bool TestRunner::concurrencyTest( IDBDataFile::Types filetype )
{
logMsg( INFO, "concurrencyTest" );
reset();
// ok - scenario is a reader opens the file then a reader/writer opens
// the same file and updates something, then the reader tries to read
// again.
m_file = IDBDataFile::open(filetype, m_fname.c_str(), "r", m_open_opts);
assert(m_file);
// read a few blocks
assert( readBlock(0, m_defbuf, 0) );
assert( readBlock(1, m_defbuf, 1) );
// open the same file for read/write
IDBDataFile* rdwrFile = IDBDataFile::open(filetype, m_fname.c_str(), "r+", m_open_opts);
assert(rdwrFile);
// read a block, write a block
assert( rdwrFile->pread(m_defbuf, 0, BLK_SIZE) == (ssize_t) BLK_SIZE );
assert( m_defbuf[0] == 0 );
m_defbuf[0] = 95;
assert( rdwrFile->seek(0, 0) == 0 );
assert( rdwrFile->write(m_defbuf, BLK_SIZE) );
// close file
delete rdwrFile;
// this 5 seconds is important in the HDFS case because it gives HDFS
// time to age off (or whatever) the blocks for the original file that
// have now been rewritten. The value was determined experimentally -
// 3 secs works fine most times but not all. If HDFS hasn't aged out
// the blocks then the read will return the old data
if( filetype == IDBDataFile::HDFS )
sleep(10);
// go back to the reader and make sure he got the new value, then close
assert( readBlock(0, m_defbuf, 95) );
delete m_file;
// now put block 0 back the way it was
m_file = IDBDataFile::open(filetype, m_fname.c_str(), "r+", m_open_opts);
assert(m_file);
assert( writeBlock(0, m_defbuf, 0) );
return true;
}
bool TestRunner::randomReadTest()
{
logMsg( INFO, "randomReadTest" );
struct drand48_data d48data;
srand48_r(pthread_self(), &d48data);
for( int i = 0; i < 10; ++i )
{
long int blk_num;
lrand48_r( &d48data, &blk_num);
blk_num = blk_num % m_opts.numblocks;
unsigned char readbuf[4];
assert( m_file->pread( readbuf, blk_num * BLK_SIZE, 4 ) == 4 );
assert( readbuf[0] == (unsigned char) blk_num );
}
return true;
}
bool TestRunner::readBlock(unsigned int blocknum, unsigned char* buf, unsigned char tag)
{
size_t rc = m_file->pread(buf,blocknum*BLK_SIZE,BLK_SIZE);
//cout << "DEBUG: read " << rc << " bytes at offset " << blocknum * BLK_SIZE << endl;
if (rc != BLK_SIZE)
{
ostringstream errstr;
errstr << "readBlock failed for block " << blocknum << ", read " << rc << " bytes, expecting " << BLK_SIZE;
logMsg( ERROR, errstr.str() );
return false;
}
else if (tag != buf[0])
{
ostringstream errstr;
errstr << "read tag 0x" << setw(2) << hex << setfill('0') << (int) buf[0] << " at block " << dec
<< blocknum << ", expected 0x" << (int) tag;
logMsg( ERROR, errstr.str() );
return false;
}
return true;
}
bool TestRunner::hdfsRdwrExhaustTest()
{
// this is going to be a self-contained test that attempts to test
// all logic inherent in HdfsRdwr
// choose a new filename that is specific to our thread
ostringstream oss;
// embed pid so that this is a new directory path
oss << "/tmp/hdfsrdwr-" << getpid() << "-" << m_id;
string newpath = oss.str();
// open a file with arbitrarily small buffer
IDBDataFile* file = IDBDataFile::open(IDBDataFile::HDFS, newpath.c_str(), "r+", 0, 8);
assert( file );
// check various empty file conditions
assert( file->size() == 0 );
assert( file->tell() == 0 );
assert( file->seek(-1, SEEK_CUR) == -1);
assert( file->seek(0, SEEK_SET) == 0);
unsigned char buf[4];
assert( file->read(buf, 4) == 0);
// write some data
buf[0] = 0xde; buf[1] = 0xad; buf[2] = 0xbe; buf[3] = 0xef;
assert( file->write(buf, 4) == 4);
assert( file->size() == 4 );
assert( file->tell() == 4 );
assert( file->truncate(-1) == -1 );
// now make file empty again
assert( file->truncate(0) == 0 );
assert( file->size() == 0 );
assert( file->seek(0, SEEK_SET) == 0);
assert( file->tell() == 0 );
assert( file->read(buf, 4) == 0);
// write data again, this time exactly up to allocated size
assert( file->write(buf, 4) == 4);
assert( file->write(buf, 4) == 4);
assert( file->size() == 8 );
assert( file->tell() == 8 );
// truncate back to 4
assert( file->truncate(4) == 0 );
assert( file->size() == 4 );
assert( file->seek(4, SEEK_SET) == 0);
assert( file->tell() == 4 );
// now trigger a buffer reallocation
assert( file->write(buf, 4) == 4);
assert( file->write(buf, 4) == 4);
assert( file->size() == 12 );
// now delete and close.
delete file;
// check the file size through the file system
IDBFileSystem& fs = IDBFileSystem::getFs( IDBDataFile::HDFS );
assert( fs.size( newpath.c_str() ) == 12);
// open again - the file is bigger than the default buffer so it triggers alternate
// logic in the constructor
file = IDBDataFile::open(IDBDataFile::HDFS, newpath.c_str(), "r+", 0, 8);
assert( file );
assert( file->size() == 12);
unsigned char newbuf[4];
assert( file->pread(newbuf, 4, 4) == 4);
assert( newbuf[0] == 0xde && newbuf[1] == 0xad && newbuf[2] == 0xbe &&newbuf[3] == 0xef);
delete file;
fs.remove(newpath.c_str());
return true;
}
bool TestRunner::rdwrTest( IDBDataFile::Types filetype )
{
reset();
m_file = IDBDataFile::open(filetype, m_fname.c_str(), "r+", m_open_opts);
if( !m_file )
{
ostringstream errstr;
errstr << "Unable to open " << m_fname << " for reading";
logMsg( ERROR, errstr.str() );
return false;
}
struct drand48_data d48data;
srand48_r(0xdeadbeef, &d48data);
// we will write to 5 random blocks and then come back and
// verify the contents and undo them
int blocks_to_touch = min( 5, m_opts.numBlocks);
vector<int> touched;
for( int i = 0; i < blocks_to_touch; ++i )
{
long int blk_num;
// we need to make sure all the blocks we touch are unique or
// the pattern used by this test won't work
bool found = false;
while(!found)
{
lrand48_r( &d48data, &blk_num);
blk_num = blk_num % m_opts.numBlocks;
vector<int>::iterator pos = find( touched.begin(), touched.end(), blk_num );
if( pos == touched.end())
found = true;
}
if( m_file->seek(blk_num * BLK_SIZE, SEEK_SET) )
{
ostringstream errstr;
errstr << "failed to seek block " << blk_num;
logMsg( ERROR, errstr.str() );
return false;
}
unsigned char writeval = 0xb0;
size_t writect = m_file->write(&writeval,1);
if( writect != 1 )
{
ostringstream errstr;
errstr << "Only wrote " << writect << " bytes, expected 1";
logMsg( ERROR, errstr.str() );
return false;
}
touched.push_back(blk_num);
}
m_file->flush();
for( int i = 0; i < (int) touched.size(); ++i )
{
unsigned char readbuf;
size_t readct = m_file->pread(&readbuf,touched[i] * BLK_SIZE,1);
if( readct != 1 || readbuf != (unsigned char) 0xb0 )
{
ostringstream errstr;
errstr << "Error reading expected value, readct=" << readct << " bytes, value" << (int) readbuf;
logMsg( ERROR, errstr.str() );
return false;
}
readbuf = touched[i];
if( m_file->seek(-1, SEEK_CUR) )
{
ostringstream errstr;
errstr << "failed to seek -1";
logMsg( ERROR, errstr.str() );
return false;
}
size_t writect = m_file->write(&readbuf,1);
if( writect != 1 )
{
ostringstream errstr;
errstr << "Only wrote " << writect << " bytes, expected 1";
logMsg( ERROR, errstr.str() );
return false;
}
}
return true;
}
bool TestRunner::flushTest( IDBDataFile::Types filetype )
{
logMsg( INFO, "flushTest" );
reset();
string scratch = "/tmp/rdwr_scratch" + m_fname; // scratch file name if exists
boost::filesystem::remove(scratch);
IDBPolicy::remove(m_fname.c_str());
// scenario: writer opens the file, writes 8 bytes, flush;
// reader opens the file, reads 8 bytes, verifys the data, then closes file;
// writer writes 8M bytes (force to buffered file) if -m option used correctly;
// reader opens the file, verifys the file size and content, then closes the file;
// writer closes the file.
ostringstream errstr;
m_file = IDBDataFile::open(filetype, m_fname.c_str(), "w+", m_open_opts, /*default:4*/ 1);
if (!m_file)
{
errstr << "flushTest: open " << m_fname.c_str() << " for write failed";
logMsg( ERROR, errstr.str() );
return false;
}
// write 8 "1" through mem buff
const char w1[] = "11111111";
ssize_t bytes = 0;
if ((bytes = m_file->write(w1, 8)) != 8)
{
errstr << "flushTest: write count = 8, return = " << bytes;
logMsg( ERROR, errstr.str() );
return false;
}
// for HDFS, force writing out to disk.
m_file->flush();
if (!IDBPolicy::exists(m_fname.c_str()))
{
errstr << "flushTest: " << m_fname.c_str() << " does not exist";
logMsg( ERROR, errstr.str() );
return false;
}
if (filetype == IDBDataFile::HDFS && boost::filesystem::exists(scratch))
{
errstr << "flushTest: " << scratch << " exists after 1st write";
logMsg( ERROR, errstr.str() );
return false;
}
// read from file in "r" mode
IDBDataFile* file = IDBDataFile::open(filetype, m_fname.c_str(), "r", m_open_opts, 1);
if (!file)
{
errstr << "flushTest: 1st open " << m_fname.c_str() << " to read failed";
logMsg( ERROR, errstr.str() );
return false;
}
char r1[9] = {0};
ssize_t n1 = file->pread(r1, 0, 8);
if (n1 != 8 || strncmp(r1, w1, 8) != 0)
{
errstr << "flushTest: read " << n1 << " != 8 OR " << r1 << "!= 11111111";
logMsg( ERROR, errstr.str() );
return false;
}
delete file;
file = NULL;
// write 8M "2", switched to file buffer if max size for mem buffer is small.
//char w2[] = {[0 ... 8*1024*1024] = '2'};
ssize_t m9 = 9*1024*1024; // must be large than EXTENTSIZE = 8390656 to swith to file buffer
boost::scoped_array<char> w2(new char[m9]);
memset(w2.get(), '2', m9);
m_file->write(w2.get(), m9);
m_file->flush();
// check file size
if (IDBPolicy::size(m_fname.c_str()) != 8 + m9)
{
errstr << "flushTest: size of " << m_fname.c_str() << " is "
<< IDBPolicy::size(m_fname.c_str()) << ", expecting " << (8 + m9);
logMsg( ERROR, errstr.str() );
return false;
}
if (filetype == IDBDataFile::HDFS &&
!boost::filesystem::exists(scratch) &&
m_opts.hdfsMaxMem < m9)
{
errstr << "flushTest: " << scratch << " does not exist after 2nd write";
logMsg( ERROR, errstr.str() );
return false;
}
// 2nd read
file = IDBDataFile::open(filetype, m_fname.c_str(), "r", m_open_opts, 1);
if (!file)
{
errstr << "flushTest: 2nd open " << m_fname.c_str() << " to read failed";
logMsg( ERROR, errstr.str() );
return false;
}
//char r2[9*1024*1024 + 8 + 1] = {0};
boost::scoped_array<char> r2(new char[9*1024*1024 + 8 + 1]);
memset(r2.get(), 0, m9 + 9);
ssize_t n2 = file->pread(r2.get(), 0, m9 + 8);
if (n2 != (m9+8) || strncmp(r2.get(), w1, 8) != 0 || memcmp(r2.get()+8, w2.get(), m9) != 0)
{
errstr << "flushTest: 2nd read " << m_fname.c_str() << " failed" << endl
<< " return value: " << n2 << " bytes -- " << r2; // need hex dump?
logMsg( ERROR, errstr.str() );
return false;
}
delete file;
file = NULL;
delete m_file;
m_file = NULL;
return true;
}
bool TestRunner::seekTest( IDBDataFile::Types filetype )
{
logMsg( INFO, "seekTest" );
reset();
// scenario: writer opens the file with w+, writes 8 bytes, seek to 4 from 0, write 4 bytes
// reader opens the file with r, verify size and contents,
// writer seeks 4 bytes beyond EOF, write 4 bytes, and close the file,
// reader rewinds, verify size and contents, and close the file.
ostringstream errstr;
m_file = IDBDataFile::open(filetype, m_fname.c_str(), "w+", m_open_opts);
if (!m_file)
{
errstr << "seekTest: open " << m_fname.c_str() << " for write failed";
logMsg( ERROR, errstr.str() );
return false;
}
// write 8 "1" through mem buff
const char w1[] = "11111111";
ssize_t bytes = 0;
if ((bytes = m_file->write(w1, 8)) != 8)
{
errstr << "seekTest: write1 count = 8, return = " << bytes;
logMsg( ERROR, errstr.str() );
return false;
}
if (m_file->seek(4, SEEK_SET) != 0)
{
errstr << "seekTest: seek(4, SEEK_SET) failed";
logMsg( ERROR, errstr.str() );
return false;
}
const char w2[] = "2222";
if ((bytes = m_file->write(w2, 4)) != 4)
{
errstr << "seekTest: write2 count = 4, return = " << bytes;
logMsg( ERROR, errstr.str() );
return false;
}
// for HDFS, force writing out to disk.
m_file->flush();
// read from file in "r" mode
IDBDataFile* file = IDBDataFile::open(filetype, m_fname.c_str(), "r", m_open_opts);
if (!file)
{
errstr << "seekTest: 1st open " << m_fname.c_str() << " to read failed";
logMsg( ERROR, errstr.str() );
return false;
}
char r1[9] = {0};
ssize_t n1 = file->pread(r1, 0, 8);
if (IDBPolicy::size(m_fname.c_str()) != 8)
{
errstr << "seekTest: size of " << m_fname.c_str() << " is "
<< IDBPolicy::size(m_fname.c_str()) << ", expecting 8";
logMsg( ERROR, errstr.str() );
return false;
}
if (n1 != 8 || strncmp(r1, w1, 4) != 0 || strncmp(r1+4, w2, 4) != 0)
{
errstr << "seekTest: read " << n1 << " != 8 OR " << r1 << "!= 11112222";
logMsg( ERROR, errstr.str() );
return false;
}
// now seek beyond the eof, and write 4 bytes.
const char w3[] = "3333";
if (m_file->seek(4, SEEK_END) != 0)
{
errstr << "seekTest: seek(4, SEEK_END) failed";
logMsg( ERROR, errstr.str() );
return false;
}
if ((bytes = m_file->write(w3, 4)) != 4)
{
errstr << "seekTest: write3 count = 4, return = " << bytes;
logMsg( ERROR, errstr.str() );
return false;
}
m_file->flush();
delete m_file;
m_file = NULL;
// check file size
if (IDBPolicy::size(m_fname.c_str()) != 16)
{
errstr << "seekTest: size of " << m_fname.c_str() << " is "
<< IDBPolicy::size(m_fname.c_str()) << ", expecting 16";
logMsg( ERROR, errstr.str() );
return false;
}
// 2nd read
file = IDBDataFile::open(filetype, m_fname.c_str(), "r", m_open_opts);
if (!file)
{
errstr << "seekTest: 2nd open " << m_fname.c_str() << " to read failed";
logMsg( ERROR, errstr.str() );
return false;
}
char r2[17] = {0};
const char pd[4] = {0}; // padding
ssize_t n2 = file->pread(r2, 0, 16);
if (n2 != 16 ||
strncmp(r2, w1, 4) != 0 || memcmp(r2+4, w2, 4) != 0 ||
strncmp(r2+8, pd, 4) != 0 || memcmp(r2+12, w3, 4) != 0)
{
errstr << "seekTest: 2nd read " << m_fname.c_str() << " failed" << endl
<< " return value: " << n2 << " bytes -- " << r2; // need hex dump?
logMsg( ERROR, errstr.str() );
return false;
}
delete file;
file = NULL;
return true;
}