizuo would matrix work for your use case?

@wuhuizuo would matrix work for your use case?

It doesn't as it doesn't "apply" to PipelineRun and in that particular case, taskRunTemplate/taskRunspecs (because this is where we can set topology constraint, …).

@wuhuizuo would matrix work for your use case?

No, it can not. That's I want: set arch in params or matrix to make parallel tasks on different arch nodes.

@lbernick @jerop this is a "shortcoming" of the matrix work that is making it a bit less useful than I thought it would to be honest 🙃

@vdemeester would matrix at the pipeline level be more useful?

@jerop I don't think it would solve too much (or even what it would look like — except if it's a typo and you meant pipelinerun 😛). The main trick is, when you want to use matrix for some runtime information. The basic example that comes to mind is for multi-arch build on nodes with different architecture (on a heterogenous cluster). Today, we cannot do this with matrix, because those are runtime information, that cannot benefit from the matrix feature I believe.

yes, I meant pipelinerun 🙃

pipeline declares string params, and pipelinerun provides array params to fan out to multiple child pipelineruns

seems to me that will help with the multi-arch use case, right?

seems to me that will help with the multi-arch use case, right?

It helps if we use cross-compilation to do multi-arch build, but it doesn't if we need the "host" to be a certain architecture (as of today I mean).

seems to me that will help with the multi-arch use case, right?

It helps if we use cross-compilation to do multi-arch build, but it doesn't if we need the "host" to be a certain architecture (as of today I mean).

@vdemeester
Sorry for late reply, not just for building, may be some tests(unit tests or integration tests ...) need run on different-arch nodes.

I would like to implement it. ❤️

/cc wuhuizuo

/assign wuhuizuo

@wuhuizuo Any update on your efforts to implement this feature?

Any update ?
In our case, we have changed the parallel matrix build tasks for amd64 and arm64 to two independent tasks.

have no time to finish the feature.

/unassign

/assign dorzel

Currently working on this in the linked PR, appreciate any feedback/suggestions