Dictionary Library source: Persistent per-dictionary custom data store (#1917)

* LibraryDictionarySource: Support numeric types, initial support for strings

* Dictionary Library source with string types

* clang-format

* Dictionary Library source: Persistent per-dictionary custom data store

* Fix unused

* Allocate -> New

* Update examples

* fix
This commit is contained in:
proller 2018-02-26 19:57:14 +03:00 committed by alexey-milovidov
parent 1362e0be68
commit 566d66e858
4 changed files with 130 additions and 80 deletions

View File

@ -61,7 +61,7 @@ namespace
if (bracket_pos != std::string::npos && bracket_pos > 0)
key_name = key.substr(0, bracket_pos);
strings.emplace_back(key_name);
strings.emplace_back(config.getString(config_root + '.' + key));
strings.emplace_back(config.getString(config_root + "." + key));
}
return CStringsHolder(strings);
}
@ -118,13 +118,14 @@ LibraryDictionarySource::LibraryDictionarySource(const DictionaryStructure & dic
, context(context)
{
if (!Poco::File(path).exists())
{
throw Exception("LibraryDictionarySource: Can't load lib " + toString() + ": " + Poco::File(path).path() + " - File doesn't exist",
ErrorCodes::FILE_DOESNT_EXIST);
}
description.init(sample_block);
library = std::make_shared<SharedLibrary>(path);
settings = std::make_shared<CStringsHolder>(getLibSettings(config, config_prefix + lib_config_settings));
auto fptr = library->tryGet<void * (*)()>("ClickHouseDictionary_v2_libNew");
if (fptr)
lib_data = fptr();
}
LibraryDictionarySource::LibraryDictionarySource(const LibraryDictionarySource & other)
@ -137,9 +138,17 @@ LibraryDictionarySource::LibraryDictionarySource(const LibraryDictionarySource &
, library{other.library}
, description{other.description}
, settings{other.settings}
, lib_data{other.lib_data}
{
}
LibraryDictionarySource::~LibraryDictionarySource()
{
auto fptr = library->tryGet<void * (*)(void *)>("ClickHouseDictionary_v2_libDelete");
if (fptr)
fptr(lib_data);
}
BlockInputStreamPtr LibraryDictionarySource::loadAll()
{
LOG_TRACE(log, "loadAll " + toString());
@ -155,13 +164,13 @@ BlockInputStreamPtr LibraryDictionarySource::loadAll()
}
void * data_ptr = nullptr;
/// Get function pointer before dataAllocate call because library->get may throw.
/// Get function pointer before dataNew call because library->get may throw.
auto fptr
= library->get<void * (*)(decltype(data_ptr), decltype(&settings->strings), decltype(&columns))>("ClickHouseDictionary_v2_loadAll");
data_ptr = library->get<void * (*)()>("ClickHouseDictionary_v2_dataAllocate")();
data_ptr = library->get<void * (*)(void *)>("ClickHouseDictionary_v2_dataNew")(lib_data);
auto data = fptr(data_ptr, &settings->strings, &columns);
auto block = dataToBlock(description.sample_block, data);
library->get<void (*)(void *)>("ClickHouseDictionary_v2_dataDelete")(data_ptr);
library->get<void (*)(void *, void *)>("ClickHouseDictionary_v2_dataDelete")(lib_data, data_ptr);
return std::make_shared<OneBlockInputStream>(block);
}
@ -181,13 +190,13 @@ BlockInputStreamPtr LibraryDictionarySource::loadIds(const std::vector<UInt64> &
}
void * data_ptr = nullptr;
/// Get function pointer before dataAllocate call because library->get may throw.
/// Get function pointer before dataNew call because library->get may throw.
auto fptr = library->get<void * (*)(decltype(data_ptr), decltype(&settings->strings), decltype(&columns_pass), decltype(&ids_data))>(
"ClickHouseDictionary_v2_loadIds");
data_ptr = library->get<void * (*)()>("ClickHouseDictionary_v2_dataAllocate")();
data_ptr = library->get<void * (*)(void *)>("ClickHouseDictionary_v2_dataNew")(lib_data);
auto data = fptr(data_ptr, &settings->strings, &columns_pass, &ids_data);
auto block = dataToBlock(description.sample_block, data);
library->get<void (*)(void * data_ptr)>("ClickHouseDictionary_v2_dataDelete")(data_ptr);
library->get<void (*)(void *, void *)>("ClickHouseDictionary_v2_dataDelete")(lib_data, data_ptr);
return std::make_shared<OneBlockInputStream>(block);
}
@ -208,30 +217,31 @@ BlockInputStreamPtr LibraryDictionarySource::loadKeys(const Columns & key_column
ext::bit_cast<decltype(ClickHouseLibrary::VectorUInt64::data)>(requested_rows.data()), requested_rows.size()};
void * data_ptr = nullptr;
/// Get function pointer before dataAllocate call because library->get may throw.
/// Get function pointer before dataNew call because library->get may throw.
auto fptr
= library->get<void * (*)(decltype(data_ptr), decltype(&settings->strings), decltype(&columns_pass), decltype(&requested_rows_c))>(
"ClickHouseDictionary_v2_loadKeys");
data_ptr = library->get<void * (*)()>("ClickHouseDictionary_v2_dataAllocate")();
data_ptr = library->get<void * (*)(void *)>("ClickHouseDictionary_v2_dataNew")(lib_data);
auto data = fptr(data_ptr, &settings->strings, &columns_pass, &requested_rows_c);
auto block = dataToBlock(description.sample_block, data);
library->get<void (*)(void * data_ptr)>("ClickHouseDictionary_v2_dataDelete")(data_ptr);
library->get<void (*)(void *, void *)>("ClickHouseDictionary_v2_dataDelete")(lib_data, data_ptr);
return std::make_shared<OneBlockInputStream>(block);
}
bool LibraryDictionarySource::isModified() const
{
auto fptr = library->tryGet<void * (*)(decltype(&settings->strings))>("ClickHouseDictionary_v2_isModified");
auto fptr = library->tryGet<void * (*)(decltype(lib_data), decltype(&settings->strings))>("ClickHouseDictionary_v2_isModified");
if (fptr)
return fptr(&settings->strings);
return fptr(lib_data, &settings->strings);
return true;
}
bool LibraryDictionarySource::supportsSelectiveLoad() const
{
auto fptr = library->tryGet<void * (*)(decltype(&settings->strings))>("ClickHouseDictionary_v2_supportsSelectiveLoad");
auto fptr
= library->tryGet<void * (*)(decltype(lib_data), decltype(&settings->strings))>("ClickHouseDictionary_v2_supportsSelectiveLoad");
if (fptr)
return fptr(&settings->strings);
return fptr(lib_data, &settings->strings);
return true;
}

View File

@ -24,7 +24,7 @@ class CStringsHolder;
/// Allows loading dictionaries from dynamic libraries (.so)
/// Experimental version
/// Now supports only uint64 types
/// Example: dbms/tests/external_dictionaries/dictionary_library/dictionary_library.cpp
class LibraryDictionarySource final : public IDictionarySource
{
public:
@ -36,6 +36,8 @@ public:
LibraryDictionarySource(const LibraryDictionarySource & other);
~LibraryDictionarySource();
BlockInputStreamPtr loadAll() override;
BlockInputStreamPtr loadIds(const std::vector<UInt64> & ids) override;
@ -63,5 +65,6 @@ private:
SharedLibraryPtr library;
ExternalResultDescription description;
std::shared_ptr<CStringsHolder> settings;
void * lib_data = nullptr;
};
}

View File

@ -1,5 +1,7 @@
/// c++ sample dictionary library
/// proller: TODO: describe
#include <cstdint>
#include <iostream>
#include <memory>
@ -10,13 +12,46 @@
//#define DUMPS(VAR) #VAR " = " << VAR
//#define DUMP(VAR) std::cerr << __FILE__ << ":" << __LINE__ << " " << DUMPS(VAR) << "\n";
struct LibHolder
{
//Some your data, maybe service connection
};
struct DataHolder
{
std::vector<std::vector<uint64_t>> vector;
std::unique_ptr<ClickHouseLibrary::VectorUInt64[]> columnsHolder;
ClickHouseLibrary::ColumnsUInt64 columns;
std::vector<std::vector<uint64_t>> dataHolder; // Actual data storage
std::vector<std::vector<ClickHouseLibrary::Field>> fieldHolder; // Pointers and sizes of data
std::unique_ptr<ClickHouseLibrary::Row[]> rowHolder;
ClickHouseLibrary::Table ctable; // Result data prepared for transfer via c-style interface
LibHolder * lib = nullptr;
};
void MakeColumnsFromVector(DataHolder * ptr)
{
for (const auto & row : ptr->dataHolder)
{
std::vector<ClickHouseLibrary::Field> fields;
for (const auto & field : row)
fields.push_back({&field, sizeof(field)});
ptr->fieldHolder.push_back(fields);
}
const auto rows_num = ptr->fieldHolder.size();
ptr->rowHolder = std::make_unique<ClickHouseLibrary::Row[]>(rows_num);
size_t i = 0;
for (auto & row : ptr->fieldHolder)
{
ptr->rowHolder[i].size = row.size();
ptr->rowHolder[i].data = row.data();
++i;
}
ptr->ctable.size = rows_num;
ptr->ctable.data = ptr->rowHolder.get();
}
extern "C" {
void * ClickHouseDictionary_v2_loadIds(void * data_ptr,
@ -29,6 +64,9 @@ void * ClickHouseDictionary_v2_loadIds(void * data_ptr,
if (ids)
std::cerr << "loadIds lib call ptr=" << data_ptr << " => " << ptr << " size=" << ids->size << "\n";
if (!ptr)
return nullptr;
if (settings)
{
std::cerr << "settings passed: " << settings->size << "\n";
@ -42,42 +80,27 @@ void * ClickHouseDictionary_v2_loadIds(void * data_ptr,
for (size_t i = 0; i < columns->size; ++i)
std::cerr << "column " << i << " :" << columns->data[i] << "\n";
}
if (ptr)
{
if (ids)
{
std::cerr << "ids passed: " << ids->size << "\n";
for (size_t i = 0; i < ids->size; ++i)
{
std::cerr << "id " << i << " :" << ids->data[i] << " replying.\n";
ptr->vector.emplace_back(std::vector<uint64_t>{ids->data[i], ids->data[i] + 1, (1 + ids->data[i]) * 10, 65});
}
}
ptr->columnsHolder = std::make_unique<ClickHouseLibrary::VectorUInt64[]>(ptr->vector.size());
size_t i = 0;
for (auto & col : ptr->vector)
if (ids)
{
std::cerr << "ids passed: " << ids->size << "\n";
for (size_t i = 0; i < ids->size; ++i)
{
//DUMP(i);
//DUMP(col);
ptr->columnsHolder[i].size = col.size();
ptr->columnsHolder[i].data = col.data();
++i;
std::cerr << "id " << i << " :" << ids->data[i] << " replying.\n";
ptr->dataHolder.emplace_back(std::vector<uint64_t>{ids->data[i], ids->data[i] + 1, (1 + ids->data[i]) * 10, 65});
}
ptr->columns.size = ptr->vector.size();
//DUMP(ptr->columns.size);
ptr->columns.data = ptr->columnsHolder.get();
//DUMP(ptr->columns.columns);
return static_cast<void *>(&ptr->columns);
}
return nullptr;
MakeColumnsFromVector(ptr);
return static_cast<void *>(&ptr->ctable);
}
void * ClickHouseDictionary_v2_loadAll(void * data_ptr, ClickHouseLibrary::CStrings * settings, ClickHouseLibrary::CStrings * /*columns*/)
{
auto ptr = static_cast<DataHolder *>(data_ptr);
std::cerr << "loadAll lib call ptr=" << data_ptr << " => " << ptr << "\n";
if (!ptr)
return nullptr;
if (settings)
{
std::cerr << "settings passed: " << settings->size << "\n";
@ -87,31 +110,15 @@ void * ClickHouseDictionary_v2_loadAll(void * data_ptr, ClickHouseLibrary::CStri
}
}
if (ptr)
for (size_t i = 0; i < 7; ++i)
{
for (size_t i = 0; i < 7; ++i)
{
std::cerr << "id " << i << " :"
<< " generating.\n";
ptr->vector.emplace_back(std::vector<uint64_t>{i, i + 1, (1 + i) * 10, 65});
}
ptr->columnsHolder = std::make_unique<ClickHouseLibrary::VectorUInt64[]>(ptr->vector.size());
size_t i = 0;
for (auto & col : ptr->vector)
{
ptr->columnsHolder[i].size = col.size();
ptr->columnsHolder[i].data = col.data();
++i;
}
ptr->columns.size = ptr->vector.size();
//DUMP(ptr->columns.size);
ptr->columns.data = ptr->columnsHolder.get();
//DUMP(ptr->columns.columns);
return static_cast<void *>(&ptr->columns);
std::cerr << "id " << i << " :"
<< " generating.\n";
ptr->dataHolder.emplace_back(std::vector<uint64_t>{i, i + 1, (1 + i) * 10, 65});
}
//return;
return nullptr;
MakeColumnsFromVector(ptr);
return static_cast<void *>(&ptr->ctable);
}
void * ClickHouseDictionary_v2_loadKeys(void * data_ptr,
@ -146,16 +153,32 @@ void * ClickHouseDictionary_v2_loadKeys(void * data_ptr,
}
}
//MakeColumnsFromVector(ptr);
return nullptr;
}
void * ClickHouseDictionary_v2_dataAllocate()
void * ClickHouseDictionary_v2_libNew()
{
auto lib_ptr = new LibHolder;
return lib_ptr;
}
void ClickHouseDictionary_v2_libDelete(void * lib_ptr)
{
auto ptr = static_cast<LibHolder *>(lib_ptr);
delete ptr;
return;
}
void * ClickHouseDictionary_v2_dataNew(void * lib_ptr)
{
auto data_ptr = new DataHolder;
data_ptr->lib = static_cast<decltype(data_ptr->lib)>(lib_ptr);
return data_ptr;
}
void ClickHouseDictionary_v2_dataDelete(void * data_ptr)
void ClickHouseDictionary_v2_dataDelete(void * /*lib_ptr*/, void * data_ptr)
{
auto ptr = static_cast<DataHolder *>(data_ptr);
delete ptr;

View File

@ -19,39 +19,53 @@ typedef struct
CString * data;
} ClickHouseLibCStrings;
void * ClickHouseDictionary_v1_loadIds(void * data_ptr, ClickHouseLibCStrings * settings, ClickHouseLibCStrings * columns, ClickHouseLibVectorUInt64 * ids)
void * ClickHouseDictionary_v2_loadIds(
void * data_ptr, ClickHouseLibCStrings * settings, ClickHouseLibCStrings * columns, ClickHouseLibVectorUInt64 * ids)
{
printf("loadIds c lib call ptr=%p size=%" PRIu64 "\n", data_ptr, ids->size);
return 0;
}
void * ClickHouseDictionary_v1_loadAll(void * data_ptr, ClickHouseLibCStrings * settings, ClickHouseLibCStrings * columns)
void * ClickHouseDictionary_v2_loadAll(void * data_ptr, ClickHouseLibCStrings * settings, ClickHouseLibCStrings * columns)
{
printf("loadAll c lib call ptr=%p \n", data_ptr);
return 0;
}
void * ClickHouseDictionary_v1_loadKeys(void * data_ptr,
ClickHouseLibCStrings * settings,
ClickHouseLibCStrings * columns,
const ClickHouseLibVectorUInt64 * requested_rows)
void * ClickHouseDictionary_v2_loadKeys(
void * data_ptr, ClickHouseLibCStrings * settings, ClickHouseLibCStrings * columns, const ClickHouseLibVectorUInt64 * requested_rows)
{
printf("loadKeys c lib call ptr=%p size=%" PRIu64 "\n", data_ptr, requested_rows->size);
return 0;
}
void * ClickHouseDictionary_v2_libNew()
{
int size = 101;
void * lib_ptr = malloc(size);
printf("libNew c lib call lib_ptr=%p \n", lib_ptr);
return lib_ptr;
}
void * ClickHouseDictionary_v1_dataAllocate()
void ClickHouseDictionary_v2_libDelete(void * lib_ptr)
{
printf("libDelete c lib call lib_ptr=%p \n", lib_ptr);
free(lib_ptr);
return;
}
void * ClickHouseDictionary_v2_dataNew(void * lib_ptr)
{
int size = 100;
void * data_ptr = malloc(size);
printf("dataAllocate c lib call ptr=%p \n", data_ptr);
printf("dataNew c lib call lib_ptr=%p data_ptr=%p \n", lib_ptr, data_ptr);
return data_ptr;
}
void ClickHouseDictionary_v1_dataDelete(void * data_ptr)
void ClickHouseDictionary_v2_dataDelete(void * lib_ptr, void * data_ptr)
{
printf("dataDelete c lib call ptr=%p \n", data_ptr);
printf("dataDelete c lib call lib_ptr=%p data_ptr=%p \n", lib_ptr, data_ptr);
free(data_ptr);
return;
}