[
https://issues.apache.org/jira/browse/PHOENIX-3578?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15875665#comment-15875665
]
chenglei edited comment on PHOENIX-3578 at 2/21/17 9:54 AM:
------------------------------------------------------------
This issue is caused by the join dynamic filter, from following RHS, we get
d.id is in (1,2):
{code}
select d.seq,d.col2,d.id from detail d where d.id between 1 and 2
{code}
so with join dynamic filter, m.id is also in (1,2). Before applying join
dynamic filter,LHS is:
{code}
select m,id,m.col1,d.seq,d.col2 from master m order by m.id desc
{code}
Obviously, LHS's OrderBy is {{OrderBy.REV_ROW_KEY_ORDER_BY}},after applying
join dynamic filter LHS turns to :
{code}
select m,id,m.col1,d.seq,d.col2 from master m where m.id in (1,2) order by
m.id desc
{code}
Notice LHS's OrderBy is still {{OrderBy.REV_ROW_KEY_ORDER_BY}} now,then
{{WhereOptimizer.pushKeyExpressionsToScan}} was called to push {{m.id in
(1,2)}} into Scan , and useSkipScan is true in following line 274 of
{{WhereOptimizer.pushKeyExpressionsToScan}} method,so the Scan would use
SkipScanFilter:
{code:borderStyle=solid}
273 stopExtracting |= (hasUnboundedRange && !forcedSkipScan) ||
(hasRangeKey && forcedRangeScan);
274 useSkipScan |= !stopExtracting && !forcedRangeScan &&
(keyRanges.size() > 1 || hasRangeKey);
{code}
next step the {{startRow}} and {{endRow}} of LHS's Scan was computed in
{{ScanRanges.create}} method, in following line 112 the LHS's RowKeySchema is
turned to SchemaUtil.VAR_BINARY_SCHEMA:
{code}
111 if (keys.size() > 1 ||
SchemaUtil.getSeparatorByte(schema.rowKeyOrderOptimizable(), false,
schema.getField(schema.getFieldCount()-1)) ==
QueryConstants.DESC_SEPARATOR_BYTE) {
112 schema = SchemaUtil.VAR_BINARY_SCHEMA;
113 slotSpan = ScanUtil.SINGLE_COLUMN_SLOT_SPAN;
114 } else {
{code}
so in following line 135 and line 136 of {{ScanRanges.create}} method,minKey is
{{\x80\x00\x00\x01}},and maxKey is {{\x80\x00\x00\x02\x00}}, and
correspondingly,the Scan's startRow is {{\x80\x00\x00\x01}}, and Scan's endRow
is {{\x80\x00\x00\x02\x00}}:
{code:borderStyle=solid}
134 if (nBuckets == null || !isPointLookup || !useSkipScan) {
135 byte[] minKey = ScanUtil.getMinKey(schema, sortedRanges,
slotSpan);
136 byte[] maxKey = ScanUtil.getMaxKey(schema, sortedRanges,
slotSpan);
{code}
In summary, when we scan the LHS {{master}} table, the Scan range is
{{[\x80\x00\x00\x01,\x80\x00\x00\x02\x00)}} ,and the Scan uses
{{SkipScanFilter}}.Furthermore,because the LHS's OrderBy is
{{OrderBy.REV_ROW_KEY_ORDER_By}},so the Scan range should be reversed.In
{{BaseScannerRegionObserver.preScannerOpen}} method,following
{{ScanUtil.setupReverseScan}} method is called to reverse the Scan's startRow
and endRow.Unfortunately, the reversed Scan's range computed by
{{ScanUtil.setupReverseScan}} method is
[\x80\x00\x00\x01\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF,
\x80\x00\x00\x00\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF),
so we can only get the rows of {{master}} table which id is 1, the rows which
id is 2 is excluded.
{code}
621 public static void setupReverseScan(Scan scan) {
622 if (isReversed(scan)) {
623 byte[] newStartRow = getReversedRow(scan.getStartRow());
624 byte[] newStopRow = getReversedRow(scan.getStopRow());
625 scan.setStartRow(newStopRow);
626 scan.setStopRow(newStartRow);
627 scan.setReversed(true);
628 }
629 }
{code}
In conclusion, following two problems causes this issue:
(1) the {{ScanUtil.getReversedRow}} method is not right for
{{\x80\x00\x00\x02\x00}},which should return {{\x80\x00\x00\x02}},not
{{\x80\x00\x00\x01\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF}}.
(2) even though {{ScanUtil.getReversedRow}} method is right,there may be
another problem,if I change the table data as following :
{noformat}
UPSERT INTO master VALUES (1, 'A1');
UPSERT INTO master VALUES (2, 'A2');
UPSERT INTO master VALUES (3, 'A3');
UPSERT INTO master VALUES (4, 'A4');
UPSERT INTO master VALUES (5, 'A5');
UPSERT INTO master VALUES (6, 'A6');
UPSERT INTO master VALUES (8, 'A8');
UPSERT INTO detail VALUES (1, 1, 'B1');
UPSERT INTO detail VALUES (2, 2, 'B2');
UPSERT INTO detail VALUES (3, 3, 'B3');
UPSERT INTO detail VALUES (4, 4, 'B4');
UPSERT INTO detail VALUES (5, 5, 'B5');
UPSERT INTO detail VALUES (6, 6, 'B6');
UPSERT INTO detail VALUES (7, 7, 'B7');
UPSERT INTO detail VALUES (8, 8, 'B8');
{noformat}
and modify the sql as :
{noformat}
select m.id, m.col1,d.col2 from master m, detail d where m.id = d.id and
d.id in (3,5,7) order by m.id desc
{noformat}
because the {{master}} table does not have rows which id=7,so the result of
{{ScanUtil.getReversedRow}} method is right,but executing the above sql would
throw following Exception(Under HBase 1.2.0 or lower):
{noformat}
org.apache.phoenix.exception.PhoenixIOException:
org.apache.hadoop.hbase.DoNotRetryIOException:
T000006,,1487348132415.92ad11583fe62ed8604f259d555b8f32.: The next hint must
come after previous hint
(prev=\x80\x00\x00\x07//LATEST_TIMESTAMP/Maximum/vlen=0/seqid=0,
next=\x80\x00\x00\x07//LATEST_TIMESTAMP/Maximum/vlen=0/seqid=0,
kv=\x80\x00\x00\x06/0:COL1/1487348135903/Put/vlen=2/seqid=4)
at
org.apache.phoenix.util.ServerUtil.createIOException(ServerUtil.java:89)
at
org.apache.phoenix.util.ServerUtil.throwIOException(ServerUtil.java:55)
at
org.apache.phoenix.coprocessor.BaseScannerRegionObserver$1.nextRaw(BaseScannerRegionObserver.java:489)
at
org.apache.phoenix.coprocessor.HashJoinRegionScanner.nextRaw(HashJoinRegionScanner.java:274)
at
org.apache.phoenix.coprocessor.DelegateRegionScanner.nextRaw(DelegateRegionScanner.java:77)
at
org.apache.phoenix.coprocessor.DelegateRegionScanner.nextRaw(DelegateRegionScanner.java:77)
at
org.apache.phoenix.coprocessor.BaseScannerRegionObserver$RegionScannerHolder.nextRaw(BaseScannerRegionObserver.java:261)
at
org.apache.hadoop.hbase.regionserver.RSRpcServices.scan(RSRpcServices.java:2555)
at
org.apache.hadoop.hbase.protobuf.generated.ClientProtos$ClientService$2.callBlockingMethod(ClientProtos.java:33648)
at org.apache.hadoop.hbase.ipc.RpcServer.call(RpcServer.java:2170)
at org.apache.hadoop.hbase.ipc.CallRunner.run(CallRunner.java:109)
at
org.apache.hadoop.hbase.ipc.RpcExecutor.consumerLoop(RpcExecutor.java:133)
at org.apache.hadoop.hbase.ipc.RpcExecutor$1.run(RpcExecutor.java:108)
at java.lang.Thread.run(Thread.java:745)
Caused by: java.lang.IllegalStateException: The next hint must come after
previous hint (prev=\x80\x00\x00\x07//LATEST_TIMESTAMP/Maximum/vlen=0/seqid=0,
next=\x80\x00\x00\x07//LATEST_TIMESTAMP/Maximum/vlen=0/seqid=0,
kv=\x80\x00\x00\x06/0:COL1/1487348135903/Put/vlen=2/seqid=4)
at
org.apache.phoenix.filter.SkipScanFilter.setNextCellHint(SkipScanFilter.java:171)
at
org.apache.phoenix.filter.SkipScanFilter.filterKeyValue(SkipScanFilter.java:145)
at
org.apache.hadoop.hbase.regionserver.ScanQueryMatcher.match(ScanQueryMatcher.java:427)
at
org.apache.hadoop.hbase.regionserver.StoreScanner.next(StoreScanner.java:522)
at
org.apache.hadoop.hbase.regionserver.KeyValueHeap.next(KeyValueHeap.java:147)
at
org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.populateResult(HRegion.java:5661)
at
org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.nextInternal(HRegion.java:5820)
at
org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.nextRaw(HRegion.java:5598)
at
org.apache.phoenix.coprocessor.BaseScannerRegionObserver$1.nextRaw(BaseScannerRegionObserver.java:459)
... 11 more
{noformat}
That is because {{SkipScanFilter}} can not support
{{OrderBy.REV_ROW_KEY_ORDER_BY}} now, if the OrderBy is
{{OrderBy.REV_ROW_KEY_ORDER_BY}}, we can not use {{SkipScanFilter}} after
applying join dynamic filter.
was (Author: comnetwork):
This issue is caused by the join dynamic filter, from following RHS, we get
d.id is in (1,2):
{code}
select d.seq,d.col2,d.id from detail d where d.id between 1 and 2
{code}
so with join dynamic filter, m.id is also in (1,2). Before applying join
dynamic filter,LHS is:
{code}
select m,id,m.col1,d.seq,d.col2 from master m order by m.id desc
{code}
Obviously, LHS's OrderBy is {{OrderBy.REV_ROW_KEY_ORDER_BY}},after applying
join dynamic filter LHS turns to :
{code}
select m,id,m.col1,d.seq,d.col2 from master m where m.id in (1,2) order by
m.id desc
{code}
Notice LHS's OrderBy is still {{OrderBy.REV_ROW_KEY_ORDER_BY}} now,then
{{WhereOptimizer.pushKeyExpressionsToScan}} was called to push {{m.id in
(1,2)}} into Scan , and useSkipScan is true in following line 274 of
{{WhereOptimizer.pushKeyExpressionsToScan}} method:
{code:borderStyle=solid}
273 stopExtracting |= (hasUnboundedRange && !forcedSkipScan) ||
(hasRangeKey && forcedRangeScan);
274 useSkipScan |= !stopExtracting && !forcedRangeScan &&
(keyRanges.size() > 1 || hasRangeKey);
{code}
next step the {{startRow}} and {{endRow}} of LHS's Scan was computed in
{{ScanRanges.create}} method, in following line 112 the LHS's RowKeySchema is
turned to SchemaUtil.VAR_BINARY_SCHEMA:
{code}
111 if (keys.size() > 1 ||
SchemaUtil.getSeparatorByte(schema.rowKeyOrderOptimizable(), false,
schema.getField(schema.getFieldCount()-1)) ==
QueryConstants.DESC_SEPARATOR_BYTE) {
112 schema = SchemaUtil.VAR_BINARY_SCHEMA;
113 slotSpan = ScanUtil.SINGLE_COLUMN_SLOT_SPAN;
114 } else {
{code}
so in following line 135 and line 136 of {{ScanRanges.create}} method,minKey is
{{\x80\x00\x00\x01}},and maxKey is {{\x80\x00\x00\x02\x00}}, and
correspondingly,the Scan's startRow is {{\x80\x00\x00\x01}}, and Scan's endRow
is {{\x80\x00\x00\x02\x00}}:
{code:borderStyle=solid}
134 if (nBuckets == null || !isPointLookup || !useSkipScan) {
135 byte[] minKey = ScanUtil.getMinKey(schema, sortedRanges,
slotSpan);
136 byte[] maxKey = ScanUtil.getMaxKey(schema, sortedRanges,
slotSpan);
{code}
In summary, when we scan the LHS {{master}} table, the Scan range is
{{[\x80\x00\x00\x01,\x80\x00\x00\x02\x00)}} ,and the Scan uses
{{SkipScanFilter}}.Furthermore,because the LHS's OrderBy is
{{OrderBy.REV_ROW_KEY_ORDER_By}},so the Scan range should be reversed.In
{{BaseScannerRegionObserver.preScannerOpen}} method,following
{{ScanUtil.setupReverseScan}} method is called to reverse the Scan's startRow
and endRow.Unfortunately, the reversed Scan's range computed by
{{ScanUtil.setupReverseScan}} method is
[\x80\x00\x00\x01\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF,
\x80\x00\x00\x00\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF),
so we can only get the rows of {{master}} table which id is 1, the rows which
id is 2 is excluded.
{code}
621 public static void setupReverseScan(Scan scan) {
622 if (isReversed(scan)) {
623 byte[] newStartRow = getReversedRow(scan.getStartRow());
624 byte[] newStopRow = getReversedRow(scan.getStopRow());
625 scan.setStartRow(newStopRow);
626 scan.setStopRow(newStartRow);
627 scan.setReversed(true);
628 }
629 }
{code}
In conclusion, following two problems causes this issue:
(1) the {{ScanUtil.getReversedRow}} method is not right for
{{\x80\x00\x00\x02\x00}},which should return {{\x80\x00\x00\x02}},not
{{\x80\x00\x00\x01\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF\xFF}}.
(2) even though {{ScanUtil.getReversedRow}} method is right,there may be
another problem,if I change the table data as following :
{noformat}
UPSERT INTO master VALUES (1, 'A1');
UPSERT INTO master VALUES (2, 'A2');
UPSERT INTO master VALUES (3, 'A3');
UPSERT INTO master VALUES (4, 'A4');
UPSERT INTO master VALUES (5, 'A5');
UPSERT INTO master VALUES (6, 'A6');
UPSERT INTO master VALUES (8, 'A8');
UPSERT INTO detail VALUES (1, 1, 'B1');
UPSERT INTO detail VALUES (2, 2, 'B2');
UPSERT INTO detail VALUES (3, 3, 'B3');
UPSERT INTO detail VALUES (4, 4, 'B4');
UPSERT INTO detail VALUES (5, 5, 'B5');
UPSERT INTO detail VALUES (6, 6, 'B6');
UPSERT INTO detail VALUES (7, 7, 'B7');
UPSERT INTO detail VALUES (8, 8, 'B8');
{noformat}
and modify the sql as :
{noformat}
select m.id, m.col1,d.col2 from master m, detail d where m.id = d.id and
d.id in (3,5,7) order by m.id desc
{noformat}
because the {{master}} table does not have rows which id=7,so the result of
{{ScanUtil.getReversedRow}} method is right,but executing the above sql would
throw following Exception(Under HBase 1.2.0 or lower):
{noformat}
org.apache.phoenix.exception.PhoenixIOException:
org.apache.hadoop.hbase.DoNotRetryIOException:
T000006,,1487348132415.92ad11583fe62ed8604f259d555b8f32.: The next hint must
come after previous hint
(prev=\x80\x00\x00\x07//LATEST_TIMESTAMP/Maximum/vlen=0/seqid=0,
next=\x80\x00\x00\x07//LATEST_TIMESTAMP/Maximum/vlen=0/seqid=0,
kv=\x80\x00\x00\x06/0:COL1/1487348135903/Put/vlen=2/seqid=4)
at
org.apache.phoenix.util.ServerUtil.createIOException(ServerUtil.java:89)
at
org.apache.phoenix.util.ServerUtil.throwIOException(ServerUtil.java:55)
at
org.apache.phoenix.coprocessor.BaseScannerRegionObserver$1.nextRaw(BaseScannerRegionObserver.java:489)
at
org.apache.phoenix.coprocessor.HashJoinRegionScanner.nextRaw(HashJoinRegionScanner.java:274)
at
org.apache.phoenix.coprocessor.DelegateRegionScanner.nextRaw(DelegateRegionScanner.java:77)
at
org.apache.phoenix.coprocessor.DelegateRegionScanner.nextRaw(DelegateRegionScanner.java:77)
at
org.apache.phoenix.coprocessor.BaseScannerRegionObserver$RegionScannerHolder.nextRaw(BaseScannerRegionObserver.java:261)
at
org.apache.hadoop.hbase.regionserver.RSRpcServices.scan(RSRpcServices.java:2555)
at
org.apache.hadoop.hbase.protobuf.generated.ClientProtos$ClientService$2.callBlockingMethod(ClientProtos.java:33648)
at org.apache.hadoop.hbase.ipc.RpcServer.call(RpcServer.java:2170)
at org.apache.hadoop.hbase.ipc.CallRunner.run(CallRunner.java:109)
at
org.apache.hadoop.hbase.ipc.RpcExecutor.consumerLoop(RpcExecutor.java:133)
at org.apache.hadoop.hbase.ipc.RpcExecutor$1.run(RpcExecutor.java:108)
at java.lang.Thread.run(Thread.java:745)
Caused by: java.lang.IllegalStateException: The next hint must come after
previous hint (prev=\x80\x00\x00\x07//LATEST_TIMESTAMP/Maximum/vlen=0/seqid=0,
next=\x80\x00\x00\x07//LATEST_TIMESTAMP/Maximum/vlen=0/seqid=0,
kv=\x80\x00\x00\x06/0:COL1/1487348135903/Put/vlen=2/seqid=4)
at
org.apache.phoenix.filter.SkipScanFilter.setNextCellHint(SkipScanFilter.java:171)
at
org.apache.phoenix.filter.SkipScanFilter.filterKeyValue(SkipScanFilter.java:145)
at
org.apache.hadoop.hbase.regionserver.ScanQueryMatcher.match(ScanQueryMatcher.java:427)
at
org.apache.hadoop.hbase.regionserver.StoreScanner.next(StoreScanner.java:522)
at
org.apache.hadoop.hbase.regionserver.KeyValueHeap.next(KeyValueHeap.java:147)
at
org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.populateResult(HRegion.java:5661)
at
org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.nextInternal(HRegion.java:5820)
at
org.apache.hadoop.hbase.regionserver.HRegion$RegionScannerImpl.nextRaw(HRegion.java:5598)
at
org.apache.phoenix.coprocessor.BaseScannerRegionObserver$1.nextRaw(BaseScannerRegionObserver.java:459)
... 11 more
{noformat}
That is because {{SkipScanFilter}} can not support
{{OrderBy.REV_ROW_KEY_ORDER_BY}} now, if the OrderBy is
{{OrderBy.REV_ROW_KEY_ORDER_BY}}, we can not use {{SkipScanFilter}} after
applying join dynamic filter.
> Incorrect query results when applying inner join and orderby desc
> -----------------------------------------------------------------
>
> Key: PHOENIX-3578
> URL: https://issues.apache.org/jira/browse/PHOENIX-3578
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 4.8.0
> Environment: hbase-1.1.2
> Reporter: sungmin.cho
>
> Step to reproduce:
> h4. 1. Create two tables
> {noformat}
> CREATE TABLE IF NOT EXISTS master (
> id integer not null,
> col1 varchar,
> constraint pk_master primary key(id)
> );
> CREATE TABLE IF NOT EXISTS detail (
> id integer not null,
> seq integer not null,
> col2 varchar,
> constraint pk_master primary key(id, seq)
> );
> {noformat}
> h4. 2. Upsert values
> {noformat}
> upsert into master values(1, 'A1');
> upsert into master values(2, 'A2');
> upsert into master values(3, 'A3');
> upsert into detail values(1, 1, 'B1');
> upsert into detail values(1, 2, 'B2');
> upsert into detail values(2, 1, 'B1');
> upsert into detail values(2, 2, 'B2');
> upsert into detail values(3, 1, 'B1');
> upsert into detail values(3, 2, 'B2');
> upsert into detail values(3, 3, 'B3');
> {noformat}
> h4. 3. Execute query
> {noformat}
> select m.id, m.col1, d.seq, d.col2
> from master m, detail d
> where m.id = d.id
> and d.id between 1 and 2
> order by m.id desc
> {noformat}
> h4. (/) Expected result
> {noformat}
> +-------+---------+--------+---------+
> | M.ID | M.COL1 | D.SEQ | D.COL2 |
> +-------+---------+--------+---------+
> | 2 | A2 | 1 | B1 |
> | 2 | A2 | 2 | B2 |
> | 1 | A1 | 1 | B1 |
> | 1 | A1 | 2 | B2 |
> +-------+---------+--------+---------+
> {noformat}
> h4. (!) Incorrect result
> {noformat}
> +-------+---------+--------+---------+
> | M.ID | M.COL1 | D.SEQ | D.COL2 |
> +-------+---------+--------+---------+
> | 1 | A1 | 1 | B1 |
> | 1 | A1 | 2 | B2 |
> +-------+---------+--------+---------+
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
