Logo ROOT  
Reference Guide
RNTupleDescriptorFmt.cxx
Go to the documentation of this file.
1/// \file RNTupleDescriptorFmt.cxx
2/// \ingroup NTuple ROOT7
3/// \author Jakob Blomer <jblomer@cern.ch>
4/// \date 2019-08-25
5/// \warning This is part of the ROOT 7 prototype! It will change without notice. It might trigger earthquakes. Feedback
6/// is welcome!
7
8/*************************************************************************
9 * Copyright (C) 1995-2019, Rene Brun and Fons Rademakers. *
10 * All rights reserved. *
11 * *
12 * For the licensing terms see $ROOTSYS/LICENSE. *
13 * For the list of contributors see $ROOTSYS/README/CREDITS. *
14 *************************************************************************/
15
17#include <ROOT/RColumnModel.hxx>
19#include <ROOT/RNTupleUtil.hxx>
20
21#include <algorithm>
22#include <iomanip>
23#include <ostream>
24#include <unordered_map>
25#include <vector>
26
27namespace {
28
29struct ClusterInfo {
30 std::uint64_t fFirstEntry = 0;
31 std::uint32_t fNPages = 0;
32 std::uint32_t fNEntries = 0;
33 std::uint32_t fBytesOnStorage = 0;
34 std::uint32_t fBytesInMemory = 0;
35
36 bool operator ==(const ClusterInfo &other) const {
37 return fFirstEntry == other.fFirstEntry;
38 }
39
40 bool operator <(const ClusterInfo &other) const {
41 return fFirstEntry < other.fFirstEntry;
42 }
43};
44
45struct ColumnInfo {
47 std::uint64_t fLocalOrder = 0;
48 std::uint64_t fNElements = 0;
49 std::uint64_t fNPages = 0;
50 std::uint64_t fBytesOnStorage = 0;
51 std::uint32_t fElementSize = 0;
53 std::string fFieldName;
54
55 bool operator <(const ColumnInfo &other) const {
56 if (fFieldName == other.fFieldName)
57 return fLocalOrder < other.fLocalOrder;
58 return fFieldName < other.fFieldName;
59 }
60};
61
62static std::string GetFieldName(ROOT::Experimental::DescriptorId_t fieldId,
64{
65 const auto &fieldDesc = ntupleDesc.GetFieldDescriptor(fieldId);
66 if (fieldDesc.GetParentId() == ROOT::Experimental::kInvalidDescriptorId)
67 return fieldDesc.GetFieldName();
68 return GetFieldName(fieldDesc.GetParentId(), ntupleDesc) + "." + fieldDesc.GetFieldName();
69}
70
71static std::string GetColumnTypeName(ROOT::Experimental::EColumnType type)
72{
73 switch (type) {
75 return "Bit";
77 return "Byte";
79 return "Int32";
81 return "Int64";
83 return "Real32";
85 return "Real64";
87 return "Index";
89 return "Switch";
90 default:
91 return "UNKNOWN";
92 }
93}
94
95} // anonymous namespace
96
98{
99 std::vector<ColumnInfo> columns;
100 std::vector<ClusterInfo> clusters;
101 std::unordered_map<DescriptorId_t, unsigned int> cluster2Idx;
102 for (const auto &cluster : fClusterDescriptors) {
103 ClusterInfo info;
104 info.fFirstEntry = cluster.second.GetFirstEntryIndex();
105 info.fNEntries = cluster.second.GetNEntries();
106 cluster2Idx[cluster.first] = clusters.size();
107 clusters.emplace_back(info);
108 }
109
110 std::uint64_t bytesOnStorage = 0;
111 std::uint64_t bytesInMemory = 0;
112 std::uint64_t nPages = 0;
113 int compression = -1;
114 for (const auto &column : fColumnDescriptors) {
115 auto element = Detail::RColumnElementBase::Generate(column.second.GetModel().GetType());
116 auto elementSize = element.GetSize();
117
118 ColumnInfo info;
119 info.fFieldId = column.second.GetFieldId();
120 info.fLocalOrder = column.second.GetIndex();
121 info.fElementSize = elementSize;
122 info.fType = column.second.GetModel().GetType();
123
124 for (const auto &cluster : fClusterDescriptors) {
125 auto columnRange = cluster.second.GetColumnRange(column.first);
126 info.fNElements += columnRange.fNElements;
127 if (compression == -1) {
128 compression = columnRange.fCompressionSettings;
129 }
130 const auto &pageRange = cluster.second.GetPageRange(column.first);
131 auto idx = cluster2Idx[cluster.first];
132 for (const auto &page : pageRange.fPageInfos) {
133 bytesOnStorage += page.fLocator.fBytesOnStorage;
134 bytesInMemory += page.fNElements * elementSize;
135 clusters[idx].fBytesOnStorage += page.fLocator.fBytesOnStorage;
136 clusters[idx].fBytesInMemory += page.fNElements * elementSize;
137 ++clusters[idx].fNPages;
138 info.fBytesOnStorage += page.fLocator.fBytesOnStorage;
139 ++info.fNPages;
140 ++nPages;
141 }
142 }
143 columns.emplace_back(info);
144 }
145 auto headerSize = SerializeHeader(nullptr);
146 auto footerSize = SerializeFooter(nullptr);
147 output << "============================================================" << std::endl;
148 output << "NTUPLE: " << GetName() << std::endl;
149 output << "Compression: " << compression << std::endl;
150 output << "------------------------------------------------------------" << std::endl;
151 output << " # Entries: " << GetNEntries() << std::endl;
152 output << " # Fields: " << GetNFields() << std::endl;
153 output << " # Columns: " << GetNColumns() << std::endl;
154 output << " # Pages: " << nPages << std::endl;
155 output << " # Clusters: " << GetNClusters() << std::endl;
156 output << " Size on storage: " << bytesOnStorage << " B" << std::endl;
157 output << " Compression rate: " << std::fixed << std::setprecision(2)
158 << float(bytesInMemory) / float(bytesOnStorage) << std::endl;
159 output << " Header size: " << headerSize << " B" << std::endl;
160 output << " Footer size: " << footerSize << " B" << std::endl;
161 output << " Meta-data / data: " << std::fixed << std::setprecision(3)
162 << float(headerSize + footerSize) / float(bytesOnStorage) << std::endl;
163 output << "------------------------------------------------------------" << std::endl;
164 output << "CLUSTER DETAILS" << std::endl;
165 output << "------------------------------------------------------------" << std::endl;
166
167 std::sort(clusters.begin(), clusters.end());
168 for (unsigned int i = 0; i < clusters.size(); ++i) {
169 output << " # " << std::setw(5) << i
170 << " Entry range: [" << clusters[i].fFirstEntry << ".."
171 << clusters[i].fFirstEntry + clusters[i].fNEntries << ") -- " << clusters[i].fNEntries << std::endl;
172 output << " "
173 << " # Pages: " << clusters[i].fNPages << std::endl;
174 output << " "
175 << " Size on storage: " << clusters[i].fBytesOnStorage << " B" << std::endl;
176 output << " "
177 << " Compression: " << std::fixed << std::setprecision(2)
178 << float(clusters[i].fBytesInMemory) / float(float(clusters[i].fBytesOnStorage)) << std::endl;
179 }
180
181 output << "------------------------------------------------------------" << std::endl;
182 output << "COLUMN DETAILS" << std::endl;
183 output << "------------------------------------------------------------" << std::endl;
184 for (auto &col : columns)
185 col.fFieldName = GetFieldName(col.fFieldId, *this).substr(1);
186 std::sort(columns.begin(), columns.end());
187 for (const auto &col : columns) {
188 auto avgPageSize = (col.fNPages == 0) ? 0 : (col.fBytesOnStorage / col.fNPages);
189 auto avgElementsPerPage = (col.fNPages == 0) ? 0 : (col.fNElements / col.fNPages);
190 output << " " << col.fFieldName << " [#" << col.fLocalOrder << "]" << " -- "
191 << GetColumnTypeName(col.fType) << std::endl;
192 output << " # Elements: " << col.fNElements << std::endl;
193 output << " # Pages: " << col.fNPages << std::endl;
194 output << " Avg elements / page: " << avgElementsPerPage << std::endl;
195 output << " Avg page size: " << avgPageSize << " B" << std::endl;
196 output << " Size on storage: " << col.fBytesOnStorage << " B" << std::endl;
197 output << " Compression: " << std::fixed << std::setprecision(2)
198 << float(col.fElementSize * col.fNElements) / float(col.fBytesOnStorage) << std::endl;
199 output << "............................................................" << std::endl;
200 }
201}
Bool_t operator<(const TDatime &d1, const TDatime &d2)
Definition: TDatime.h:106
Bool_t operator==(const TDatime &d1, const TDatime &d2)
Definition: TDatime.h:102
int type
Definition: TGX11.cxx:120
static RColumnElementBase Generate(EColumnType type)
The available trivial, native content types of a column.
The on-storage meta-data of an ntuple.
std::unordered_map< DescriptorId_t, RClusterDescriptor > fClusterDescriptors
May contain only a subset of all the available clusters, e.g.
std::uint32_t SerializeHeader(void *buffer) const
We deliberately do not use ROOT's built-in serialization in order to allow for use of RNTuple's witho...
std::unordered_map< DescriptorId_t, RColumnDescriptor > fColumnDescriptors
std::uint32_t SerializeFooter(void *buffer) const
Serializes cluster meta data. Returns the number of bytes and fills buffer if it is not nullptr.
const RFieldDescriptor & GetFieldDescriptor(DescriptorId_t fieldId) const
void PrintInfo(std::ostream &output) const
std::uint64_t DescriptorId_t
Distriniguishes elements of the same type within a descriptor, e.g. different fields.
Definition: RNTupleUtil.hxx:78
constexpr DescriptorId_t kInvalidDescriptorId
Definition: RNTupleUtil.hxx:79
static void output(int code)
Definition: gifencode.c:226