private void ValidateCache(PTable.Midpoint[] cache, int count, int depth)
{
if (count == 0 || depth == 0)
{
Assert.IsNull(cache);
return;
}
if (count == 1)
{
Assert.IsNotNull(cache);
Assert.AreEqual(2, cache.Length);
Assert.AreEqual(0, cache[1].ItemIndex);
Assert.AreEqual(0, cache[1].ItemIndex);
return;
}
Assert.IsNotNull(cache);
Assert.AreEqual(Math.Min(count, 1<<depth), cache.Length);
Assert.AreEqual(0, cache[0].ItemIndex);
for (int i = 1; i < cache.Length; ++i)
{
Assert.IsTrue(cache[i - 1].Key.GreaterEqualsThan(cache[i].Key), "Expected {0} to be >= {1}", cache[i - 1].Key, cache[i].Key);
Assert.Less(cache[i-1].ItemIndex, cache[i].ItemIndex);
}
Assert.AreEqual(count-1, cache[cache.Length-1].ItemIndex);
}
public void Setup()
{
var mtable = new HashListMemTable(maxSize: 10);
mtable.Add(0x0101, 0x0001, 0x0001);
mtable.Add(0x0105, 0x0001, 0x0002);
_table = PTable.FromMemtable(mtable, Filename);
_table.MarkForDestruction();
}
public override void TestFixtureSetUp()
{
base.TestFixtureSetUp();
var table = new HashListMemTable(maxSize: 2000);
table.Add(0x0101, 0x0001, 0x0001);
_ptable = PTable.FromMemtable(table, Filename, cacheDepth: 0);
}
public void when_ptable_file_is_deleted()
{
_ptable.MarkForDestruction();
_ptable = null;
File.Delete(_ptableFileName);
Assert.Throws<CorruptIndexException>(() => IndexMap.FromFile(_indexMapFileName, 2));
}
public void Setup()
{
_filename = Path.GetRandomFileName();
var table = new HashListMemTable();
AddItemsForScenario(table);
PTable = PTable.FromMemtable(table, _filename, cacheDepth: _midpointCacheDepth);
}
public void Setup()
{
_filename = Path.GetRandomFileName();
var mtable = new HashListMemTable();
mtable.Add(0x0101, 0x0001, 0x0001);
mtable.Add(0x0105, 0x0001, 0x0002);
_table = PTable.FromMemtable(mtable, _filename);
_table.MarkForDestruction();
}
public override void TestFixtureSetUp()
{
base.TestFixtureSetUp();
_ptableCount = (long)(uint.MaxValue + 10000000L);
_size = _ptableCount * (long)indexEntrySize + PTableHeader.Size + PTable.MD5Size;
Console.WriteLine("Creating PTable at {0}. Size of PTable: {1}", Filename, _size);
CreatePTableFile(Filename, _size, indexEntrySize);
_ptable = PTable.FromFile(Filename, _ptableVersion, 22);
}
public override void SetUp()
{
base.SetUp();
var table = new HashListMemTable(maxSize: 50);
AddItemsForScenario(table);
PTable = PTable.FromMemtable(table, Filename, cacheDepth: _midpointCacheDepth);
}
public override void SetUp()
{
base.SetUp();
_indexMapFileName = GetFilePathFor("index.map");
_ptableFileName = GetFilePathFor("ptable");
_emptyIndexMap = IndexMap.FromFile(_indexMapFileName, _ptableVersion);
var memTable = new HashListMemTable(_ptableVersion, maxSize: 10);
memTable.Add(0, 1, 2);
_ptable = PTable.FromMemtable(memTable, _ptableFileName, _ptableVersion);
}
public override void SetUp()
{
base.SetUp();
_indexMapFileName = Path.Combine(PathName, "index.map");
_ptableFileName = Path.Combine(PathName, "ptable");
_emptyIndexMap = IndexMap.FromFile(_indexMapFileName, x => false);
var memTable = new HashListMemTable();
memTable.Add(0, 1, 2);
_ptable = PTable.FromMemtable(memTable, _ptableFileName);
}
public override void TestFixtureSetUp()
{
base.TestFixtureSetUp();
var table = new HashListMemTable();
table.Add(0x0101, 0x0001, 0x0001);
table.Add(0x0105, 0x0001, 0x0002);
table.Add(0x0102, 0x0001, 0x0003);
table.Add(0x0102, 0x0002, 0x0004);
table.Add(0x0103, 0x0001, 0xFFF1);
table.Add(0x0103, 0x0003, 0xFFF3);
table.Add(0x0103, 0x0005, 0xFFF5);
_ptable = PTable.FromMemtable(table, Filename, cacheDepth: 0);
}
public override void SetUp()
{
base.SetUp();
_indexMapFileName = GetFilePathFor("index.map");
_ptableFileName = GetFilePathFor("ptable");
var indexMap = IndexMap.FromFile(_indexMapFileName, maxTablesPerLevel: 2);
var memtable = new HashListMemTable(_ptableVersion, maxSize: 10);
memtable.Add(0,0,0);
memtable.Add(1,1,100);
_ptable = PTable.FromMemtable(memtable, _ptableFileName);
indexMap = indexMap.AddPTable(_ptable, 0, 0, (streamId, hash) => hash, _ => true, _ => new Tuple<string, bool>("", true), new GuidFilenameProvider(PathName), _ptableVersion).MergedMap;
indexMap.SaveToFile(_indexMapFileName);
}
public override void SetUp()
{
base.SetUp();
_indexMapFileName = Path.Combine(PathName, "index.map");
_ptableFileName = Path.Combine(PathName, "ptable");
var indexMap = IndexMap.FromFile(_indexMapFileName, x => false, maxTablesPerLevel: 2);
var memtable = new HashListMemTable(maxSize: 2000);
memtable.Add(0,0,0);
memtable.Add(1,1,100);
_ptable = PTable.FromMemtable(memtable, _ptableFileName);
indexMap = indexMap.AddFile(_ptable, 0, 0, new GuidFilenameProvider(PathName)).MergedMap;
indexMap.SaveToFile(_indexMapFileName);
}
public override void SetUp()
{
base.SetUp();
_indexMapFileName = GetFilePathFor("index.map");
_ptableFileName = GetFilePathFor("ptable");
var indexMap = IndexMap.FromFile(_indexMapFileName, maxTablesPerLevel: 2);
var memtable = new HashListMemTable(maxSize: 10);
memtable.Add(0,0,0);
memtable.Add(1,1,100);
_ptable = PTable.FromMemtable(memtable, _ptableFileName);
indexMap = indexMap.AddPTable(_ptable, 0, 0, _ => true, new GuidFilenameProvider(PathName)).MergedMap;
indexMap.SaveToFile(_indexMapFileName);
}
private static void InsertTableToTables(List <List <PTable> > tables, int level, int position, PTable table)
{
while (level >= tables.Count)
{
tables.Add(new List <PTable>());
}
var innerTables = tables[level] ?? (tables[level] = new List <PTable>());
while (position >= innerTables.Count)
{
innerTables.Add(null);
}
innerTables[position] = table;
}
public void Setup()
{
for (int i = 0; i < 2; i++)
{
_files.Add(Path.GetRandomFileName());
var table = new HashListMemTable();
for (int j = 0; j < 10; j++)
{
table.Add((UInt32)j + 1, i + 1, i * j);
}
_tables.Add(PTable.FromMemtable(table, _files[i]));
}
_files.Add(Path.GetRandomFileName());
_newtable = PTable.MergeTo(_tables, _files[2], x => false);
}
public MergeResult AddPTableForManualMerge(long prepareCheckpoint,
long commitCheckpoint,
Func <string, ulong, ulong> upgradeHash,
Func <IndexEntry, bool> existsAt,
Func <IndexEntry, Tuple <string, bool> > recordExistsAt,
IIndexFilenameProvider filenameProvider,
byte version,
int indexCacheDepth = 16,
bool skipIndexVerify = false)
{
Ensure.Nonnegative(prepareCheckpoint, "prepareCheckpoint");
Ensure.Nonnegative(commitCheckpoint, "commitCheckpoint");
var tables = CopyFrom(_map);
if (tables.Count < _maxTableLevelsForAutomaticMerge)
{
return(new MergeResult(this, new List <PTable>()));
}
var toDelete = new List <PTable>();
var tablesToMerge = tables.Skip(_maxTableLevelsForAutomaticMerge).SelectMany(a => a).ToList();
if (tablesToMerge.Count == 1)
{
return(new MergeResult(this, new List <PTable>()));
}
var filename = filenameProvider.GetFilenameNewTable();
PTable mergedTable = PTable.MergeTo(tablesToMerge, filename, upgradeHash, existsAt, recordExistsAt,
version, ESConsts.PTableInitialReaderCount, _pTableMaxReaderCount,
indexCacheDepth, skipIndexVerify);
for (int i = tables.Count - 1; i > _maxTableLevelsForAutomaticMerge; i--)
{
tables.RemoveAt(i);
}
tables[_maxTableLevelsForAutomaticMerge].Clear();
AddTableToTables(tables, _maxTableLevelsForAutomaticMerge + 1, mergedTable);
toDelete.AddRange(tablesToMerge);
var indexMap = new IndexMap(Version, tables, prepareCheckpoint, commitCheckpoint, _maxTablesPerLevel,
_maxTableLevelsForAutomaticMerge, _pTableMaxReaderCount);
return(new MergeResult(indexMap, toDelete));
}
public void Setup()
{
for (int i = 0; i < 4; i++)
{
_files.Add(Path.GetRandomFileName());
var table = new HashListMemTable();
for (int j = 0; j < 10; j++)
{
table.Add(0, 0, 1000000 - i*1000 - j);
table.Add(0, 0, i*1000 + j);
}
_tables.Add(PTable.FromMemtable(table, _files[i]));
}
_files.Add(Path.GetRandomFileName());
_newtable = PTable.MergeTo(_tables, _files[4], x => false);
}
public void setup()
{
_filename = Path.GetRandomFileName();
_copiedfilename = Path.GetRandomFileName();
var mtable = new HashListMemTable();
mtable.Add(0x0101, 0x0001, 0x0001);
mtable.Add(0x0105, 0x0001, 0x0002);
_table = PTable.FromMemtable(mtable, _filename);
File.Copy(_filename, _copiedfilename);
_table.MarkForDestruction();
using(var f = new FileStream(_copiedfilename, FileMode.Open, FileAccess.ReadWrite, FileShare.ReadWrite))
{
f.Seek(130, SeekOrigin.Begin);
f.WriteByte(0x22);
}
_table = PTable.FromFile(_copiedfilename);
}
public override void TestFixtureSetUp()
{
base.TestFixtureSetUp();
for (int i = 0; i < 4; i++)
{
_files.Add(GetTempFilePath());
var table = new HashListMemTable(maxSize: 30);
for (int j = 0; j < 10; j++)
{
table.Add((uint)i, j, i*10 + j);
}
_tables.Add(PTable.FromMemtable(table, _files[i]));
}
_files.Add(GetTempFilePath());
_newtable = PTable.MergeTo(_tables, _files[4], x => x.Position % 2 == 0);
}
public override void TestFixtureSetUp()
{
base.TestFixtureSetUp();
for (int i = 0; i < 4; i++)
{
_files.Add(GetTempFilePath());
var table = new HashListMemTable(_ptableVersion, maxSize: 30);
for (int j = 0; j < 10; j++)
{
table.Add((ulong)(0x010100000000 << i), j, i*10 + j);
}
_tables.Add(PTable.FromMemtable(table, _files[i]));
}
_files.Add(GetTempFilePath());
_newtable = PTable.MergeTo(_tables, _files[4], (streamId, hash) => hash, x => x.Position % 2 == 0, x => new Tuple<string, bool>("", x.Position % 2 == 0), _ptableVersion);
}
public MergeResult TryMergeOneLevel <TStreamId>(
Func <TStreamId, ulong, ulong> upgradeHash,
Func <IndexEntry, bool> existsAt,
Func <IndexEntry, Tuple <TStreamId, bool> > recordExistsAt,
IIndexFilenameProvider filenameProvider,
byte version,
int indexCacheDepth = 16,
bool skipIndexVerify = false,
bool useBloomFilter = true,
int lruCacheSize = 1_000_000)
{
var tables = CopyFrom(_map);
var hasMergedAny = false;
var canMergeAny = false;
var toDelete = new List <PTable>();
var maxTableLevelsToMerge = Math.Min(tables.Count, _maxTableLevelsForAutomaticMerge);
for (int level = 0; level < maxTableLevelsToMerge; level++)
{
if (tables[level].Count >= _maxTablesPerLevel)
{
if (hasMergedAny)
{
canMergeAny = true;
break;
}
var filename = filenameProvider.GetFilenameNewTable();
PTable mergedTable = PTable.MergeTo(tables[level], filename, upgradeHash, existsAt, recordExistsAt,
version,
ESConsts.PTableInitialReaderCount, _pTableMaxReaderCount,
indexCacheDepth, skipIndexVerify,
useBloomFilter, lruCacheSize);
hasMergedAny = true;
AddTableToTables(tables, level + 1, mergedTable);
toDelete.AddRange(tables[level]);
tables[level].Clear();
}
}
var indexMap = new IndexMap(Version, tables, PrepareCheckpoint, CommitCheckpoint, _maxTablesPerLevel,
_maxTableLevelsForAutomaticMerge, _pTableMaxReaderCount);
return(new MergeResult(indexMap, toDelete, hasMergedAny, canMergeAny));
}
public void Setup()
{
_filename = GetTempFilePath();
_copiedfilename = GetTempFilePath();
var mtable = new HashListMemTable(_ptableVersion, maxSize: 10);
mtable.Add(0x010100000000, 0x0001, 0x0001);
mtable.Add(0x010500000000, 0x0001, 0x0002);
_table = PTable.FromMemtable(mtable, _filename);
_table.Dispose();
File.Copy(_filename, _copiedfilename);
using (var f = new FileStream(_copiedfilename, FileMode.Open, FileAccess.ReadWrite, FileShare.ReadWrite))
{
f.Seek(22, SeekOrigin.Begin);
f.WriteByte(0x22);
}
}
public override void TestFixtureSetUp()
{
base.TestFixtureSetUp();
for (int i = 0; i < 2; i++)
{
_files.Add(GetTempFilePath());
var table = new HashListMemTable(maxSize: 20);
for (int j = 0; j < 10; j++)
{
table.Add((UInt32)j + 1, i + 1, i * j);
}
_tables.Add(PTable.FromMemtable(table, _files[i]));
}
_files.Add(GetTempFilePath());
_newtable = PTable.MergeTo(_tables, _files[2], x => true);
}
public override void TestFixtureSetUp()
{
base.TestFixtureSetUp();
for (int i = 0; i < 2; i++)
{
_files.Add(GetTempFilePath());
var table = new HashListMemTable(_ptableVersion, maxSize: 20);
for (int j = 0; j < 10; j++)
{
table.Add((ulong)(0x010100000000 << (j + 1)), i + 1, i*j);
}
_tables.Add(PTable.FromMemtable(table, _files[i]));
}
_files.Add(GetTempFilePath());
_newtable = PTable.MergeTo(_tables, _files[2], (streamId, hash) => hash, x => true, x => new System.Tuple<string, bool>("", true), _ptableVersion);
}
public MergeResult TryManualMerge <TStreamId>(
Func <TStreamId, ulong, ulong> upgradeHash,
Func <IndexEntry, bool> existsAt,
Func <IndexEntry, Tuple <TStreamId, bool> > recordExistsAt,
IIndexFilenameProvider filenameProvider,
byte version,
int indexCacheDepth = 16,
bool skipIndexVerify = false,
bool useBloomFilter = true,
int lruCacheSize = 1_000_000)
{
var tables = CopyFrom(_map);
if (tables.Count <= _maxTableLevelsForAutomaticMerge)
{
return(new MergeResult(this, new List <PTable>(), false, false));
}
var toDelete = new List <PTable>();
var tablesToMerge = tables.Skip(_maxTableLevelsForAutomaticMerge).SelectMany(a => a).ToList();
if (tablesToMerge.Count == 1)
{
return(new MergeResult(this, new List <PTable>(), false, false));
}
var filename = filenameProvider.GetFilenameNewTable();
PTable mergedTable = PTable.MergeTo(tablesToMerge, filename, upgradeHash, existsAt, recordExistsAt,
version, ESConsts.PTableInitialReaderCount, _pTableMaxReaderCount,
indexCacheDepth, skipIndexVerify,
useBloomFilter, lruCacheSize);
for (int i = tables.Count - 1; i > _maxTableLevelsForAutomaticMerge; i--)
{
tables.RemoveAt(i);
}
tables[_maxTableLevelsForAutomaticMerge].Clear();
AddTableToTables(tables, _maxTableLevelsForAutomaticMerge + 1, mergedTable);
toDelete.AddRange(tablesToMerge);
var indexMap = new IndexMap(Version, tables, PrepareCheckpoint, CommitCheckpoint, _maxTablesPerLevel,
_maxTableLevelsForAutomaticMerge, _pTableMaxReaderCount);
return(new MergeResult(indexMap, toDelete, true, false));
}
public override void TestFixtureSetUp()
{
base.TestFixtureSetUp();
for (int i = 0; i < 4; i++)
{
_files.Add(GetTempFilePath());
var table = new HashListMemTable(maxSize: 50);
for (int j = 0; j < 10; j++)
{
table.Add(0, 0, 1000000 - i*1000 - j);
table.Add(0, 0, i*1000 + j);
}
_tables.Add(PTable.FromMemtable(table, _files[i]));
}
_files.Add(GetTempFilePath());
_newtable = PTable.MergeTo(_tables, _files[4], x => false);
}
public override void TestFixtureSetUp()
{
base.TestFixtureSetUp();
_files.Add(GetTempFilePath());
var table = new HashListMemTable(PTableVersions.Index32Bit, maxSize: 20);
table.Add(0x010100000000, 0, 0x0101);
table.Add(0x010200000000, 0, 0x0102);
table.Add(0x010300000000, 0, 0x0103);
table.Add(0x010400000000, 0, 0x0104);
_tables.Add(PTable.FromMemtable(table, GetTempFilePath()));
table = new HashListMemTable(PTableVersions.Index32Bit, maxSize: 20);
table.Add(0x010500000000, 0, 0x0105);
table.Add(0x010600000000, 0, 0x0106);
table.Add(0x010700000000, 0, 0x0107);
table.Add(0x010800000000, 0, 0x0108);
_tables.Add(PTable.FromMemtable(table, GetTempFilePath()));
_newtable = PTable.MergeTo(_tables, GetTempFilePath(), (streamId, hash) => hash + 1, x => true, x => new Tuple<string, bool>(x.Stream.ToString(), true), PTableVersions.Index32Bit);
}
private void ReclaimMemoryIfNeeded(List <TableItem> awaitingMemTables)
{
var toPutOnDisk = awaitingMemTables.OfType <IMemTable>().Count() - MaxMemoryTables;
for (var i = awaitingMemTables.Count - 1; i >= 1 && toPutOnDisk > 0; i--)
{
var memtable = awaitingMemTables[i].Table as IMemTable;
if (memtable == null || !memtable.MarkForConversion())
{
continue;
}
Log.Debug("Putting awaiting file as PTable instead of MemTable [{id}].", memtable.Id);
var ptable = PTable.FromMemtable(memtable, _fileNameProvider.GetFilenameNewTable(),
ESConsts.PTableInitialReaderCount,
_pTableMaxReaderCount,
_indexCacheDepth,
_skipIndexVerify);
var swapped = false;
lock (_awaitingTablesLock) {
for (var j = _awaitingMemTables.Count - 1; j >= 1; j--)
{
var tableItem = _awaitingMemTables[j];
if (!(tableItem.Table is IMemTable) || tableItem.Table.Id != ptable.Id)
{
continue;
}
swapped = true;
_awaitingMemTables[j] = new TableItem(ptable,
tableItem.PrepareCheckpoint,
tableItem.CommitCheckpoint,
tableItem.Level);
break;
}
}
if (!swapped)
{
ptable.MarkForDestruction();
}
toPutOnDisk--;
}
}
public override void SetUp()
{
base.SetUp();
_filename = GetTempFilePath();
_copiedfilename = GetTempFilePath();
var mtable = new HashListMemTable(maxSize: 10);
mtable.Add(0x0101, 0x0001, 0x0001);
mtable.Add(0x0105, 0x0001, 0x0002);
_table = PTable.FromMemtable(mtable, _filename);
_table.Dispose();
File.Copy(_filename, _copiedfilename);
using (var f = new FileStream(_copiedfilename, FileMode.Open, FileAccess.ReadWrite, FileShare.ReadWrite))
{
f.Seek(130, SeekOrigin.Begin);
f.WriteByte(0x22);
}
_table = PTable.FromFile(_copiedfilename);
}
public ScavengeResult Scavenge <TStreamId>(Guid toScavenge, CancellationToken ct,
Func <TStreamId, ulong, ulong> upgradeHash,
Func <IndexEntry, bool> existsAt,
Func <IndexEntry, Tuple <TStreamId, bool> > recordExistsAt,
IIndexFilenameProvider filenameProvider,
byte version,
int indexCacheDepth = 16,
bool skipIndexVerify = false,
bool useBloomFilter = true,
int lruCacheSize = 1_000_000)
{
var scavengedMap = CopyFrom(_map);
for (int level = 0; level < scavengedMap.Count; level++)
{
for (int i = 0; i < scavengedMap[level].Count; i++)
{
if (scavengedMap[level][i].Id == toScavenge)
{
long spaceSaved;
var filename = filenameProvider.GetFilenameNewTable();
var oldTable = scavengedMap[level][i];
PTable scavenged = PTable.Scavenged(oldTable, filename, upgradeHash, existsAt, recordExistsAt,
version, out spaceSaved, ESConsts.PTableInitialReaderCount, _pTableMaxReaderCount, indexCacheDepth, skipIndexVerify, useBloomFilter, lruCacheSize, ct);
if (scavenged == null)
{
return(ScavengeResult.Failed(oldTable, level, i));
}
scavengedMap[level][i] = scavenged;
var indexMap = new IndexMap(Version, scavengedMap, PrepareCheckpoint, CommitCheckpoint,
_maxTablesPerLevel, _maxTableLevelsForAutomaticMerge, _pTableMaxReaderCount);
return(ScavengeResult.Success(indexMap, oldTable, scavenged, spaceSaved, level, i));
}
}
}
throw new ArgumentException("Unable to find table in map.", nameof(toScavenge));
}
public MergeResult AddPTableForAutomaticMerge(PTable tableToAdd,
long prepareCheckpoint,
long commitCheckpoint,
Func <string, ulong, ulong> upgradeHash,
Func <IndexEntry, bool> existsAt,
Func <IndexEntry, Tuple <string, bool> > recordExistsAt,
IIndexFilenameProvider filenameProvider,
byte version,
int indexCacheDepth = 16,
bool skipIndexVerify = false)
{
Ensure.Nonnegative(prepareCheckpoint, "prepareCheckpoint");
Ensure.Nonnegative(commitCheckpoint, "commitCheckpoint");
var tables = CopyFrom(_map);
AddTableToTables(tables, 0, tableToAdd);
var toDelete = new List <PTable>();
var maxTableLevelsToMerge = Math.Min(tables.Count, _maxTableLevelsForAutomaticMerge);
for (int level = 0; level < maxTableLevelsToMerge; level++)
{
if (tables[level].Count >= _maxTablesPerLevel)
{
var filename = filenameProvider.GetFilenameNewTable();
PTable mergedTable = PTable.MergeTo(tables[level], filename, upgradeHash, existsAt, recordExistsAt,
version,
ESConsts.PTableInitialReaderCount, _pTableMaxReaderCount,
indexCacheDepth, skipIndexVerify);
AddTableToTables(tables, level + 1, mergedTable);
toDelete.AddRange(tables[level]);
tables[level].Clear();
}
}
var indexMap = new IndexMap(Version, tables, prepareCheckpoint, commitCheckpoint, _maxTablesPerLevel,
_maxTableLevelsForAutomaticMerge, _pTableMaxReaderCount);
return(new MergeResult(indexMap, toDelete));
}
public MergeResult AddPTable(PTable tableToAdd,
long prepareCheckpoint,
long commitCheckpoint,
Func <string, ulong, ulong> upgradeHash,
Func <IndexEntry, bool> existsAt,
Func <IndexEntry, Tuple <string, bool> > recordExistsAt,
IIndexFilenameProvider filenameProvider,
byte version,
int level,
int indexCacheDepth = 16,
bool skipIndexVerify = false)
{
if (level < _maxTableLevelsForAutomaticMerge)
{
return(AddPTableForAutomaticMerge(tableToAdd, prepareCheckpoint, commitCheckpoint, upgradeHash,
existsAt, recordExistsAt, filenameProvider, version, indexCacheDepth, skipIndexVerify));
}
//For manual merge, we are never adding any extra entries, just merging existing files, so the index p/c checkpoint won't change
return(AddPTableForManualMerge(PrepareCheckpoint, CommitCheckpoint, upgradeHash, existsAt, recordExistsAt,
filenameProvider, version, indexCacheDepth, skipIndexVerify));
}
public MergeResult AddPTable(PTable tableToAdd,
long prepareCheckpoint,
long commitCheckpoint,
Func <string, ulong, ulong> upgradeHash,
Func <IndexEntry, bool> existsAt,
Func <IndexEntry, Tuple <string, bool> > recordExistsAt,
IIndexFilenameProvider filenameProvider,
byte version,
int level,
int indexCacheDepth = 16,
bool skipIndexVerify = false)
{
bool isManual;
//if (_maxTableLevelsForAutomaticMerge == 0)
//{
// //when we are not auto merging at all, a manual merge will only be triggered if
// //there are entries in the index map. the table it passes is always the first table
// //at the maximum automerge level, so we only need to hit the first table and see if it
// //matches, otherwise it must be an an add of a memtable.
// //although we are not automatically merging, the automerge process is also responsible
// //for writing out the memtable that just got persisted so we want to call auto merge still
// isManual = _map.Count != 0 && _map[0].FirstOrDefault() == tableToAdd;
//}
//else
//{
isManual = level > _maxTableLevelsForAutomaticMerge;
//}
if (isManual)
{
//For manual merge, we are never adding any extra entries, just merging existing files, so the index p/c checkpoint won't change
return(AddPTableForManualMerge(PrepareCheckpoint, CommitCheckpoint, upgradeHash, existsAt,
recordExistsAt,
filenameProvider, version, indexCacheDepth, skipIndexVerify));
}
return(AddPTableForAutomaticMerge(tableToAdd, prepareCheckpoint, commitCheckpoint, upgradeHash,
existsAt, recordExistsAt, filenameProvider, version, indexCacheDepth, skipIndexVerify));
}
public ScavengeResult Scavenge(Guid toScavenge, CancellationToken ct,
Func <string, ulong, ulong> upgradeHash,
Func <IndexEntry, bool> existsAt,
Func <IndexEntry, Tuple <string, bool> > recordExistsAt,
IIndexFilenameProvider filenameProvider,
byte version,
int indexCacheDepth = 16,
bool skipIndexVerify = false)
{
var scavengedMap = CopyFrom(_map);
for (int level = 0; level < scavengedMap.Count; level++)
{
for (int i = 0; i < scavengedMap[level].Count; i++)
{
if (scavengedMap[level][i].Id == toScavenge)
{
long spaceSaved;
var filename = filenameProvider.GetFilenameNewTable();
var oldTable = scavengedMap[level][i];
PTable scavenged = PTable.Scavenged(oldTable, filename, upgradeHash, existsAt, recordExistsAt, version, out spaceSaved, indexCacheDepth, skipIndexVerify, ct);
if (scavenged == null)
{
return(ScavengeResult.Failed(oldTable, level, i));
}
scavengedMap[level][i] = scavenged;
var indexMap = new IndexMap(Version, scavengedMap, PrepareCheckpoint, CommitCheckpoint, _maxTablesPerLevel);
return(ScavengeResult.Success(indexMap, oldTable, scavenged, spaceSaved, level, i));
}
}
}
throw new ArgumentException("Unable to find table in map.", nameof(toScavenge));
}
private void ReclaimMemoryIfNeeded(List <TableItem> awaitingMemTables)
{
var toPutOnDisk = awaitingMemTables.OfType <IMemTable>().Count() - MaxMemoryTables;
for (var i = awaitingMemTables.Count - 1; i >= 1 && toPutOnDisk > 0; i--)
{
var memtable = awaitingMemTables[i].Table as IMemTable;
if (memtable == null || !memtable.MarkForConversion())
{
continue;
}
Log.Trace("Putting awaiting file as PTable instead of MemTable [{0}].", memtable.Id);
var ptable = PTable.FromMemtable(memtable, _fileNameProvider.GetFilenameNewTable());
bool swapped = false;
lock (_awaitingTablesLock)
{
for (int j = _awaitingMemTables.Count - 1; j >= 1; j--)
{
var tableItem = _awaitingMemTables[j];
if (tableItem.Table is IMemTable && tableItem.Table.Id == ptable.Id)
{
swapped = true;
_awaitingMemTables[j] = new TableItem(ptable,
tableItem.PrepareCheckpoint,
tableItem.CommitCheckpoint);
break;
}
}
}
if (!swapped)
{
ptable.MarkForDestruction();
}
toPutOnDisk--;
}
}
public MergeResult AddPTable(PTable tableToAdd,
long prepareCheckpoint,
long commitCheckpoint,
Func <string, ulong, ulong> upgradeHash,
Func <IndexEntry, bool> existsAt,
Func <IndexEntry, Tuple <string, bool> > recordExistsAt,
IIndexFilenameProvider filenameProvider,
byte version,
int indexCacheDepth = 16,
bool skipIndexVerify = false)
{
Ensure.Nonnegative(prepareCheckpoint, "prepareCheckpoint");
Ensure.Nonnegative(commitCheckpoint, "commitCheckpoint");
var tables = CopyFrom(_map);
CreateIfNeeded(0, tables);
tables[0].Add(tableToAdd);
var toDelete = new List <PTable>();
for (int level = 0; level < tables.Count; level++)
{
if (tables[level].Count >= _maxTablesPerLevel)
{
var filename = filenameProvider.GetFilenameNewTable();
PTable table = PTable.MergeTo(tables[level], filename, upgradeHash, existsAt, recordExistsAt, version, indexCacheDepth, skipIndexVerify);
CreateIfNeeded(level + 1, tables);
tables[level + 1].Add(table);
toDelete.AddRange(tables[level]);
tables[level].Clear();
}
}
var indexMap = new IndexMap(Version, tables, prepareCheckpoint, commitCheckpoint, _maxTablesPerLevel);
return(new MergeResult(indexMap, toDelete));
}
public void Setup()
{
for (int i = 0; i < 4; i++)
{
_files.Add(Path.GetRandomFileName());
var table = new HashListMemTable();
for (int j = 0; j < 10; j++)
{
table.Add((UInt32)j%8, i, i * j * 100 + i + j); // 0 collisions with 8, 1 collisions with 9
}
if (i == 3)
{
table.Add(0, int.MaxValue, 45);
table.Add(1, int.MaxValue, 45);
table.Add(2, int.MaxValue, 45);
table.Add(3, int.MaxValue, 45);
}
_tables.Add(PTable.FromMemtable(table, _files[i]));
}
_files.Add(Path.GetRandomFileName());
_newtable = PTable.MergeTo(_tables, _files[4], x => x.Stream <= 1);
}
public void Setup()
{
for (int i = 0; i < 2; i++)
{
_files.Add(Path.GetRandomFileName());
var table = new HashListMemTable(maxSize: 2000);
for (int j = 0; j < 10; j++)
{
table.Add((UInt32)j + 1, i + 1, i * j);
}
if (i == 1)
{
table.Add(1, int.MaxValue, 45);
table.Add(2, int.MaxValue, 45);
table.Add(3, int.MaxValue, 45);
table.Add(4, int.MaxValue, 45);
}
_tables.Add(PTable.FromMemtable(table, _files[i]));
}
_files.Add(Path.GetRandomFileName());
_newtable = PTable.MergeTo(_tables, _files[2], x => false);
}
public override void TestFixtureSetUp()
{
base.TestFixtureSetUp();
for (int i = 0; i < 2; i++)
{
_files.Add(GetTempFilePath());
var table = new HashListMemTable(maxSize: 30);
for (int j = 0; j < 10; j++)
{
table.Add((UInt32)j % 8, i, i * j * 100 + i + j); // 0 collisions with 8, 1 collisions with 9
}
if (i == 1)
{
table.Add(0, int.MaxValue, 45);
table.Add(1, int.MaxValue, 45);
table.Add(2, int.MaxValue, 45);
table.Add(3, int.MaxValue, 45);
}
_tables.Add(PTable.FromMemtable(table, _files[i]));
}
_files.Add(GetTempFilePath());
_newtable = PTable.MergeTo(_tables, _files[2], x => x.Stream <= 1);
}
public override void TestFixtureSetUp()
{
base.TestFixtureSetUp();
for (int i = 0; i < 4; i++)
{
_files.Add(GetTempFilePath());
var table = new HashListMemTable(maxSize: 20);
for (int j = 0; j < 10; j++)
{
table.Add((UInt32)j + 1, i + 1, i * j);
}
if (i == 3)
{
table.Add(1, int.MaxValue, 45);
table.Add(2, int.MaxValue, 45);
table.Add(3, int.MaxValue, 45);
table.Add(4, int.MaxValue, 45);
}
_tables.Add(PTable.FromMemtable(table, _files[i]));
}
_files.Add(GetTempFilePath());
_newtable = PTable.MergeTo(_tables, _files[4], x => false);
}
public static IndexMap FromFile(string filename, Func <IndexEntry, bool> isHashCollision, int maxTablesPerLevel = 4, bool loadPTables = true)
{
var tables = new List <List <PTable> >();
int version;
long prepareCheckpoint = -1;
long commitCheckpoint = -1;
if (!File.Exists(filename))
{
return(new IndexMap(IndexMapVersion, tables, prepareCheckpoint, commitCheckpoint, isHashCollision, maxTablesPerLevel));
}
using (var f = File.OpenRead(filename))
using (var reader = new StreamReader(f))
{
// calculate real MD5 hash except first 32 bytes which are string representation of stored hash
f.Position = 32;
var realHash = MD5Hash.GetHashFor(f);
f.Position = 0;
// read stored MD5 hash and convert it from string to byte array
string text;
if ((text = reader.ReadLine()) == null)
{
throw new CorruptIndexException("IndexMap file is empty.");
}
if (text.Length != 32 || !text.All(x => char.IsDigit(x) || (x >= 'A' && x <= 'F')))
{
throw new CorruptIndexException("Corrupted MD5 hash.");
}
// check expected and real hashes are the same
var expectedHash = new byte[16];
for (int i = 0; i < 16; ++i)
{
expectedHash[i] = Convert.ToByte(text.Substring(i * 2, 2), 16);
}
if (expectedHash.Length != realHash.Length)
{
throw new InvalidOperationException("Invalid length of expected and real hash.");
}
for (int i = 0; i < realHash.Length; ++i)
{
if (expectedHash[i] != realHash[i])
{
throw new CorruptIndexException("Expected and real hash are different.");
}
}
// at this point we can assume the format is ok, so actually no need to check errors.
if ((text = reader.ReadLine()) == null)
{
throw new CorruptIndexException("Corrupted version.");
}
version = int.Parse(text);
// read and check prepare/commit checkpoint
if ((text = reader.ReadLine()) == null)
{
throw new CorruptIndexException("Corrupted commit checkpoint.");
}
try
{
var checkpoints = text.Split('/');
if (!long.TryParse(checkpoints[0], out prepareCheckpoint) || prepareCheckpoint < -1)
{
throw new CorruptIndexException("Invalid prepare checkpoint.");
}
if (!long.TryParse(checkpoints[1], out commitCheckpoint) || commitCheckpoint < -1)
{
throw new CorruptIndexException("Invalid commit checkpoint.");
}
}
catch (Exception exc)
{
throw new CorruptIndexException("Corrupted prepare/commit checkpoints pair.", exc);
}
// all next lines are PTables sorted by levels
while ((text = reader.ReadLine()) != null)
{
if (prepareCheckpoint < 0 || commitCheckpoint < 0)
{
throw new CorruptIndexException("Negative prepare/commit checkpoint in non-empty IndexMap.");
}
if (!loadPTables)
{
break;
}
PTable ptable = null;
var pieces = text.Split(',');
try
{
var level = int.Parse(pieces[0]);
var position = int.Parse(pieces[1]);
var file = pieces[2];
var path = Path.GetDirectoryName(filename);
var ptablePath = Path.Combine(path, file);
ptable = PTable.FromFile(ptablePath);
ptable.VerifyFileHash();
CreateIfNeeded(level, tables);
tables[level].Insert(position, ptable);
}
catch (Exception exc)
{
// if PTable file path was correct, but data is corrupted, we still need to dispose opened streams
if (ptable != null)
{
ptable.Dispose();
}
// also dispose all previously loaded correct PTables
for (int i = 0; i < tables.Count; ++i)
{
for (int j = 0; j < tables[i].Count; ++j)
{
tables[i][j].Dispose();
}
}
throw new CorruptIndexException("Error while loading IndexMap.", exc);
}
}
}
return(new IndexMap(version, tables, prepareCheckpoint, commitCheckpoint, isHashCollision, maxTablesPerLevel));
}
public void when_ptable_data_is_corrupted()
{
_ptable.Dispose();
_ptable = null;
using (var fs = File.Open(_ptableFileName, FileMode.Open))
{
fs.Position = new Random().Next(PTableHeader.Size, (int)fs.Length);
var b = (byte)fs.ReadByte();
b ^= 1;
fs.Position -= 1;
fs.WriteByte(b);
}
Assert.Throws<CorruptIndexException>(() => IndexMap.FromFile(_indexMapFileName, 2));
}
public EnumerablePTable(PTable table, IEnumerator <IndexEntry> enumerator)
{
Table = table;
_enumerator = enumerator;
}
private void ReadOffQueue()
{
try {
while (true)
{
var indexmapFile = Path.Combine(_directory, IndexMapFilename);
if (_isManualMergePending)
{
Log.Debug("Performing manual index merge.");
_isManualMergePending = false;
using (var reader = _tfReaderFactory()) {
var manualMergeResult = _indexMap.TryManualMerge(
(streamId, currentHash) => UpgradeHash(streamId, currentHash),
entry => reader.ExistsAt(entry.Position),
entry => ReadEntry(reader, entry.Position),
_fileNameProvider,
_ptableVersion,
_indexCacheDepth,
_skipIndexVerify);
if (manualMergeResult.HasMergedAny)
{
_indexMap = manualMergeResult.MergedMap;
_indexMap.SaveToFile(indexmapFile);
manualMergeResult.ToDelete.ForEach(x => x.MarkForDestruction());
}
Log.Debug("Manual index merge completed: {numMergedPTables} PTable(s) merged.", manualMergeResult.ToDelete.Count);
}
}
TableItem tableItem;
//ISearchTable table;
lock (_awaitingTablesLock) {
Log.Debug("Awaiting tables queue size is: {awaitingMemTables}.", _awaitingMemTables.Count);
if (_awaitingMemTables.Count == 1)
{
return;
}
tableItem = _awaitingMemTables[_awaitingMemTables.Count - 1];
}
PTable ptable;
var memtable = tableItem.Table as IMemTable;
if (memtable != null)
{
memtable.MarkForConversion();
ptable = PTable.FromMemtable(memtable, _fileNameProvider.GetFilenameNewTable(),
ESConsts.PTableInitialReaderCount,
_pTableMaxReaderCount,
_indexCacheDepth, _skipIndexVerify);
}
else
{
ptable = (PTable)tableItem.Table;
}
var addResult = _indexMap.AddPTable(ptable, tableItem.PrepareCheckpoint, tableItem.CommitCheckpoint);
_indexMap = addResult.NewMap;
_indexMap.SaveToFile(indexmapFile);
if (addResult.CanMergeAny)
{
using (var reader = _tfReaderFactory()) {
MergeResult mergeResult;
do
{
mergeResult = _indexMap.TryMergeOneLevel(
(streamId, currentHash) => UpgradeHash(streamId, currentHash),
entry => reader.ExistsAt(entry.Position),
entry => ReadEntry(reader, entry.Position),
_fileNameProvider,
_ptableVersion,
_indexCacheDepth,
_skipIndexVerify);
if (mergeResult.HasMergedAny)
{
_indexMap = mergeResult.MergedMap;
_indexMap.SaveToFile(indexmapFile);
mergeResult.ToDelete.ForEach(x => x.MarkForDestruction());
}
} while (mergeResult.CanMergeAny);
}
}
lock (_awaitingTablesLock) {
var memTables = _awaitingMemTables.ToList();
var corrTable = memTables.First(x => x.Table.Id == ptable.Id);
memTables.Remove(corrTable);
// parallel thread could already switch table,
// so if we have another PTable instance with same ID,
// we need to kill that instance as we added ours already
if (!ReferenceEquals(corrTable.Table, ptable) && corrTable.Table is PTable)
{
((PTable)corrTable.Table).MarkForDestruction();
}
Log.Debug("There are now {awaitingMemTables} awaiting tables.", memTables.Count);
_awaitingMemTables = memTables;
}
}
} catch (FileBeingDeletedException exc) {
Log.Error(exc,
"Could not acquire chunk in TableIndex.ReadOffQueue. It is OK if node is shutting down.");
} catch (Exception exc) {
Log.Error(exc, "Error in TableIndex.ReadOffQueue");
throw;
} finally {
lock (_awaitingTablesLock) {
_backgroundRunning = false;
_backgroundRunningEvent.Set();
}
}
}
private static void AddTableToTables(List <List <PTable> > tables, int level, PTable table)
{
while (level >= tables.Count)
{
tables.Add(new List <PTable>());
}
var innerTables = tables[level] ?? (tables[level] = new List <PTable>());
innerTables.Add(table);
}
private void ReadOffQueue()
{
try
{
while (true)
{
TableItem tableItem;
//ISearchTable table;
lock (_awaitingTablesLock)
{
Log.Trace("Awaiting tables queue size is: {0}.", _awaitingMemTables.Count);
if (_awaitingMemTables.Count == 1)
{
_backgroundRunning = false;
_backgroundRunningEvent.Set();
return;
}
tableItem = _awaitingMemTables[_awaitingMemTables.Count - 1];
}
PTable ptable;
var memtable = tableItem.Table as IMemTable;
if (memtable != null)
{
memtable.MarkForConversion();
ptable = PTable.FromMemtable(memtable, _fileNameProvider.GetFilenameNewTable());
}
else
{
ptable = (PTable)tableItem.Table;
}
_indexMap.EnterUnsafeState(_directory);
var mergeResult = _indexMap.AddFile(ptable,
tableItem.PrepareCheckpoint,
tableItem.CommitCheckpoint,
_fileNameProvider);
_indexMap = mergeResult.MergedMap;
_indexMap.SaveToFile(Path.Combine(_directory, IndexMapFilename));
_indexMap.LeaveUnsafeState(_directory);
lock (_awaitingTablesLock)
{
//oh well at least its only a small lock that is very unlikely to ever be hit
var memTables = _awaitingMemTables.ToList();
var corrTable = memTables.First(x => x.Table.Id == ptable.Id);
memTables.Remove(corrTable);
// parallel thread could already switch table,
// so if we have another PTable instance with same ID,
// we need to kill that instance as we added ours already
if (!ReferenceEquals(corrTable.Table, ptable) && corrTable.Table is PTable)
{
((PTable)corrTable.Table).MarkForDestruction();
}
Log.Trace("There are now {0} awaiting tables.", memTables.Count);
_awaitingMemTables = memTables;
}
mergeResult.ToDelete.ForEach(x => x.MarkForDestruction());
}
}
catch (Exception exc)
{
Log.ErrorException(exc, "Error in TableIndex.ReadOffQueue");
throw;
}
}
private static List <List <PTable> > LoadPTables(StreamReader reader, string indexmapFilename, TFPos checkpoints, int cacheDepth, bool skipIndexVerify,
int threads)
{
var tables = new List <List <PTable> >();
try
{
try
{
Parallel.ForEach(GetAllLines(reader).Reverse(), // Reverse so we load the highest levels (biggest files) first - ensures we use concurrency in the most efficient way.
new ParallelOptions {
MaxDegreeOfParallelism = threads
},
indexMapEntry =>
{
if (checkpoints.PreparePosition < 0 || checkpoints.CommitPosition < 0)
{
throw new CorruptIndexException(
string.Format("Negative prepare/commit checkpoint in non-empty IndexMap: {0}.", checkpoints));
}
PTable ptable = null;
var pieces = indexMapEntry.Split(',');
try
{
var level = int.Parse(pieces[0]);
var position = int.Parse(pieces[1]);
var file = pieces[2];
var path = Path.GetDirectoryName(indexmapFilename);
var ptablePath = Path.Combine(path, file);
ptable = PTable.FromFile(ptablePath, cacheDepth, skipIndexVerify);
lock (tables)
{
InsertTableToTables(tables, level, position, ptable);
}
}
catch (Exception)
{
// if PTable file path was correct, but data is corrupted, we still need to dispose opened streams
if (ptable != null)
{
ptable.Dispose();
}
throw;
}
});
// Verify map is correct
for (int i = 0; i < tables.Count; ++i)
{
for (int j = 0; j < tables[i].Count; ++j)
{
if (tables[i][j] == null)
{
throw new CorruptIndexException($"indexmap is missing contiguous level,position {i},{j}");
}
}
}
}
catch (AggregateException aggEx)
{
// We only care that *something* has gone wrong, throw the first exception
throw aggEx.InnerException;
}
}
catch (Exception exc)
{
// also dispose all previously loaded correct PTables
for (int i = 0; i < tables.Count; ++i)
{
for (int j = 0; j < tables[i].Count; ++j)
{
if (tables[i][j] != null)
{
tables[i][j].Dispose();
}
}
}
throw new CorruptIndexException("Error while loading IndexMap.", exc);
}
return(tables);
}
private void ReadOffQueue()
{
try
{
while (true)
{
TableItem tableItem;
//ISearchTable table;
lock (_awaitingTablesLock)
{
Log.Trace("Awaiting tables queue size is: {0}.", _awaitingMemTables.Count);
if (_awaitingMemTables.Count == 1)
{
_backgroundRunning = false;
_backgroundRunningEvent.Set();
return;
}
tableItem = _awaitingMemTables[_awaitingMemTables.Count - 1];
}
PTable ptable;
var memtable = tableItem.Table as IMemTable;
if (memtable != null)
{
memtable.MarkForConversion();
ptable = PTable.FromMemtable(memtable, _fileNameProvider.GetFilenameNewTable(), _indexCacheDepth);
}
else
{
ptable = (PTable)tableItem.Table;
}
var indexmapFile = Path.Combine(_directory, IndexMapFilename);
MergeResult mergeResult;
using (var reader = _tfReaderFactory())
{
mergeResult = _indexMap.AddPTable(ptable, tableItem.PrepareCheckpoint, tableItem.CommitCheckpoint,
(streamId, currentHash) => UpgradeHash(streamId, currentHash),
entry => reader.ExistsAt(entry.Position),
entry => ReadEntry(reader, entry.Position), _fileNameProvider, _ptableVersion, _indexCacheDepth);
}
_indexMap = mergeResult.MergedMap;
_indexMap.SaveToFile(indexmapFile);
lock (_awaitingTablesLock)
{
var memTables = _awaitingMemTables.ToList();
var corrTable = memTables.First(x => x.Table.Id == ptable.Id);
memTables.Remove(corrTable);
// parallel thread could already switch table,
// so if we have another PTable instance with same ID,
// we need to kill that instance as we added ours already
if (!ReferenceEquals(corrTable.Table, ptable) && corrTable.Table is PTable)
{
((PTable)corrTable.Table).MarkForDestruction();
}
Log.Trace("There are now {0} awaiting tables.", memTables.Count);
_awaitingMemTables = memTables;
}
mergeResult.ToDelete.ForEach(x => x.MarkForDestruction());
}
}
catch (FileBeingDeletedException exc)
{
Log.ErrorException(exc, "Could not acquire chunk in TableIndex.ReadOffQueue. It is OK if node is shutting down.");
}
catch (Exception exc)
{
Log.ErrorException(exc, "Error in TableIndex.ReadOffQueue");
throw;
}
}
public static PTable Scavenged(PTable table, string outputFile, Func <string, ulong, ulong> upgradeHash,
Func <IndexEntry, bool> existsAt, Func <IndexEntry, Tuple <string, bool> > readRecord, byte version, out long spaceSaved,
int cacheDepth = 16, bool skipIndexVerify = false, CancellationToken ct = default(CancellationToken))
{
Ensure.NotNull(table, "table");
Ensure.NotNullOrEmpty(outputFile, "outputFile");
Ensure.Nonnegative(cacheDepth, "cacheDepth");
var indexEntrySize = GetIndexEntrySize(version);
var numIndexEntries = table.Count;
var fileSizeUpToIndexEntries = GetFileSizeUpToIndexEntries(numIndexEntries, version);
Log.Trace("PTables scavenge started with {numIndexEntries} entries.", numIndexEntries);
var watch = Stopwatch.StartNew();
long keptCount = 0L;
long droppedCount;
try
{
using (var f = new FileStream(outputFile, FileMode.CreateNew, FileAccess.ReadWrite, FileShare.None,
DefaultSequentialBufferSize, FileOptions.SequentialScan))
{
f.SetLength(fileSizeUpToIndexEntries);
f.Seek(0, SeekOrigin.Begin);
using (var md5 = MD5.Create())
using (var cs = new CryptoStream(f, md5, CryptoStreamMode.Write))
using (var bs = new BufferedStream(cs, DefaultSequentialBufferSize))
{
// WRITE HEADER
var headerBytes = new PTableHeader(version).AsByteArray();
cs.Write(headerBytes, 0, headerBytes.Length);
// WRITE SCAVENGED INDEX ENTRIES
var buffer = new byte[indexEntrySize];
using (var enumerator = new EnumerableTable(version, table, upgradeHash, existsAt, readRecord))
{
while (enumerator.MoveNext())
{
ct.ThrowIfCancellationRequested();
if (existsAt(enumerator.Current))
{
AppendRecordTo(bs, buffer, version, enumerator.Current, indexEntrySize);
keptCount++;
}
}
}
// We calculate this as the EnumerableTable can silently drop entries too.
droppedCount = numIndexEntries - keptCount;
var forceKeep = version > table.Version;
if (droppedCount == 0 && !forceKeep)
{
Log.Trace(
"PTable scavenge finished in {elapsed}. No entries removed so not keeping scavenged table.",
watch.Elapsed);
try
{
bs.Close();
File.Delete(outputFile);
}
catch (Exception ex)
{
Log.ErrorException(ex, "Unable to delete unwanted scavenged PTable: {outputFile}", outputFile);
}
spaceSaved = 0;
return(null);
}
if (droppedCount == 0 && forceKeep)
{
Log.Trace("Keeping scavenged index even though it isn't smaller; version upgraded.");
}
//CALCULATE AND WRITE MIDPOINTS
if (version >= PTableVersions.IndexV4)
{
var requiredMidpointCount = GetRequiredMidpointCount(keptCount, version, cacheDepth);
var midpoints = ComputeMidpoints(bs, f, version, indexEntrySize, keptCount, requiredMidpointCount, new List <Midpoint>(), ct);
WriteMidpointsTo(bs, f, version, indexEntrySize, buffer, keptCount, keptCount, requiredMidpointCount, midpoints);
}
bs.Flush();
cs.FlushFinalBlock();
f.FlushToDisk();
f.SetLength(f.Position + MD5Size);
// WRITE MD5
var hash = md5.Hash;
f.Write(hash, 0, hash.Length);
f.FlushToDisk();
}
}
Log.Trace("PTable scavenge finished in {elapsed} ({droppedCount} entries removed, {keptCount} remaining).", watch.Elapsed,
droppedCount, keptCount);
var scavengedTable = new PTable(outputFile, Guid.NewGuid(), depth: cacheDepth,
skipIndexVerify: skipIndexVerify);
spaceSaved = table._size - scavengedTable._size;
return(scavengedTable);
}
catch (Exception)
{
try
{
File.Delete(outputFile);
}
catch (Exception ex)
{
Log.ErrorException(ex, "Unable to delete unwanted scavenged PTable: {outputFile}", outputFile);
}
throw;
}
}
private void ReadOffQueue()
{
try
{
while (true)
{
TableItem tableItem;
//ISearchTable table;
lock (_awaitingTablesLock)
{
Log.Trace("Awaiting tables queue size is: {0}.", _awaitingMemTables.Count);
if (_awaitingMemTables.Count == 1)
{
_backgroundRunning = false;
_backgroundRunningEvent.Set();
return;
}
tableItem = _awaitingMemTables[_awaitingMemTables.Count - 1];
}
PTable ptable;
var memtable = tableItem.Table as IMemTable;
if (memtable != null)
{
memtable.MarkForConversion();
ptable = PTable.FromMemtable(memtable, _fileNameProvider.GetFilenameNewTable());
}
else
{
ptable = (PTable)tableItem.Table;
}
// backup current version of IndexMap in case following switch will be left in unsafe state
// this will allow to rebuild just part of index
var backupFile = Path.Combine(_directory, IndexMapBackupFilename);
var indexmapFile = Path.Combine(_directory, IndexMapFilename);
Helper.EatException(() =>
{
if (File.Exists(backupFile))
{
File.Delete(backupFile);
}
if (File.Exists(indexmapFile))
{
// same as File.Copy(indexmapFile, backupFile); but with forced flush
var indexmapContent = File.ReadAllBytes(indexmapFile);
using (var f = File.Create(backupFile))
{
f.Write(indexmapContent, 0, indexmapContent.Length);
f.FlushToDisk();
}
}
});
EnterUnsafeState(_directory);
MergeResult mergeResult;
using (var reader = _tfReaderFactory())
{
mergeResult = _indexMap.AddPTable(ptable, tableItem.PrepareCheckpoint, tableItem.CommitCheckpoint,
entry => reader.ExistsAt(entry.Position), _fileNameProvider);
}
_indexMap = mergeResult.MergedMap;
_indexMap.SaveToFile(indexmapFile);
LeaveUnsafeState(_directory);
lock (_awaitingTablesLock)
{
var memTables = _awaitingMemTables.ToList();
var corrTable = memTables.First(x => x.Table.Id == ptable.Id);
memTables.Remove(corrTable);
// parallel thread could already switch table,
// so if we have another PTable instance with same ID,
// we need to kill that instance as we added ours already
if (!ReferenceEquals(corrTable.Table, ptable) && corrTable.Table is PTable)
{
((PTable)corrTable.Table).MarkForDestruction();
}
Log.Trace("There are now {0} awaiting tables.", memTables.Count);
_awaitingMemTables = memTables;
}
// We'll keep indexmap.backup in case of crash. In case of crash we hope that all necessary
// PTables for previous version of IndexMap are still there, so we can rebuild
// from last step, not to do full rebuild.
mergeResult.ToDelete.ForEach(x => x.MarkForDestruction());
}
}
catch (FileBeingDeletedException exc)
{
Log.ErrorException(exc, "Couldn't acquire chunk in TableIndex.ReadOffQueue. It is ok if node is shutting down.");
}
catch (Exception exc)
{
Log.ErrorException(exc, "Error in TableIndex.ReadOffQueue");
throw;
}
}
