[refactor](nereids) make forbid_unknown_col_stats check more accurate (#24061)

ignore unknown col stats check if:
colunm not used in query
column is Array/Json/Map/Struct type
This commit is contained in:
minghong
2023-09-09 22:42:17 +08:00
committed by GitHub
parent e09e030652
commit 7c7e44fcc8
4 changed files with 129 additions and 32 deletions

View File

@ -126,6 +126,11 @@ import org.apache.doris.nereids.trees.plans.physical.PhysicalUnion;
import org.apache.doris.nereids.trees.plans.physical.PhysicalWindow;
import org.apache.doris.nereids.trees.plans.physical.RuntimeFilter;
import org.apache.doris.nereids.trees.plans.visitor.DefaultPlanVisitor;
import org.apache.doris.nereids.types.ArrayType;
import org.apache.doris.nereids.types.DataType;
import org.apache.doris.nereids.types.JsonType;
import org.apache.doris.nereids.types.MapType;
import org.apache.doris.nereids.types.StructType;
import org.apache.doris.nereids.util.ExpressionUtils;
import org.apache.doris.nereids.util.JoinUtils;
import org.apache.doris.nereids.util.Utils;
@ -204,7 +209,6 @@ import java.util.stream.Stream;
public class PhysicalPlanTranslator extends DefaultPlanVisitor<PlanFragment, PlanTranslatorContext> {
private static final Logger LOG = LogManager.getLogger(PhysicalPlanTranslator.class);
private final StatsErrorEstimator statsErrorEstimator;
private final PlanTranslatorContext context;
@ -236,6 +240,14 @@ public class PhysicalPlanTranslator extends DefaultPlanVisitor<PlanFragment, Pla
Collections.reverse(context.getPlanFragments());
// TODO: maybe we need to trans nullable directly? and then we could remove call computeMemLayout
context.getDescTable().computeMemLayout();
if (ConnectContext.get() != null && ConnectContext.get().getSessionVariable().forbidUnknownColStats) {
Set<ScanNode> scans = context.getScanNodeWithUnknownColumnStats();
if (!scans.isEmpty()) {
StringBuilder builder = new StringBuilder();
scans.forEach(scanNode -> builder.append(scanNode));
throw new AnalysisException("tables with unknown column stats: " + builder);
}
}
return rootFragment;
}
@ -542,6 +554,15 @@ public class PhysicalPlanTranslator extends DefaultPlanVisitor<PlanFragment, Pla
// TODO: move all node set cardinality into one place
if (olapScan.getStats() != null) {
olapScanNode.setCardinality((long) olapScan.getStats().getRowCount());
if (ConnectContext.get().getSessionVariable().forbidUnknownColStats) {
for (int i = 0; i < slots.size(); i++) {
Slot slot = slots.get(i);
if (olapScan.getStats().findColumnStatistics(slot).isUnKnown()
&& !isComplexDataType(slot.getDataType())) {
context.addUnknownStatsColumn(olapScanNode, tupleDescriptor.getSlots().get(i).getId());
}
}
}
}
// TODO: Do we really need tableName here?
TableName tableName = new TableName(null, "", "");
@ -2000,6 +2021,14 @@ public class PhysicalPlanTranslator extends DefaultPlanVisitor<PlanFragment, Pla
scanNode.getTupleDesc().getSlots().add(smallest);
}
try {
if (ConnectContext.get() != null && ConnectContext.get().getSessionVariable().forbidUnknownColStats) {
for (SlotId slotId : requiredByProjectSlotIdSet) {
if (context.isColumnStatsUnknown(scanNode, slotId)) {
throw new AnalysisException("meet unknown column stats on table " + scanNode);
}
}
context.removeScanFromStatsUnknownColumnsMap(scanNode);
}
scanNode.updateRequiredSlots(context, requiredByProjectSlotIdSet);
} catch (UserException e) {
Util.logAndThrowRuntimeException(LOG,
@ -2262,4 +2291,9 @@ public class PhysicalPlanTranslator extends DefaultPlanVisitor<PlanFragment, Pla
}
return outputExprs;
}
private boolean isComplexDataType(DataType dataType) {
return dataType instanceof ArrayType || dataType instanceof MapType || dataType instanceof JsonType
|| dataType instanceof StructType;
}
}

View File

@ -47,11 +47,13 @@ import org.apache.doris.thrift.TPushAggOp;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.collect.Lists;
import com.google.common.collect.Maps;
import com.google.common.collect.Sets;
import java.util.IdentityHashMap;
import java.util.List;
import java.util.Map;
import java.util.Optional;
import java.util.Set;
import java.util.stream.Collectors;
import javax.annotation.Nullable;
@ -97,6 +99,8 @@ public class PlanTranslatorContext {
private final Map<RelationId, TPushAggOp> tablePushAggOp = Maps.newHashMap();
private final Map<ScanNode, Set<SlotId>> statsUnknownColumnsMap = Maps.newHashMap();
public PlanTranslatorContext(CascadesContext ctx) {
this.translator = new RuntimeFilterTranslator(ctx.getRuntimeFilterContext());
}
@ -106,6 +110,34 @@ public class PlanTranslatorContext {
translator = null;
}
/**
* remember the unknown-stats column and its scan, used for forbid_unknown_col_stats check
*/
public void addUnknownStatsColumn(ScanNode scan, SlotId slotId) {
Set<SlotId> slots = statsUnknownColumnsMap.get(scan);
if (slots == null) {
statsUnknownColumnsMap.put(scan, Sets.newHashSet(slotId));
} else {
statsUnknownColumnsMap.get(scan).add(slotId);
}
}
public boolean isColumnStatsUnknown(ScanNode scan, SlotId slotId) {
Set<SlotId> unknownSlots = statsUnknownColumnsMap.get(scan);
if (unknownSlots == null) {
return false;
}
return unknownSlots.contains(slotId);
}
public void removeScanFromStatsUnknownColumnsMap(ScanNode scan) {
statsUnknownColumnsMap.remove(scan);
}
public Set<ScanNode> getScanNodeWithUnknownColumnStats() {
return statsUnknownColumnsMap.keySet();
}
public List<PlanFragment> getPlanFragments() {
return planFragments;
}

View File

@ -28,7 +28,6 @@ import org.apache.doris.common.Config;
import org.apache.doris.common.FeConstants;
import org.apache.doris.common.Pair;
import org.apache.doris.nereids.CascadesContext;
import org.apache.doris.nereids.exceptions.AnalysisException;
import org.apache.doris.nereids.memo.Group;
import org.apache.doris.nereids.memo.GroupExpression;
import org.apache.doris.nereids.trees.expressions.Alias;
@ -126,7 +125,6 @@ import org.apache.doris.statistics.StatisticConstants;
import org.apache.doris.statistics.StatisticRange;
import org.apache.doris.statistics.Statistics;
import org.apache.doris.statistics.StatisticsBuilder;
import org.apache.doris.statistics.util.StatisticsUtil;
import com.google.common.base.Preconditions;
import com.google.common.collect.Maps;
@ -639,36 +637,16 @@ public class StatsCalculator extends DefaultPlanVisitor<Statistics, Void> {
.setAvgSizeByte(slotReference.getColumn().get().getType().getSlotSize())
.build();
}
if (cache.isUnKnown) {
if (forbidUnknownColStats && !shouldIgnoreThisCol) {
if (StatisticsUtil.statsTblAvailable()) {
throw new AnalysisException(String.format("Found unknown stats for column:%s.%s.\n"
+ "It may caused by:\n"
+ "\n"
+ "1. This column never got analyzed\n"
+ "2. This table is empty\n"
+ "3. Stats load failed caused by unstable of backends,"
+ "and FE cached the unknown stats by default in this scenario\n"
+ "4. There is a bug, please report it to Doris community\n"
+ "\n"
+ "If an unknown stats for this column is tolerable,"
+ "you could set session variable `forbid_unknown_col_stats` to false to make planner"
+ " ignore this error and keep planning.", table.getName(), colName));
} else {
throw new AnalysisException("BE is not available!");
}
if (!cache.isUnKnown) {
rowCount = Math.max(rowCount, cache.count);
cache = setOlapPartitionInfo(table, cache);
Histogram histogram = getColumnHistogram(table, colName);
if (histogram != null) {
ColumnStatisticBuilder columnStatisticBuilder =
new ColumnStatisticBuilder(cache).setHistogram(histogram);
columnStatisticMap.put(slotReference, columnStatisticBuilder.build());
cache = columnStatisticBuilder.build();
}
columnStatisticMap.put(slotReference, cache);
continue;
}
rowCount = Math.max(rowCount, cache.count);
cache = setOlapPartitionInfo(table, cache);
Histogram histogram = getColumnHistogram(table, colName);
if (histogram != null) {
ColumnStatisticBuilder columnStatisticBuilder =
new ColumnStatisticBuilder(cache).setHistogram(histogram);
columnStatisticMap.put(slotReference, columnStatisticBuilder.build());
cache = columnStatisticBuilder.build();
}
columnStatisticMap.put(slotReference, cache);
}