Github user JamesRTaylor commented on a diff in the pull request: https://github.com/apache/phoenix/pull/156#discussion_r58637402 --- Diff: phoenix-core/src/main/java/org/apache/phoenix/iterate/TableResultIterator.java --- @@ -107,8 +127,37 @@ public synchronized void close() throws SQLException { @Override public synchronized Tuple next() throws SQLException { initScanner(); - Tuple t = scanIterator.next(); - return t; + try { + lastTuple = scanIterator.next(); + if (lastTuple != null) { + ImmutableBytesWritable ptr = new ImmutableBytesWritable(); + lastTuple.getKey(ptr); + } + } catch (SQLException e) { + try { + throw ServerUtil.parseServerException(e); + } catch(StaleRegionBoundaryCacheException e1) { + if(scan.getAttribute(NON_AGGREGATE_QUERY)!=null) { + Scan newScan = ScanUtil.newScan(scan); + if(lastTuple != null) { + lastTuple.getKey(ptr); + byte[] startRowSuffix = ByteUtil.copyKeyBytesIfNecessary(ptr); + if(ScanUtil.isLocalIndex(newScan)) { + newScan.setAttribute(SCAN_START_ROW_SUFFIX, ByteUtil.nextKey(startRowSuffix)); + } else { + newScan.setStartRow(ByteUtil.nextKey(startRowSuffix)); + } + } + plan.getContext().getConnection().getQueryServices().clearTableRegionCache(htable.getTableName()); + this.scanIterator = + plan.iterator(DefaultParallelScanGrouper.getInstance(), newScan); --- End diff -- So we're re-running the scan, only if it's an aggregate scan? Does this apply only for an "ordered" group by (which is grouping by the leading part of the pk and doing so in-place)? In the unordered case, each scan is done during the post open and the results are cached on the RS, so a split won't impact the results. There may be a race condition between the pre open (where we check if we're in the region) and the post open (where we start the region operation), but I think that could be solved by doing the in-region check again). In these cases, we already handle this through the exception catching code in BaseResultIterators.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---