private void HandleNestedValuesReduction(TransactionOperationContext indexContext, IndexingStatsScope stats,
CancellationToken token, MapReduceResultsStore modifiedStore,
IndexWriteOperation writer, LazyStringValue reduceKeyHash)
{
EnsureValidNestedValuesReductionStats(stats);
var numberOfEntriesToReduce = 0;
try
{
var section = modifiedStore.GetNestedResultsSection();
if (section.IsModified == false)
{
return;
}
using (_nestedValuesReductionStats.NestedValuesRead.Start())
{
numberOfEntriesToReduce += section.GetResults(indexContext, _aggregationBatch.Items);
}
stats.RecordReduceAttempts(numberOfEntriesToReduce);
AggregationResult result;
using (_nestedValuesReductionStats.NestedValuesAggregation.Start())
{
result = AggregateOn(_aggregationBatch.Items, indexContext, token);
}
if (section.IsNew == false)
{
writer.DeleteReduceResult(reduceKeyHash, stats);
}
foreach (var output in result.GetOutputs())
{
writer.IndexDocument(reduceKeyHash, output, stats, indexContext);
}
_index.ReducesPerSec.Mark(numberOfEntriesToReduce);
_metrics.MapReduceIndexes.ReducedPerSec.Mark(numberOfEntriesToReduce);
stats.RecordReduceSuccesses(numberOfEntriesToReduce);
}
catch (Exception e)
{
_index.HandleError(e);
var message = $"Failed to execute reduce function for reduce key '{reduceKeyHash}' on nested values of '{_indexDefinition.Name}' index.";
if (_logger.IsInfoEnabled)
{
_logger.Info(message, e);
}
stats.RecordReduceErrors(numberOfEntriesToReduce);
stats.AddReduceError(message + $" Exception: {e}");
}
}
private void LogReductionError(Exception error, LazyStringValue reduceKeyHash, IndexingStatsScope stats, bool updateStats, TreePage page,
int numberOfNestedValues = -1)
{
var builder = new StringBuilder("Failed to execute reduce function on ");
if (page != null)
{
builder.Append($"page {page} ");
}
else
{
builder.Append("nested values ");
}
builder.Append($"of '{_indexDefinition.Name}' index (reduce key hash: {reduceKeyHash}");
var sampleItem = _aggregationBatch?.Items?.FirstOrDefault();
if (sampleItem != null)
{
builder.Append($", sample item to reduce: {sampleItem}");
}
builder.Append(")");
var message = builder.ToString();
if (_logger.IsInfoEnabled)
{
_logger.Info(message, error);
}
if (updateStats)
{
var errorCount = page?.NumberOfEntries ?? numberOfNestedValues;
Debug.Assert(errorCount != -1);
stats.RecordReduceErrors(errorCount);
stats.AddReduceError(message + $" Exception: {error}");
}
}
private void HandleReductionError(Exception error, LazyStringValue reduceKeyHash, IndexWriteOperation writer, IndexingStatsScope stats, bool updateStats, TreePage page,
int numberOfNestedValues = -1)
{
var builder = new StringBuilder("Failed to execute reduce function on ");
if (page != null)
{
builder.Append($"page {page} ");
}
else
{
builder.Append("nested values ");
}
builder.Append($"of '{_indexDefinition.Name}' index. The relevant reduce result is going to be removed from the index ");
builder.Append($"as it would be incorrect due to encountered errors (reduce key hash: {reduceKeyHash}");
var sampleItem = _aggregationBatch?.Items?.FirstOrDefault();
if (sampleItem != null)
{
builder.Append($", sample item to reduce: {sampleItem}");
}
builder.Append(")");
var message = builder.ToString();
if (_logger.IsInfoEnabled)
{
_logger.Info(message, error);
}
try
{
writer.DeleteReduceResult(reduceKeyHash, stats);
}
catch (Exception e)
{
if (_logger.IsInfoEnabled)
{
_logger.Info($"Failed to delete an index result from '${_indexDefinition.Name}' index on reduce error (reduce key hash: ${reduceKeyHash})", e);
}
}
if (updateStats)
{
var numberOfEntries = page?.NumberOfEntries ?? numberOfNestedValues;
Debug.Assert(numberOfEntries != -1);
// we'll only want to record exceptions on some of these, to give the
// user information about what is going on, otherwise we'll have to wait a lot if we
// are processing a big batch, and this can be a perf killer. See: RavenDB-11038
stats.RecordReduceErrors(numberOfEntries);
if (stats.NumberOfKeptReduceErrors < IndexStorage.MaxNumberOfKeptErrors)
{
stats.AddReduceError(message + $" Exception: {error}");
}
var failureInfo = new IndexFailureInformation
{
Name = _index.Name,
MapErrors = stats.MapErrors,
MapAttempts = stats.MapAttempts,
ReduceErrors = stats.ReduceErrors,
ReduceAttempts = stats.ReduceAttempts
};
if (failureInfo.IsInvalidIndex(_isStaleBecauseOfRunningReduction))
{
throw new ExcessiveNumberOfReduceErrorsException("Excessive number of errors during the reduce phase for the current batch. Failure info: " +
failureInfo.GetErrorMessage());
}
}
}
private void HandleTreeReduction(TransactionOperationContext indexContext, IndexingStatsScope stats,
CancellationToken token, MapReduceResultsStore modifiedStore, LowLevelTransaction lowLevelTransaction,
IndexWriteOperation writer, LazyStringValue reduceKeyHash, Table table)
{
EnsureValidTreeReductionStats(stats);
var tree = modifiedStore.Tree;
var branchesToAggregate = new HashSet <long>();
var parentPagesToAggregate = new HashSet <long>();
var page = new TreePage(null, Constants.Storage.PageSize);
foreach (var modifiedPage in modifiedStore.ModifiedPages)
{
token.ThrowIfCancellationRequested();
page.Base = lowLevelTransaction.GetPage(modifiedPage).Pointer;
stats.RecordReduceTreePageModified(page.IsLeaf);
if (page.IsLeaf == false)
{
Debug.Assert(page.IsBranch);
branchesToAggregate.Add(modifiedPage);
continue;
}
var leafPage = page;
var compressed = leafPage.IsCompressed;
if (compressed)
{
stats.RecordCompressedLeafPage();
}
using (compressed ? (DecompressedLeafPage)(leafPage = tree.DecompressPage(leafPage, skipCache: true)) : null)
{
if (leafPage.NumberOfEntries == 0)
{
if (leafPage.PageNumber != tree.State.RootPageNumber)
{
throw new InvalidOperationException(
$"Encountered empty page which isn't a root. Page #{leafPage.PageNumber} in '{tree.Name}' tree.");
}
writer.DeleteReduceResult(reduceKeyHash, stats);
var emptyPageNumber = Bits.SwapBytes(leafPage.PageNumber);
using (Slice.External(indexContext.Allocator, (byte *)&emptyPageNumber, sizeof(long), out Slice pageNumSlice))
table.DeleteByKey(pageNumSlice);
continue;
}
var parentPage = tree.GetParentPageOf(leafPage);
stats.RecordReduceAttempts(leafPage.NumberOfEntries);
try
{
using (var result = AggregateLeafPage(leafPage, lowLevelTransaction, indexContext, token))
{
if (parentPage == -1)
{
writer.DeleteReduceResult(reduceKeyHash, stats);
foreach (var output in result.GetOutputs())
{
writer.IndexDocument(reduceKeyHash, output, stats, indexContext);
}
}
else
{
StoreAggregationResult(leafPage.PageNumber, leafPage.NumberOfEntries, table, result);
parentPagesToAggregate.Add(parentPage);
}
_metrics.MapReduceReducedPerSecond.Mark(leafPage.NumberOfEntries);
stats.RecordReduceSuccesses(leafPage.NumberOfEntries);
}
}
catch (Exception e)
{
_index.HandleError(e);
var message =
$"Failed to execute reduce function for reduce key '{tree.Name}' on a leaf page #{leafPage} of '{_indexDefinition.Name}' index.";
if (_logger.IsInfoEnabled)
{
_logger.Info(message, e);
}
if (parentPage == -1)
{
stats.RecordReduceErrors(leafPage.NumberOfEntries);
stats.AddReduceError(message + $" Exception: {e}");
}
}
}
}
long tmp = 0;
using (Slice.External(indexContext.Allocator, (byte *)&tmp, sizeof(long), out Slice pageNumberSlice))
{
foreach (var freedPage in modifiedStore.FreedPages)
{
tmp = Bits.SwapBytes(freedPage);
table.DeleteByKey(pageNumberSlice);
}
}
while (parentPagesToAggregate.Count > 0 || branchesToAggregate.Count > 0)
{
token.ThrowIfCancellationRequested();
var branchPages = parentPagesToAggregate;
parentPagesToAggregate = new HashSet <long>();
foreach (var pageNumber in branchPages)
{
page.Base = lowLevelTransaction.GetPage(pageNumber).Pointer;
try
{
if (page.IsBranch == false)
{
throw new InvalidOperationException("Parent page was found that wasn't a branch, error at " +
page.PageNumber);
}
stats.RecordReduceAttempts(page.NumberOfEntries);
var parentPage = tree.GetParentPageOf(page);
using (var result = AggregateBranchPage(page, table, indexContext, branchesToAggregate, token))
{
if (parentPage == -1)
{
writer.DeleteReduceResult(reduceKeyHash, stats);
foreach (var output in result.GetOutputs())
{
writer.IndexDocument(reduceKeyHash, output, stats, indexContext);
}
}
else
{
parentPagesToAggregate.Add(parentPage);
StoreAggregationResult(page.PageNumber, page.NumberOfEntries, table, result);
}
_metrics.MapReduceReducedPerSecond.Mark(page.NumberOfEntries);
stats.RecordReduceSuccesses(page.NumberOfEntries);
}
}
catch (Exception e)
{
_index.HandleError(e);
var message =
$"Failed to execute reduce function for reduce key '{tree.Name}' on a branch page #{page} of '{_indexDefinition.Name}' index.";
if (_logger.IsInfoEnabled)
{
_logger.Info(message, e);
}
stats.RecordReduceErrors(page.NumberOfEntries);
stats.AddReduceError(message + $" Exception: {e}");
}
finally
{
branchesToAggregate.Remove(pageNumber);
}
}
if (parentPagesToAggregate.Count == 0 && branchesToAggregate.Count > 0)
{
// we still have unaggregated branches which were modified but their children were not modified (branch page splitting) so we missed them
parentPagesToAggregate.Add(branchesToAggregate.First());
}
}
}
private void HandleReductionError(Exception error, LazyStringValue reduceKeyHash, Lazy <IndexWriteOperation> writer, IndexingStatsScope stats, bool updateStats, TreePage page,
int numberOfNestedValues = -1)
{
var builder = new StringBuilder("Failed to execute reduce function on ");
if (page != null)
{
builder.Append($"page {page} ");
}
else
{
builder.Append("nested values ");
}
builder.Append($"of '{_indexDefinition.Name}' index. The relevant reduce result is going to be removed from the index ");
builder.Append($"as it would be incorrect due to encountered errors (reduce key hash: {reduceKeyHash}");
var erroringResult = OnErrorResult;
if (erroringResult != null)
{
builder.Append($", current item to reduce: {erroringResult}");
}
else
{
erroringResult = _aggregationBatch?.Items?.FirstOrDefault();
if (erroringResult != null)
{
builder.Append($", sample item to reduce: {erroringResult}");
}
}
builder.Append(")");
var message = builder.ToString();
if (_logger.IsInfoEnabled)
{
_logger.Info(message, error);
}
try
{
writer.Value.DeleteReduceResult(reduceKeyHash, stats);
}
catch (Exception e)
{
if (_logger.IsInfoEnabled)
{
_logger.Info($"Failed to delete an index result from '${_indexDefinition.Name}' index on reduce error (reduce key hash: ${reduceKeyHash})", e);
}
}
if (updateStats)
{
var numberOfEntries = page?.NumberOfEntries ?? numberOfNestedValues;
Debug.Assert(numberOfEntries != -1);
// we'll only want to record exceptions on some of these, to give the
// user information about what is going on, otherwise we'll have to wait a lot if we
// are processing a big batch, and this can be a perf killer. See: RavenDB-11038
stats.RecordReduceErrors(numberOfEntries);
if (stats.NumberOfKeptReduceErrors < IndexStorage.MaxNumberOfKeptErrors)
{
var reduceKey = GetReduceKey();
stats.AddReduceError(message + $" Exception: {error}", reduceKey);
}
var failureInfo = new IndexFailureInformation
{
Name = _index.Name,
MapErrors = stats.MapErrors,
MapAttempts = stats.MapAttempts,
ReduceErrors = stats.ReduceErrors,
ReduceAttempts = stats.ReduceAttempts
};
if (failureInfo.IsInvalidIndex(true))
{
throw new ExcessiveNumberOfReduceErrorsException("Excessive number of errors during the reduce phase for the current batch. Failure info: " +
failureInfo.GetErrorMessage());
}
string GetReduceKey()
{
if (erroringResult == null)
{
return(null);
}
try
{
var mapReduceDef = _index.Definition as MapReduceIndexDefinition;
var autoMapReduceDef = _index.Definition as AutoMapReduceIndexDefinition;
var groupByKeys = (mapReduceDef?.GroupByFields.Select(x => x.Key) ??
autoMapReduceDef?.GroupByFields.Select(x => x.Key))?.ToList();
StringBuilder reduceKeyValue = null;
if (groupByKeys != null)
{
foreach (var key in groupByKeys)
{
if (erroringResult.TryGetMember(key, out var result))
{
if (reduceKeyValue == null)
{
reduceKeyValue = new StringBuilder("Reduce key: { ");
}
reduceKeyValue.Append($"'{key}' : {result?.ToString() ?? "null"}");
}
}
reduceKeyValue?.Append(" }");
}
return(reduceKeyValue?.ToString());
}
catch
{
// ignore - make sure we don't error on error reporting
return(null);
}
}
}
}
