mariadb-columnstore-engine/src/Cache.cpp

#include "Cache.h"
#include "Config.h"
#include "Downloader.h"
#include <iostream>
#include <syslog.h>
#include <boost/filesystem.hpp>
#include <boost/thread.hpp>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>

using namespace std;
using namespace boost::filesystem;

namespace storagemanager
{

Cache::Cache() : currentCacheSize(0)
{
    Config *conf = Config::get();
    logger = SMLogging::get();
    sync = Synchronizer::get();
    
    string stmp = conf->getValue("Cache", "cache_size");
    if (stmp.empty()) 
    {
        logger->log(LOG_CRIT, "Cache/cache_size is not set");
        throw runtime_error("Please set Cache/cache_size in the storagemanager.cnf file");
    }
    try
    {
        maxCacheSize = stoul(stmp);
    }
    catch (invalid_argument &)
    {
        logger->log(LOG_CRIT, "Cache/cache_size is not a number");
        throw runtime_error("Please set Cache/cache_size to a number");
    }
    //cout << "Cache got cache size " << maxCacheSize << endl;
        
    stmp = conf->getValue("ObjectStorage", "object_size");
    if (stmp.empty()) 
    {
        logger->log(LOG_CRIT, "ObjectStorage/object_size is not set");
        throw runtime_error("Please set ObjectStorage/object_size in the storagemanager.cnf file");
    }
    try
    {
        objectSize = stoul(stmp);
    }
    catch (invalid_argument &)
    {
        logger->log(LOG_CRIT, "ObjectStorage/object_size is not a number");
        throw runtime_error("Please set ObjectStorage/object_size to a number");
    }
    
    prefix = conf->getValue("Cache", "path");
    if (prefix.empty())
    {
        logger->log(LOG_CRIT, "Cache/path is not set");
        throw runtime_error("Please set Cache/path in the storagemanager.cnf file");
    }
    
    try 
    {
        boost::filesystem::create_directories(prefix);
    }
    catch (exception &e)
    {
        syslog(LOG_CRIT, "Failed to create %s, got: %s", prefix.string().c_str(), e.what());
        throw e;
    }
    //cout << "Cache got prefix " << prefix << endl;
    
    downloader.setDownloadPath(prefix.string());
    /* todo: populate structures with existing files in the cache path */
}

Cache::~Cache()
{
}

void Cache::read(const vector<string> &keys)
{
/*
    move existing keys to a do-not-evict map
    fetch keys that do not exist
    after fetching, move all keys from do-not-evict to the back of the LRU
*/
    boost::unique_lock<boost::mutex> s(lru_mutex);
    vector<const string *> keysToFetch;
    
    uint i;
    M_LRU_t::iterator mit;
    
    for (const string &key : keys)
    {
        mit = m_lru.find(key);
        if (mit != m_lru.end())
            // it's in the cache, add the entry to the do-not-evict set
            addToDNE(mit->lit);
        else
            // not in the cache, put it in the list to download
            keysToFetch.push_back(&key);
    }
    
    // TODO: get the sizes of the objects to download and make space
    // For now using an estimate
    makeSpace(keys.size() * objectSize);
    
    s.unlock();
    
    // start downloading the keys to fetch
    int dl_err;
    vector<int> dl_errnos;
    if (!keysToFetch.empty())
        dl_err = downloader.download(keysToFetch, &dl_errnos);
    
    s.lock();
    
    // move all keys to the back of the LRU
    for (i = 0; i < keys.size(); i++)
    {
        mit = m_lru.find(keys[i]);
        if (mit != m_lru.end())
        {
            lru.splice(lru.end(), lru, mit->lit);
            removeFromDNE(lru.end());
        }
        else if (dl_errnos[i] == 0)   // successful download
        {
            lru.push_back(keys[i]);
            m_lru.insert(M_LRU_element_t(&(lru.back()), lru.end()--));
        }
        else
        {
            // Downloader already logged it, anything to do here?
            /* brainstorming options for handling it.
             1) Should it be handled?  The caller will log a file-not-found error, and there will be a download
                failure in the log already.  
             2) Can't really DO anything can it?
            */
        }
    }
}

Cache::DNEElement::DNEElement(const LRU_t::iterator &k) : key(k), refCount(1)
{
}
        
void Cache::addToDNE(const LRU_t::iterator &key)
{
    DNEElement e(key);
    DNE_t::iterator it = doNotEvict.find(e);
    if (it != doNotEvict.end())
    {
        DNEElement &dnee = const_cast<DNEElement &>(*it);
        ++dnee.refCount;
    }
    else
        doNotEvict.insert(e);
}
        
void Cache::removeFromDNE(const LRU_t::iterator &key)
{
    DNEElement e(key);
    DNE_t::iterator it = doNotEvict.find(e);
    if (it == doNotEvict.end())
        return;
    DNEElement &dnee = const_cast<DNEElement &>(*it);
    if (--dnee.refCount == 0)
        doNotEvict.erase(it);
}
    
const boost::filesystem::path & Cache::getCachePath()
{
    return prefix;
}

        
void Cache::exists(const vector<string> &keys, vector<bool> *out)
{
    out->resize(keys.size());
    boost::unique_lock<boost::mutex> s(lru_mutex);
    for (int i = 0; i < keys.size(); i++)
        (*out)[i] = (m_lru.find(keys[i]) == m_lru.end());
}

void Cache::newObject(const string &key, size_t size)
{
}

void Cache::deletedObject(const string &key, size_t size)
{
}

void Cache::setMaxCacheSize(size_t size)
{
}

// call this holding lru_mutex
void Cache::makeSpace(size_t size)
{
    ssize_t thisMuch = currentCacheSize + size - maxCacheSize;
    if (thisMuch <= 0)
        return;

    struct stat statbuf;
    LRU_t::iterator it = lru.begin();
    while (it != lru.end() && thisMuch > 0)
    {
        if (doNotEvict.find(it) != doNotEvict.end())
        {
            ++it;
            continue;   // it's in the do-not-evict list
        }
        
        boost::filesystem::path cachedFile = prefix / *it;
        int err = stat(cachedFile.string().c_str(), &statbuf);
        if (err)
        {
            logger->log(LOG_WARNING, "Downloader: There seems to be a cached file that couldn't be stat'ed: %s", cachedFile.string().c_str());
            ++it;
            continue;
        }
        
        /* 
            TODO: tell Synchronizer that this key will be evicted
            delete the file
            remove it from our structs
            update current size
        */
        assert(currentCacheSize >= statbuf.st_size);
        currentCacheSize -= statbuf.st_size;
        thisMuch -= statbuf.st_size;
        sync->flushObject(*it);
        boost::filesystem::remove(cachedFile);
        LRU_t::iterator toRemove = it++;
        lru.erase(toRemove);
        m_lru.erase(*toRemove);
    }
}


/* The helper classes */

Cache::M_LRU_element_t::M_LRU_element_t(const string *k) : key(k)
{}

Cache::M_LRU_element_t::M_LRU_element_t(const string &k) : key(&k)
{}

Cache::M_LRU_element_t::M_LRU_element_t(const string *k, const LRU_t::iterator &i) : key(k), lit(i)
{}

inline size_t Cache::KeyHasher::operator()(const M_LRU_element_t &l) const
{
    return hash<string>()(*(l.key));
}

inline bool Cache::KeyEquals::operator()(const M_LRU_element_t &l1, const M_LRU_element_t &l2) const
{
    return (*(l1.key) == *(l2.key));
}

inline size_t Cache::DNEHasher::operator()(const DNEElement &l) const
{
    return hash<string>()(*(l.key));
}

inline bool Cache::DNEEquals::operator()(const DNEElement &l1, const DNEElement &l2) const
{
    return (*(l1.key) == *(l2.key));
}

}