Details
-
Improvement
-
Resolution: Unresolved
-
Major - P3
-
None
-
None
-
None
-
None
-
Query Execution
Description
We currently keep track of the updated RecordIds in a hashmap. This could end up using huge amounts of memory if a multi:true update is modifying large number of documents. The is more problematic for clustered collections, given the size of the RecordIds for clustered collections can be much larger than a regular collection. We should spill to disk instead if the memory usage crosses certain threshold.