-
Type: Improvement
-
Resolution: Won't Do
-
Priority: Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
Labels:None
keith.bostic points out that there is no reason we need to acquire ts_lock in timestamp_once(). The idea is that each worker thread updates its thread-local struct with its commit timestamp after its done the commit. Before that moment, the struct contains the commit ts from the previous commit (or 0). The utility thread that updates the oldest and stable timestamps just needs to scan all the thread local structs and get the minimum of all commit timestamps.