Julian Samaroo
Julian Samaroo
Can you try running the Julia profiler with Julia master on this, at least twice before the slowdown and twice after? We should see a noticeable slowdown during some part...
@shashi is the code for that special scheduler freely available?
This is something we should have somewhere, but not necessarily in Dagger's core, since we may want more "supervisory actions" than just retries and delay-based retry. For example, we might...
For checkpointing to work with the eager API, we'll probably need to start calculating hashes of each task's function and arguments (and recursively hashing any upstream tasks, while weakly caching...
Additionally, we should probably also setup TLS variables for the associated thunk when calling thunk checkpoint/restore. This would let us access the upcoming `storage` option, so that the callback can...
Hash implementation added in #356
We should be able to wrap the call to `compute` in a try-catch, and also forward any exceptions that occur in async calls like `async_apply` and any try-catch blocks that...
As @kshyatt pointed out, we'll also want the ability for thunks to cancel other thunks (supported by #117 ). e.g. optimization problems could use Dagger to schedule a number of...
Yeah, it's not only-one semantics, but first-to-finish semantics. `waitfirst` sounds good! > are these tasks that are both local and remote? They can be executing anywhere, but we have a...
I think that `waitfirst` should probably return the handle to the finished task, and `fetchfirst` will just return the value of the finished task.