1
0
mirror of https://github.com/mariadb-corporation/mariadb-columnstore-engine.git synced 2025-07-27 21:01:50 +03:00
Files
Roman Nozdrin 4fe9cd64a3 Revert "No boost condition (#2822)" (#2828)
This reverts commit f916e64927.
2023-04-22 15:49:50 +03:00

292 lines
8.1 KiB
C++

/* Copyright (C) 2014 InfiniDB, Inc.
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License
as published by the Free Software Foundation; version 2 of
the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
MA 02110-1301, USA. */
/******************************************************************************
* $Id: vss.h 1926 2013-06-30 21:18:14Z wweeks $
*
*****************************************************************************/
/** @file
* class XXX interface
*/
#pragma once
#include <set>
//#define NDEBUG
#include <cassert>
#include <boost/thread.hpp>
#include "brmshmimpl.h"
#include "brmtypes.h"
#include "undoable.h"
#include "mastersegmenttable.h"
#include "shmkeys.h"
#include "hasher.h"
#ifdef NONE
#undef NONE
#endif
#ifdef READ
#undef READ
#endif
#ifdef WRITE
#undef WRITE
#endif
// These config parameters need to be loaded
// will get a small hash function performance boost by using powers of 2
#define VSSSTORAGE_INITIAL_COUNT 200000
#define VSSSTORAGE_INITIAL_SIZE (VSSSTORAGE_INITIAL_COUNT * sizeof(VSSEntry))
#define VSSSTORAGE_INCREMENT_COUNT 20000
#define VSSSTORAGE_INCREMENT (VSSSTORAGE_INCREMENT_COUNT * sizeof(VSSEntry))
// (average list length = 4)
#define VSSTABLE_INITIAL_SIZE (50000 * sizeof(int))
#define VSSTABLE_INCREMENT (5000 * sizeof(int))
#define VSS_INITIAL_SIZE (sizeof(VSSShmsegHeader) + VSSSTORAGE_INITIAL_SIZE + VSSTABLE_INITIAL_SIZE)
#define VSS_INCREMENT (VSSTABLE_INCREMENT + VSSSTORAGE_INCREMENT)
#define VSS_SIZE(entries) \
((entries * sizeof(VSSEntry)) + (entries / 4 * sizeof(int)) + sizeof(VSSShmsegHeader))
#define EXPORT
namespace BRM
{
struct VSSEntry
{
LBID_t lbid;
VER_t verID;
bool vbFlag;
bool locked;
int next;
#ifndef __LP64__
uint32_t pad1;
#endif
EXPORT VSSEntry();
};
struct VSSShmsegHeader
{
int capacity;
int currentSize;
int LWM;
int numHashBuckets;
int lockedEntryCount;
// the rest of the overlay looks like this
// int hashBuckets[numHashBuckets];
// VSSEntry storage[capacity];
};
class QueryContext_vss
{
public:
QueryContext_vss() : currentScn(0)
{
txns.reset(new std::set<VER_t>());
}
QueryContext_vss(const QueryContext& qc);
VER_t currentScn;
boost::shared_ptr<std::set<VER_t> > txns;
};
class VSSImpl
{
public:
static VSSImpl* makeVSSImpl(unsigned key, off_t size, bool readOnly = false);
inline void grow(unsigned key, off_t size)
#ifndef NDBUG
{
fVSS.grow(key, size);
}
#else
{
int rc = fVSS.grow(key, size);
idbassert(rc == 0);
}
#endif
inline void makeReadOnly()
{
fVSS.setReadOnly();
}
inline void clear(unsigned key, off_t size)
{
fVSS.clear(key, size);
}
inline void swapout(BRMShmImpl& rhs)
{
fVSS.swap(rhs);
rhs.destroy();
}
inline unsigned key() const
{
return fVSS.key();
}
inline VSSShmsegHeader* get() const
{
return reinterpret_cast<VSSShmsegHeader*>(fVSS.fMapreg.get_address());
}
private:
VSSImpl(unsigned key, off_t size, bool readOnly = false);
~VSSImpl();
VSSImpl(const VSSImpl& rhs);
VSSImpl& operator=(const VSSImpl& rhs);
BRMShmImpl fVSS;
static boost::mutex fInstanceMutex;
static VSSImpl* fInstance;
};
class VBBM;
class ExtentMap;
/** @brief The Version Substitution Structure (VSS)
*
* At a high level, the VSS maintains a table that associates an LBID with
* a version number and 2 flags that indicate whether or not the block
* identified by the <LBID, VerID> pair exists in the main database files
* or the Version Buffer. The VSS's main purpose is to resolve the version of
* a specified block the caller can safely use given that there may be concurrent
* writes to that block.
*
* As implemented, it is a hash table and a set of lists that exist in
* shared memory. The hash table is keyed by LBID, and
* each valid entry points to the head of a unique list. Each list element
* contains the LBID, VerID, & the two flags that encapsulate "an entry in the
* VSS table". Every list contains all elements that collide on that hash table
* entry that points to it, "load factor" has no bearing on performance,
* and lists can grow arbitrarily large.
* Technically lookups are O(n), but in normal circumstances it'll
* be constant time. As things are right now, we expect there to be about
* 200k VSS entries. The hash table is sized such that on average there will be 4
* entries per list when it's at capacity.
*
* The memory management & structure manipulation code is nearly identical
* to that in the VBBM, so any bugs found here are likely there as well.
*
* Shared memory is managed using code similar to the ExtentMap & VBBM. When
* the shared memory segment needs to grow, it is write-locked, a new one
* is created, the contents are reinserted to the new one, the key is
* registered, and the old segment is destroyed when the last reference to it
* is detached.
*/
class VSS : public Undoable
{
public:
enum OPS
{
NONE,
READ,
WRITE
};
EXPORT VSS();
EXPORT ~VSS();
EXPORT bool isLocked(const LBIDRange& l, VER_t txnID = -1) const;
EXPORT void removeEntry(LBID_t lbid, VER_t verID, std::vector<LBID_t>* flushList);
// Note, the use_vbbm switch should be used for unit testing the VSS only
EXPORT void removeEntriesFromDB(const LBIDRange& range, VBBM& vbbm, bool use_vbbm = true);
EXPORT int lookup(LBID_t lbid, const QueryContext_vss&, VER_t txnID, VER_t* outVer, bool* vbFlag,
bool vbOnly = false) const;
/// Returns the version in the main DB files
EXPORT VER_t getCurrentVersion(LBID_t lbid, bool* isLocked) const; // returns the ver in the main DB files
/// Returns the highest version in the version buffer, less than max
EXPORT VER_t getHighestVerInVB(LBID_t lbid, VER_t max) const;
/// returns true if that block is in the version buffer, false otherwise
EXPORT bool isVersioned(LBID_t lbid, VER_t version) const;
EXPORT void setVBFlag(LBID_t lbid, VER_t verID, bool vbFlag);
EXPORT void insert(LBID_t, VER_t, bool vbFlag, bool locked, bool loading = false);
EXPORT void commit(VER_t txnID);
EXPORT void getUncommittedLBIDs(VER_t txnID, std::vector<LBID_t>& lbids);
EXPORT void getUnlockedLBIDs(BlockList_t& lbids);
EXPORT void getLockedLBIDs(BlockList_t& lbids);
EXPORT void lock(OPS op);
EXPORT void release(OPS op);
EXPORT void setReadOnly();
EXPORT int checkConsistency(const VBBM& vbbm, ExtentMap& em) const;
EXPORT int size() const;
EXPORT bool hashEmpty() const;
EXPORT void getCurrentTxnIDs(std::set<VER_t>& txnList) const;
EXPORT void clear();
EXPORT void load(std::string filename);
EXPORT void save(std::string filename);
#ifdef BRM_DEBUG
EXPORT int getShmid() const;
#endif
EXPORT bool isEmpty(bool doLock = true);
/* Bug 2293. VBBM will use this fcn to determine whether a block is
* currently in use. */
EXPORT bool isEntryLocked(LBID_t lbid, VER_t verID) const;
EXPORT bool isTooOld(LBID_t lbid, VER_t verID) const;
private:
VSS(const VSS&);
VSS& operator=(const VSS&);
struct VSSShmsegHeader* vss;
int* hashBuckets;
VSSEntry* storage;
bool r_only;
static boost::mutex mutex; // @bug5355 - made mutex static
key_t currentVSSShmkey;
int vssShmid;
MSTEntry* vssShminfo;
MasterSegmentTable mst;
static const int MAX_IO_RETRIES = 10;
key_t chooseShmkey() const;
void growVSS();
void growForLoad(int count);
void initShmseg();
void copyVSS(VSSShmsegHeader* dest);
int getIndex(LBID_t lbid, VER_t verID, int& prev, int& bucket) const;
void _insert(VSSEntry& e, VSSShmsegHeader* dest, int* destTable, VSSEntry* destStorage,
bool loading = false);
ShmKeys fShmKeys;
VSSImpl* fPVSSImpl;
utils::Hasher hasher;
};
} // namespace BRM
#undef EXPORT