Github user dhutchis commented on the pull request:

    https://github.com/apache/accumulo/pull/96#issuecomment-215535249
  
    > An alternative solution to introducing SyncingTabletLocator could be to 
change the implementation of clearLocators to the following. This clears all of 
the caches while still maintaining the references to in uses locators.
    > 
    >      public static synchronized void clearLocators() {
    >         for(TabletLocator tl : locators.values()) {
    >           tl.invalidateCache();
    >         }
    >      }
    > If worried about empty tablet location caches being left around in the 
locators map, that could possible be changed to a weak hash map.
    
    You could be on to something @keith-turner if we give it more thought.  The 
original reason why the `locators` cache is cleared every hour is to remove 
TabletLocators that refer to tablets that don't exist or are rarely accessed, 
ya?
    
    A WeakHashMap wouldn't work because it uses weak references for the keys. 
This would mean that the gc could reclaim keys as soon as the `getLocator` 
method returns, which is inefficient for a cache and has correctness issues if 
the static `invalidate` methods are called.
    
    What do you think of switching locators to a [Guava 
cache](https://github.com/google/guava/wiki/CachesExplained) with a 1 hour 
timeout on entries since their last access?
    
    I think this would let us remove the `synchronized` from `getLocator` 
because Guava caches are thread-safe and designed for concurrency.  It would 
eliminate the need for any `clearLocators` call (and `SyncingTabletLocator`).  


---
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 [email protected] or file a JIRA ticket
with INFRA.
---

Reply via email to