Flow

[dworkin]

Hydra can run tasks in parallel, as long as they don't interfere with each other. How can tasks avoid such interference?

In its most simplified version, this can be accomplished when a task modifies no files, and no other object than the one it starts in. Other objects are not modified directly, but by starting a call_out or continuation in them (recall that starting a call_out in an object without accessing data in that object does not count as a modification). When all currently running tasks follow this rule, it is guaranteed that all of them can succeed.

The Flow framework remodels I/O using this paradigm, and HTTP handling has also adopted it.