Improving the scheduler (GSoC task)
Software Heritage periodically crawls over 150 millions software origins. This is a challenge as our loading infrastructure is not infinite, and loading VCS history is a somewhat computationally expensive task.
The swh-scheduler component (see: code and doc) is in charge of scheduling loading tasks regularly, and tries to be smart about the intervals, to avoid unnecessary work. For example, by visiting active origins more often than inactive ones.
You will improve the task scheduler to make smarter choices about what origins to visit and when.
We recently redesigned the scheduler to allow "pluggable" scheduling policies, so you can write new ones and see how they compare to the simple ones we already wrote.
We also wrote a simulator to try out the scheduler with tasks on virtual origins, and you can also work on making it more realistic, as it would be very useful to see how each policy performs
- Python 3 and Git are a must to work on any Software Heritage project
- PostgreSQL, as the scheduler heavily relies on it to be efficient
- Experience with scientific simulation is a plus, but not required