Copy chunk bytes in TSDB store before sending to client

During head compaction mmaped memory gets released while gRPC is
marshaling bytes from that same memory region. This leads to a fatal
segfault and crashes the receiver. The segfault happens when marshaling
chunks specifically.

This commit modifies the TSDB store server to copy chunk bytes before
sending them to the client. I tried running this for a while and saw
no significant increase in memory usage.

Signed-off-by: Filip Petkovski <[email protected]>

pkg/store/tsdb.go

@@ -202,13 +202,15 @@ func (s *TSDBStore) Series(r *storepb.SeriesRequest, srv storepb.Store_SeriesSer
 				return status.Errorf(codes.Internal, "TSDBStore: found not populated chunk returned by SeriesSet at ref: %v", chk.Ref)
 			}
 
+			chunkBytes := make([]byte, len(chk.Chunk.Bytes()))
+			copy(chunkBytes, chk.Chunk.Bytes())
 			c := storepb.AggrChunk{
 				MinTime: chk.MinTime,
 				MaxTime: chk.MaxTime,
 				Raw: &storepb.Chunk{
 					Type: storepb.Chunk_Encoding(chk.Chunk.Encoding() - 1), // Proto chunk encoding is one off to TSDB one.
-					Data: chk.Chunk.Bytes(),
-					Hash: hashChunk(hasher, chk.Chunk.Bytes(), enableChunkHashCalculation),
+					Data: chunkBytes,
+					Hash: hashChunk(hasher, chunkBytes, enableChunkHashCalculation),
 				},
 			}
 			frameBytesLeft -= c.Size()