Github user JoshRosen commented on the pull request:
https://github.com/apache/spark/pull/1106#issuecomment-54674029
Yeah, I suppose so, but there was one corner-case that I was concerned
about (that is addressed by treating it as a circular buffer):
Let's say we have a scenario where we are trying to schedule three drivers
but there are only two workers that they can run on, and let's also say that
there's initially enough capacity to run all three drivers. In that case, I
think we would pop the head of `shuffleWorkers` until it becomes empty and not
schedule the third driver. This other driver would get scheduled on a
subsequent call to `Master.schedule()`, but I guess that only happens when new
apps join or when resource availability changes, so we might wait longer than
is necessary to launch this driver.
The circular buffer brings its own problems, though: let's say that there
are _no_ valid locations where we can schedule the driver. In this case, we
should stop looping through the buffer so that we don't go into an infinite
loop.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
