You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Because this process does not involve any epoch comparison, it may lead to gaps in the result region tree. For example, region A [m, x) merged to region B [c, m), produces new region B [c, x). If region B has a smaller region ID, it will be loaded before A and be removed when A is loaded, resulting a gap [c, m).
Possible fix is to compare region version and remove the one with older epoch.
The text was updated successfully, but these errors were encountered:
disksing
changed the title
LoadRegion from persistent storage should remove overlap region after compare epoch
LoadRegion from persistent storage should remove overlap region after compare version
Dec 12, 2019
Currently, when PD loads regions from persistent storage, it inserts regions to the region tree one by one. When there are regions have overlapped range, the region that loaded earlier will be removed from the tree. See https://github.com/pingcap/pd/blob/035f65eebbd028449d9ebb06cb166ab3c266136d/server/cluster/cluster.go#L245
Because this process does not involve any epoch comparison, it may lead to gaps in the result region tree. For example, region A
[m, x)
merged to region B[c, m)
, produces new region B[c, x)
. If region B has a smaller region ID, it will be loaded before A and be removed when A is loaded, resulting a gap[c, m)
.Possible fix is to compare region version and remove the one with older epoch.
The text was updated successfully, but these errors were encountered: