public KeyDataHolder()
{
KeyDataArray = new KeyData[100000];
Sync = new ManualResetEvent(true);
}
public static void Write(CacheyBashi cb, ushort keyLength, IEnumerable <KeyValuePair <HashBin, byte[]> > data)
{
var sw = new Stopwatch();
sw.Start();
List <string> batchFiles = new List <string>();
var batchNameFormat = Path.Combine(cb.Dir, cb.DbName) + ".keybatch_{0}";
var batchIndex = 0;
//take batches of 100k? arbitraty or maybe roughly calc mem requirements
//use 2 buffers, one for writing, and one for streaming out to file?
var keyDataArray1 = new KeyData[100000];
var keyDataArray2 = new KeyData[100000];
var activeKeyDataArray = keyDataArray1;
var index = 0;
var datFileIndex = 0;
var keyCount = (ulong)0;
Task batchWriteTask = null;
foreach (var kvp in data)
{
if (kvp.Key.Length != keyLength)
{
throw new ArgumentException($"All keys must be of the provided keyLength: {keyLength}");
}
//need to copy the key array here incase someone is re-using the buffer
activeKeyDataArray[index].Key = kvp.Key.Clone();
activeKeyDataArray[index].DataAddr.addr = (ulong)datFileIndex;
activeKeyDataArray[index].DataAddr.len = (ulong)kvp.Value.Length;
//need to write the dat file here so we can discard data from memory
//cleanup tasks
cb.CbData.UnsafeWrite(kvp.Value);
datFileIndex += kvp.Value.Length;
var newBatch = index == activeKeyDataArray.Length - 1;
if (newBatch)//time to sort and start a new batch
{
var batchFile = string.Format(batchNameFormat, batchIndex);
batchWriteTask?.Wait();
batchWriteTask?.Dispose();
var array = activeKeyDataArray;
batchWriteTask = Task.Run(() =>
{
WriteBatch(array, batchFile, array.Length);
});
batchFiles.Add(batchFile);
batchIndex++;
index = 0;
//swap the active buffer
if (activeKeyDataArray == keyDataArray1)
{
activeKeyDataArray = keyDataArray2;
}
else
{
activeKeyDataArray = keyDataArray1;
}
}
if (!newBatch)
{
index++;
}
keyCount++;
}
//did we finish processing exactly on a batch boundary?
//if so roll back a batch index.
if (index == 0 && batchIndex > 0)
{
batchIndex--;
}
if (batchWriteTask != null && !batchWriteTask.IsCompleted)
{
batchWriteTask.Wait();
}
if (index > 0) //write the remaining keys to the final batch
{
var batchFile = string.Format(batchNameFormat, batchIndex);
WriteBatch(activeKeyDataArray, batchFile, (int)index);
batchFiles.Add(batchFile);
}
//Console.WriteLine($"writing batches took: {sw.ElapsedMilliseconds}");
//no more sorting required if only 1 batch so just write the file directly
if (batchIndex == 1)
{
var outFile = cb.CbKey.FileStream;
var writer = new BinaryWriter(outFile);
writer.Write(keyCount);
//first the keys
foreach (var keyData in activeKeyDataArray)
{
outFile.Write(keyData.Key.Hash, 0, keyData.Key.Length);
}
//then the addr infos
foreach (var keyData in activeKeyDataArray)
{
writer.Write(keyData.DataAddr.addr);
writer.Write(keyData.DataAddr.len);
}
return;
}
sw.Restart();
//now sort the batches into the final file
SortAndWrite(batchFiles, keyCount, cb, keyLength, cb.KeyFile);
//Console.WriteLine($"sorting batches and writing took: {sw.ElapsedMilliseconds}");
}
