Details
-
Sub-task
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
Description
Problem: In a look aside cache pattern when the cache is down the database will move ahead and become more accurate, up-to-date, while the cache will become stale. Users would like recovery to include automatically invalidating or destroying (this can be configured) stale keys in the cache.
Some ideas here were to use a client cache to hold updated keys while the cluster is coming back online. OR have application update a DB table with updated keys... and have recovery pull those keys to invalidate / destroy.
Do not allow writes / reads before the data is being invalidated / destroyed from the list of keys above.
Have a cluster setting for "Recovering" and "recovered" to know when to open up the pool.