Details
-
Type:
Bug
-
Status: Resolved
-
Priority:
Major
-
Resolution: Fixed
-
Affects Version/s: None
-
Component/s: modules/facet
-
Labels:None
-
Lucene Fields:New, Patch Available
Description
I'm doing some facet performance tests, and I tried using the CachedOrdsCountingFacetsAggregator to cache the decoded ords per doc X field ... but noticed it was generating way too many cache entries, because it's currently using the NDV instance as the cache key.
NDV instances are thread-private so this results in way too many entries in the cache.
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
Good catch!
I guess we should use a compound cache key: coreCacheKey + clp.field?
Patch changes the map to be a WeakHashMap<Object,Map<String,CachedOrds>> so the outer map is keyed by reader.getCoreCacheKey() and the inner map is from field to CachedOrds, where field is the BinaryDV field which holds the facets ordinals.
I also added ramBytesUsed and a test which verifies that ramBytesUsed does not change between threads.
+1, thanks Shai!
Commit 1535163 from Shai Erera in branch 'dev/trunk'
[ https://svn.apache.org/r1535163 ]
LUCENE-5303: OrdinalsCache should use reader.getCoreCacheKey()
Commit 1535167 from Shai Erera in branch 'dev/branches/branch_4x'
[ https://svn.apache.org/r1535167 ]
LUCENE-5303: OrdinalsCache should use reader.getCoreCacheKey()
Committed to trunk and 4x.
Commit 1535215 from Michael McCandless in branch 'dev/trunk'
[ https://svn.apache.org/r1535215 ]
LUCENE-5303: clear cache before each test
Commit 1535216 from Michael McCandless in branch 'dev/branches/branch_4x'
[ https://svn.apache.org/r1535216 ]
LUCENE-5303: clear cache before each test
Sorry, I meant BDV (BinaryDocValues) not NDV ...