23#include <unordered_map>
29 std::uint64_t fFirstEntry = 0;
30 std::uint32_t fNPages = 0;
31 std::uint32_t fNEntries = 0;
32 std::uint32_t fBytesOnStorage = 0;
33 std::uint32_t fBytesInMemory = 0;
35 bool operator==(
const ClusterInfo &other)
const {
return fFirstEntry == other.fFirstEntry; }
37 bool operator<(
const ClusterInfo &other)
const {
return fFirstEntry < other.fFirstEntry; }
44 std::uint64_t fNElements = 0;
45 std::uint64_t fNPages = 0;
46 std::uint64_t fBytesOnStorage = 0;
47 std::uint32_t fElementSize = 0;
48 std::uint32_t fColumnIndex = 0;
49 std::uint16_t fRepresentationIndex = 0;
51 std::string fFieldName;
52 std::string fFieldDescription;
54 bool operator<(
const ColumnInfo &other)
const
56 if (fFieldName == other.fFieldName) {
57 if (fRepresentationIndex == other.fRepresentationIndex)
58 return fColumnIndex < other.fColumnIndex;
59 return fRepresentationIndex < other.fRepresentationIndex;
61 return fFieldName < other.fFieldName;
71 return GetFieldName(fieldDesc.GetParentId(), ntupleDesc) +
"." + fieldDesc.GetFieldName();
85 std::vector<ColumnInfo> columns;
86 std::vector<ClusterInfo> clusters;
87 std::unordered_map<DescriptorId_t, unsigned int> cluster2Idx;
91 info.fFirstEntry = cluster.second.GetFirstEntryIndex();
92 info.fNEntries = cluster.second.GetNEntries();
93 cluster2Idx[cluster.first] = clusters.size();
94 clusters.emplace_back(info);
97 std::uint64_t bytesOnStorage = 0;
98 std::uint64_t bytesInMemory = 0;
99 std::uint64_t nPages = 0;
100 int compression = -1;
104 if (column.second.IsAliasColumn())
112 info.fPhysicalColumnId = column.second.GetPhysicalId();
113 info.fLogicalColumnId = column.second.GetLogicalId();
114 info.fFieldId = column.second.GetFieldId();
115 info.fColumnIndex = column.second.GetIndex();
116 info.fElementSize = elementSize;
117 info.fType = column.second.GetType();
118 info.fRepresentationIndex = column.second.GetRepresentationIndex();
121 auto columnRange = cluster.second.GetColumnRange(column.second.GetPhysicalId());
122 if (columnRange.fIsSuppressed)
125 info.fNElements += columnRange.fNElements;
126 if (compression == -1) {
127 compression = columnRange.fCompressionSettings;
129 const auto &pageRange = cluster.second.GetPageRange(column.second.GetPhysicalId());
130 auto idx = cluster2Idx[cluster.first];
131 for (
const auto &page : pageRange.fPageInfos) {
132 bytesOnStorage += page.fLocator.fBytesOnStorage;
133 bytesInMemory += page.fNElements * elementSize;
134 clusters[idx].fBytesOnStorage += page.fLocator.fBytesOnStorage;
135 clusters[idx].fBytesInMemory += page.fNElements * elementSize;
136 ++clusters[idx].fNPages;
137 info.fBytesOnStorage += page.fLocator.fBytesOnStorage;
142 columns.emplace_back(info);
146 output <<
"============================================================\n";
148 output <<
"Compression: " << compression <<
"\n";
149 output <<
"------------------------------------------------------------\n";
154 output <<
" # Pages: " << nPages <<
"\n";
156 output <<
" Size on storage: " << bytesOnStorage <<
" B"
158 output <<
" Compression rate: " << std::fixed << std::setprecision(2)
159 << float(bytesInMemory) / float(bytesOnStorage) <<
"\n";
160 output <<
" Header size: " << headerSize <<
" B"
162 output <<
" Footer size: " << footerSize <<
" B"
164 output <<
" Meta-data / data: " << std::fixed << std::setprecision(3)
165 << float(headerSize + footerSize) / float(bytesOnStorage) <<
"\n";
166 output <<
"------------------------------------------------------------\n";
167 output <<
"CLUSTER DETAILS\n";
168 output <<
"------------------------------------------------------------" << std::endl;
170 std::sort(clusters.begin(), clusters.end());
171 for (
unsigned int i = 0; i < clusters.size(); ++i) {
172 output <<
" # " << std::setw(5) << i <<
" Entry range: [" << clusters[i].fFirstEntry <<
".."
173 << clusters[i].fFirstEntry + clusters[i].fNEntries - 1 <<
"] -- " << clusters[i].fNEntries <<
"\n";
175 <<
" # Pages: " << clusters[i].fNPages <<
"\n";
177 <<
" Size on storage: " << clusters[i].fBytesOnStorage <<
" B\n";
179 <<
" Compression: " << std::fixed << std::setprecision(2)
180 << float(clusters[i].fBytesInMemory) / float(
float(clusters[i].fBytesOnStorage)) << std::endl;
183 output <<
"------------------------------------------------------------\n";
184 output <<
"COLUMN DETAILS\n";
185 output <<
"------------------------------------------------------------\n";
186 for (
auto &col : columns) {
187 col.fFieldName = GetFieldName(col.fFieldId, *
this).substr(1);
188 col.fFieldDescription = GetFieldDescription(col.fFieldId, *
this);
190 std::sort(columns.begin(), columns.end());
191 for (
const auto &col : columns) {
192 auto avgPageSize = (col.fNPages == 0) ? 0 : (col.fBytesOnStorage / col.fNPages);
193 auto avgElementsPerPage = (col.fNPages == 0) ? 0 : (col.fNElements / col.fNPages);
194 std::string nameAndType = std::string(
" ") + col.fFieldName +
" [#" + std::to_string(col.fColumnIndex);
195 if (col.fRepresentationIndex > 0)
196 nameAndType +=
" / R." + std::to_string(col.fRepresentationIndex);
198 std::string
id = std::string(
"{id:") + std::to_string(col.fLogicalColumnId) +
"}";
199 if (col.fLogicalColumnId != col.fPhysicalColumnId)
200 id +=
" --alias--> " + std::to_string(col.fPhysicalColumnId);
201 output << nameAndType << std::setw(60 - nameAndType.length()) <<
id <<
"\n";
202 if (!col.fFieldDescription.empty())
203 output <<
" Description: " << col.fFieldDescription <<
"\n";
204 output <<
" # Elements: " << col.fNElements <<
"\n";
205 output <<
" # Pages: " << col.fNPages <<
"\n";
206 output <<
" Avg elements / page: " << avgElementsPerPage <<
"\n";
207 output <<
" Avg page size: " << avgPageSize <<
" B\n";
208 output <<
" Size on storage: " << col.fBytesOnStorage <<
" B\n";
209 output <<
" Compression: " << std::fixed << std::setprecision(2)
210 << float(col.fElementSize * col.fNElements) / float(col.fBytesOnStorage) <<
"\n";
211 output <<
"............................................................" << std::endl;
Bool_t operator<(const TDatime &d1, const TDatime &d2)
Bool_t operator==(const TDatime &d1, const TDatime &d2)
The available trivial, native content types of a column.
static std::string GetTypeName(EColumnType type)
static std::unique_ptr< RColumnElementBase > Generate(EColumnType type)
If CppT == void, use the default C++ type for the given column type.
const std::string & GetFieldName() const
const std::string & GetFieldDescription() const
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::size_t GetNLogicalColumns() const
const std::string & GetName() const
std::uint64_t GetOnDiskHeaderSize() const
std::size_t GetNClusters() const
std::unordered_map< DescriptorId_t, RColumnDescriptor > fColumnDescriptors
NTupleSize_t GetNEntries() const
We know the number of entries from adding the cluster summaries.
std::size_t GetNFields() const
std::uint64_t GetOnDiskFooterSize() const
const RFieldDescriptor & GetFieldDescriptor(DescriptorId_t fieldId) const
std::size_t GetNPhysicalColumns() 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.
constexpr DescriptorId_t kInvalidDescriptorId