public List <Tuple <DateTime, long> > GetSegmentsIndex(BsonDocument version)
{
var result = new List <Tuple <DateTime, long> >();
var seg_ind_buf = new ByteBuffer();
seg_ind_buf.AppendDecompress(version["segment_index"].AsByteArray);
for (int ofs = 0; ofs < seg_ind_buf.Length; ofs += 16)
{
var t = seg_ind_buf.Read <long>(ofs);
var rec = new Tuple <DateTime, long>(DateTime64.ToDateTime(t), seg_ind_buf.Read <long>(ofs + 8));
result.Add(rec);
}
return(result);
}
public async Task <BsonDocument> AppendAsync(string symbol, DataFrame df, int chunksize = 0, bool skipAlreadyWrittenDates = true)
{
if (df.Index == null)
{
throw new ArgumentException("Please specify DataFrame.Index column before saving");
}
if (chunksize > 0 && df.Rows.Count > chunksize)
{
var rng = Range.R(0, chunksize);
BsonDocument ver = null;
int chunkscount = 0;
while (rng.First < df.Rows.Count)
{
var chunk = df[rng];
ver = await AppendAsync(symbol, chunk);
rng = Range.R(rng.First + chunksize, rng.Last + chunksize);
chunkscount++;
}
return(ver);
}
int attemptNo = 0;
for (;;)
{
var previous_version = await ReadVersionAsync(symbol);
var version = await GetNewVersion(symbol, previous_version);
/*var previous_version = await (_versions.AsQueryable ()
* .Where (v => v ["symbol"] == symbol && v ["version"] < version ["version"])
* .OrderByDescending (v => v ["version"]) as IAsyncCursorSource<BsonDocument>)
* .FirstOrDefaultAsync ();*/
var dtype = version.GetValue("dtype", "").ToString();
Log.Debug("loaded dtype {0}", dtype);
if (dtype != "" && df.DType.ToString() != dtype)
{
// dtype changed. need reload old data and repack it.
throw new NotImplementedException("old dtype {0}, new dtype {1}: not implemented".Args(dtype, df.DType));
}
var sdt = df.DType.ToString();
version["metadata"] = df.Metadata;
version["dtype"] = sdt;
Log.Debug("saved dtype {0}", sdt);
version["shape"] = new BsonArray {
{ -1 }
};
version["dtype_metadata"] = new BsonDocument {
{ "index", new BsonArray {
{ df.Index.Name }
} },
{ "columns", new BsonArray(df.Columns.Select(c => c.Name).ToList()) }
};
version["type"] = "pandasdf";
version["segment_count"] = previous_version != null ? previous_version["segment_count"].AsInt32 + 1 : 1;
version["append_count"] = previous_version != null ? previous_version["append_count"].AsInt32 + 1 : 0;
var seg_ind_buf = new ByteBuffer();
int segment_offset = 0;
bool is_date_time_index = DType.DateTime64.ToString().Equals(df.Index.DType.ToString());
//version ["base_sha"] = version ["sha"];
if (previous_version != null)
{
var seg_ind = previous_version["segment_index"].AsByteArray;
seg_ind_buf.AppendDecompress(seg_ind);
segment_offset = previous_version["up_to"].AsInt32;
if (is_date_time_index && skipAlreadyWrittenDates)
{
long date = seg_ind_buf.Read <long>(seg_ind_buf.Length - 16);
DateTime dt = DateTime64.ToDateTime(date);
var range = df.Index.AsDateTime().RangeOf(dt, 0, df.FilledCount - 1, Location.GT);
if (range.Last <= range.First)
{
Log.Information($"Skipped DataFrame.Append because date {dt} already written for {symbol}");
return(null); // Hey all was skipped
}
else if (range.First != 0)
{
Log.Information($"Skipped DataFrame.Append initial {range.First} elements date {dt} already written for {symbol}");
}
df = df[range];
}
}
var up_to = segment_offset + df.Rows.Count;
var buf = new ByteBuffer();
// add index that is last datetime + 0-based int64 index of last appended record like (segment_count-1)
if (is_date_time_index && df.Rows.Count > 0)
{
var date = df.Index.AsDateTime().Source[-1];
seg_ind_buf.Append <long>(date);
seg_ind_buf.Append <long>(up_to - 1);
}
var seg_ind_buf2 = new ByteBuffer();
seg_ind_buf2.AppendCompress(seg_ind_buf.GetBytes());
version["segment_index"] = seg_ind_buf2.GetBytes();
version["up_to"] = up_to;
var bin = df.ToBuffer();
buf.AppendCompress(bin);
var sha1 = SHA1.Create();
var sha = version.GetValue("sha", null);
if (sha == null)
{
byte[] hashBytes = sha1.ComputeHash(bin);
version["sha"] = new BsonBinaryData(hashBytes);
}
#if false
var buf2 = new ByteBuffer();
buf2.AppendDecompress(buf.GetBytes());
var bin2 = buf2.GetBytes();
if (!bin.SequenceEqual(bin2))
{
throw new InvalidOperationException();
}
var df2 = DataFrame.FromBuffer(bin2, df.DType, df.Rows.Count);
#endif
var segment = new BsonDocument {
{ "symbol", symbol },
{ "data", new BsonBinaryData(buf.GetBytes()) },
{ "compressed", true },
{ "segment", segment_offset + df.Rows.Count - 1 },
{ "parent", new BsonArray {
version["_id"]
} },
};
var hash = new ByteBuffer();
hash.Append(Encoding.ASCII.GetBytes(symbol));
foreach (var key in segment.Names.OrderByDescending(x => x))
{
var value = segment.GetValue(key);
if (value is BsonBinaryData)
{
hash.Append(value.AsByteArray);
}
else
{
var str = value.ToString();
hash.Append(Encoding.ASCII.GetBytes(str));
}
}
segment["sha"] = sha1.ComputeHash(hash.GetBytes());
//await _versions.InsertOneAsync(version);
try {
await _versions.ReplaceOneAsync(BF.Eq("symbol", symbol), version, Upsert);
}catch (MongoWriteException e)
{
Log.Information("Retrying append symbol {symbol}, attempt {attemptNo}", symbol, attemptNo++);
continue;
}
await _segments.InsertOneAsync(segment);
//Log.Information ("inserted new segment {segment} for symbol {symbol}", segment["_id"], symbol);
//Log.Information ("replaced version {0} for symbol {symbol} sha1 {sha}", version["_id"], symbol, sha);
// update parents in versions
//var res = await _segments.UpdateManyAsync (BF.Eq("symbol", symbol), BU.Set ("parent", new BsonArray{ version ["_id"] }));
//Log.Debug ("updated segments parents {0}".Args(res.MatchedCount));
return(version);
}
}
