Agree about separation. I think we need to define some internal
permutations that keep coming up with bugs. I can start the list here:
1. Serializable, Externalizable, neither.
1. We should run the whole suite 3 times, once for each serialization
mode. Having 2 or 3 nodes in the cluster here should be enough.
No need to
test it on the changing cluster.
2. On-heap and off-heap with entry processor and peer-deployment
1. on-heap, entry-processor, peer-deployment=true/false
2. off-heap, entry-processor, peer-deployment=true/false
3. On-heap and off-heap with eviction policies
1. eviction policy, memory-mode=on-heap
2. eviction policy, memory-mode=off-heap
3. eviction policy, memory-mode=off-heap-values
4. On-heap and off-heap with continuous queries and
peer-deployment=true/false
5. …
I think we can come up with about 20 most important permutations. Thoughts?
D.
On Wed, Feb 10, 2016 at 9:24 AM, Alexey Goncharuk <
[email protected]> wrote:
> Dmitriy, the size of the circular buffer can be decreased by setting
> IGNITE_ATOMIC_CACHE_DELETE_HISTORY_SIZE system property.
>
> I was wondering what our next steps will be after implementing the
> framework. From the initial discussion I thought we want to convert some
> older tests to this new framework and run all new tests using this
> framework. However, from what Artem already has written, this sounds
> unrealistic because adding even a simple test case will induce hours of run
> time. I think we still need to separate more important and less important
> configurations, run important ones on a daily basis, and use all others for
> after-code-freeze runs, for example.
>
> Thoughts?
>
