batch.groupBy fails all groups in case of error in one

[georgi-lozev]

With the current implementation of groupBy in abacus-batch event if there is is an error in a call in one of the groups, all groups response with that same error.

Imagine the following situation. Two clients post their usage to the collector in parallel. The first one with valid usage and the second one with out of slack window. The requests are received in batch in the accumulator. It starts to groupReduce them and the second one which is not in the slack throws. This will result in reporting the same error to both clients, despite the fact the usage doc of the first client has been processed successfully. This could also happen for all kind of throw errors that are not handled in the dataflow reducers, not only for out of slack.

[cf-gitbot]

We have created an issue in Pivotal Tracker to manage this:

https://www.pivotaltracker.com/story/show/153771565

The labels on this github issue will be updated when the story is started.