keith-turner commented on code in PR #3640:
URL: https://github.com/apache/accumulo/pull/3640#discussion_r1298580902
##########
server/manager/src/main/java/org/apache/accumulo/manager/TabletGroupWatcher.java:
##########
@@ -789,8 +801,77 @@ private void mergeMetadataRecords(MergeInfo info) throws
AccumuloException {
for (Entry<Key,Value> entry : scanner) {
Key key = entry.getKey();
Value value = entry.getValue();
+
+ final KeyExtent keyExtent = KeyExtent.fromMetaRow(key.getRow());
+
+ // Keep track of the last Key Extent seen so we can use it to fence
+ // of RFiles when merging the metadata
+ if (lastExtent != null && !keyExtent.equals(lastExtent)) {
+ previousKeyExtent = lastExtent;
+ }
+
+ // Special case for now to handle the highest/stop tablet which is
where files are
+ // merged to so we need to handle the deletes on update here as it
won't be handled later
+ // TODO: Can this be re-written to not have a special edge case and
make it simpler?
+ if (keyExtent.equals(stopExtent)) {
+ if (previousKeyExtent != null
+ && key.getColumnFamily().equals(DataFileColumnFamily.NAME)) {
+
+ // Fence off existing files by the end row of the previous tablet
and current tablet
+ final StoredTabletFile existing =
StoredTabletFile.of(key.getColumnQualifier());
+ // The end row should be inclusive for the current tablet and the
previous end row
+ // should be exclusive for the start row
+ Range fenced = new Range(previousKeyExtent.endRow(), false,
keyExtent.endRow(), true);
+
+ // Clip range if exists
+ fenced = existing.hasRange() ? existing.getRange().clip(fenced,
true) : fenced;
+
+ // If null the range is disjoint with the new tablet range so need
to handle
+ if (fenced == null) {
Review Comment:
Ok. Ideally the split code would only move files to a new child tablet if
the file had data in the range. The split code currently looks at each files
first and last key to make this determination, so it seems like it should work.
Need to figure out why its not.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
