Details
-
Improvement
-
Status: Open
-
Major
-
Resolution: Unresolved
-
0.8.0
-
None
-
None
Description
[Documenting my discussion with Jacob]
Lack of resources (eg: memory) might lead to a particular container lagging behind. In the current implementation, this leads to the container dying and restarting with the same config.
A better strategy would be to split the work into additional containers (maybe) and then partition the job further. This is definitely possible for stateless jobs. However, whether the same can be accomplished for stateful jobs - needs to be investigated.
Attachments
Issue Links
- relates to
-
SAMZA-334 Need for asymmetric container config
- Open