Clean up memory reuse a bit. #9272
Conversation
- don't allow for soft references anymore in the recycler - remove some abusive thread locals - don't recycle independently float/double and int/long pages, they are the same and just interpret bits differently.
| @@ -273,12 +273,6 @@ public String buildFullName(BuilderContext context) { | |||
| } | |||
| } | |||
|
|
|||
| private static final ThreadLocal<List<Field>> FIELD_LIST = new ThreadLocal<List<Field>>() { | |||
There was a problem hiding this comment.
I wonder if this should be a different change and we should look at the case where some docs have 1000 fields. Are we sure the auto-growth of the field list is not worth caching?
There was a problem hiding this comment.
This list is per-field so you would need 1000 values for a single field to make it grow to a capacity of 1000. I think it is quite uncommon? I am personally more concerned about the opposite case that you suddenly index a large document and memory is never reclaimed since the list is stored in a ThreadLocal?
There was a problem hiding this comment.
OH,I see. Misread the code :( .
In that case +1 on removal , it's typically one or two values. Note how the old code initialized it with a capacity of 2 - maybe keep that?
|
I like the paged array part. Left a minor concern regarding the field mappers change. |
|
@bleskes pushed a new commit |
|
LGTM. Thx! |
|
nice!, I love the fact that this frees more memory now..., I wonder if we can back port to 1.5 as well? this is big. |
|
Reopening for backport. |
- remove some abusive thread locals - don't recycle independently float/double and int/long pages, they are the same and just interpret bits differently. Close #9272
|
Merged to 1.x. I just had to leave the support for soft references which are still used by the CacheRecycler (used for facets). |
clintongormley
changed the title
same and just interpret bits differently.