int main(int argc, char **argv) {
if (argc < 2) {
print_usage(argv[0]);
return 1;
}
if (strcmp(argv[1], "--h") == 0 || strcmp(argv[1], "--help") == 0 ) {
print_usage(argv[0]);
return 1;
}
char filepath[PATH_MAX];
char dir[PATH_MAX];
strcpy(filepath, argv[1]);
strcpy(dir, dirname(filepath));
char filename[PATH_MAX];
strcpy(filepath, argv[1]);
strcpy(filename, basename(filepath));
// initialize library
g_mem.init();
hashinit();
g_conf.init(NULL);
BigFile bigFile;
bigFile.set(dir, filename);
RdbBuckets buckets;
if (starts_with(filename, "posdb")) {
buckets.set(Posdb::getFixedDataSize(), g_conf.m_posdbMaxTreeMem, "buckets-posdb", RDB_POSDB, "posdb", Posdb::getKeySize());
if (!buckets.fastLoad(&bigFile, "posdb")) {
fprintf(stdout, "Unable to load bucket\n");
return 1;
}
buckets.printBuckets(printRecord);
} else {
fprintf(stdout, "Unsupported rdb type\n");
return 1;
}
return 0;
}
void Indexdb::deepVerify ( char *coll ) {
log ( LOG_INFO, "db: Deep Verifying Indexdb for coll %s...", coll );
g_threads.disableThreads();
Msg5 msg5;
Msg5 msg5b;
RdbList list;
key_t startKey;
key_t endKey;
startKey.setMin();
endKey.setMax();
//long minRecSizes = 64000;
collnum_t collnum = g_collectiondb.getCollnum(coll);
RdbBase *rdbBase = g_indexdb.m_rdb.getBase(collnum);
long numFiles = rdbBase->getNumFiles();
long currentFile = 0;
deepLoop:
// done after scanning all files
if ( currentFile >= numFiles ) {
g_threads.enableThreads();
log ( LOG_INFO, "db: Finished deep verify for %li files.",
numFiles );
return;
}
// scan this file
if ( ! msg5.getList ( RDB_INDEXDB ,
coll ,
&list ,
startKey ,
endKey ,
64000 , // minRecSizes ,
true , // includeTree ,
false , // add to cache?
0 , // max cache age
currentFile , // startFileNum ,
1 , // numFiles ,
NULL , // state
NULL , // callback
0 , // niceness
false , // err correction?
NULL ,
0 ,
-1 ,
true ,
-1LL ,
&msg5b ,
false )) {
g_threads.enableThreads();
log("db: HEY! it did not block");
return;
}
long count = 0;
long got = 0;
for ( list.resetListPtr() ; ! list.isExhausted() ;
list.skipCurrentRecord() ) {
key_t k = list.getCurrentKey();
count++;
//unsigned long groupId = k.n1 & g_hostdb.m_groupMask;
unsigned long groupId = getGroupId ( RDB_INDEXDB , &k );
if ( groupId == g_hostdb.m_groupId ) got++;
}
if ( got != count ) {
BigFile *f = rdbBase->getFile(currentFile);
log ("db: File %s: Out of first %li records in indexdb, "
"only %li belong to our group.",
f->getFilename(),count,got );
}
//else
// log ( LOG_INFO, "db: File %li: Indexdb passed verification "
// "successfully for %li recs.",currentFile,count );
// next file
currentFile++;
goto deepLoop;
}
// . but now that we may get a list remotely to fix data corruption,
// this may indeed block
bool Msg3::doneScanning ( ) {
QUICKPOLL(m_niceness);
// . did we have any error on any scan?
// . if so, repeat ALL of the scans
g_errno = m_errno;
// 2 retry is the default
int32_t max = 2;
// see if explicitly provided by the caller
if ( m_maxRetries >= 0 ) max = m_maxRetries;
// now use -1 (no max) as the default no matter what
max = -1;
// ENOMEM is particulary contagious, so watch out with it...
if ( g_errno == ENOMEM && m_maxRetries == -1 ) max = 0;
// msg0 sets maxRetries to 2, don't let max stay set to -1
if ( g_errno == ENOMEM && m_maxRetries != -1 ) max = m_maxRetries;
// when thread cannot alloc enough read buf it keeps the read buf
// set to NULL and BigFile.cpp sets g_errno to EBUFTOOSMALL
if ( g_errno == EBUFTOOSMALL && m_maxRetries == -1 ) max = 0;
// msg0 sets maxRetries to 2, don't let max stay set to -1
if ( g_errno == EBUFTOOSMALL && m_maxRetries != -1 ) max = m_maxRetries;
// . if no thread slots available, that hogs up serious memory.
// the size of Msg3 is 82k, so having just 5000 of them is 430MB.
// . i just made Msg3 alloc mem when it needs more than about 2k
// so this problem is greatly reduced, therefore let's keep
// retrying... forever if no thread slots in thread queue since
// we become the thread queue in a way.
if ( g_errno == ENOTHREADSLOTS ) max = -1;
// this is set above if the map has the same consecutive key repeated
// and the read is enormous
if ( g_errno == ECORRUPTDATA ) max = 0;
// usually bad disk failures, don't retry those forever
//if ( g_errno == EIO ) max = 3;
// no, now our hitachis return these even when they're good so
// we have to keep retrying forever
if ( g_errno == EIO ) max = -1;
// count these so we do not take drives offline just because
// kernel ring buffer complains...
if ( g_errno == EIO ) g_numIOErrors++;
// bail early on high priority reads for these errors
if ( g_errno == EDISKSTUCK && m_niceness == 0 ) max = 0;
if ( g_errno == EIO && m_niceness == 0 ) max = 0;
// how does this happen? we should never bail out on a low priority
// disk read... we just wait for it to complete...
if ( g_errno == EDISKSTUCK && m_niceness != 0 ) { char *xx=NULL;*xx=0;}
// on I/O, give up at call it corrupt after a while. some hitachis
// have I/O errros on little spots, like gk88, maybe we can fix him
if ( g_errno == EIO && m_retryNum >= 5 ) {
m_errno = ECORRUPTDATA;
m_hadCorruption = true;
// do not do any retries any more
max = 0;
}
// convert m_errno to ECORRUPTDATA if it is EBUFTOOSMALL and the
// max of the bytesToRead are over 500MB.
// if bytesToRead was ludicrous, then assume that the data file
// was corrupted, the map was regenerated and it patched
// over the corrupted bits which were 500MB or more in size.
// we cannot practically allocate that much, so let's just
// give back an empty buffer. treat it like corruption...
// the way it patches is to store the same key over all the corrupted
// pages, which can get pretty big. so if you read a range with that
// key you will be hurting!!
// this may be the same scenario as when the rdbmap has consecutive
// same keys. see above where we set m_errno to ECORRUPTDATA...
if ( g_errno == EBUFTOOSMALL ) {
int32_t biggest = 0;
for ( int32_t i = 0 ; i < m_numFileNums ; i++ ) {
if ( m_scans[i].m_bytesToRead < biggest ) continue;
biggest = m_scans[i].m_bytesToRead;
}
if ( biggest > 500000000 ) {
log("db: Max read size was %" PRId32" > 500000000. Assuming "
"corrupt data in data file.",biggest);
m_errno = ECORRUPTDATA;
m_hadCorruption = true;
// do not do any retries on this, the read was > 500MB
max = 0;
}
}
// if shutting down gb then limit to 20 so we can shutdown because
// it can't shutdown until all threads are out of the queue i think
if ( g_process.m_mode == EXIT_MODE && max < 0 ) {
//log("msg3: forcing retries to 0 because shutting down");
max = 0;
}
// get base, returns NULL and sets g_errno to ENOCOLLREC on error
RdbBase *base = getRdbBase( m_rdbId, m_collnum );
if ( ! base ) {
return true;
}
// this really slows things down because it blocks the cpu so
// leave it out for now
#ifdef GBSANITYCHECK
// check for corruption here, do not do it again in Msg5 if we pass
if ( ! g_errno ) { // && g_conf.m_doErrorCorrection ) {
int32_t i;
for ( i = 0 ; i < m_numFileNums ; i++ )
if ( ! m_lists[i].checkList_r ( false, false ) ) break;
if ( i < m_numFileNums ) {
g_errno = ECORRUPTDATA;
m_errno = ECORRUPTDATA;
max = g_conf.m_corruptRetries; // try 100 times
log("db: Encountered corrupt list in file %s.",
base->getFile(m_fileNums[i])->getFilename());
}
else
m_listsChecked = true;
}
#endif
// try to fix this error i've seen
if ( g_errno == EBADENGINEER && max == -1 )
max = 100;
// . if we had a ETRYAGAIN error, then try again now
// . it usually means the whole file or a part of it was deleted
// before we could finish reading it, so we should re-read all now
// . RdbMerge deletes BigFiles after it merges them and also chops
// off file heads
// . now that we have threads i'd imagine we'd get EBADFD or something
// . i've also seen "illegal seek" as well
if ( m_errno && (m_retryNum < max || max < 0) &&
// this will complete in due time, we can't call a sleep wrapper
// on it because the read is really still pending...
m_errno != EDISKSTUCK ) {
// print the error
static time_t s_time = 0;
time_t now = getTime();
if ( now - s_time > 5 || g_errno != ENOTHREADSLOTS ) {
log("net: Had error reading %s: %s. Retrying. "
"(retry #%" PRId32")",
base->m_dbname,mstrerror(m_errno) , m_retryNum );
s_time = now;
}
// send email alert if in an infinite loop, but don't send
// more than once every 2 hours
static int32_t s_lastSendTime = 0;
if ( m_retryNum == 100 && getTime() - s_lastSendTime > 3600*2){
// remove this for now it is going off all the time
//g_pingServer.sendEmail(NULL,//g_hostdb.getMyHost(),
// "100 read retries",true);
s_lastSendTime = getTime();
}
// clear g_errno cuz we should for call to readList()
g_errno = 0;
// free the list buffer since if we have 1000 Msg3s retrying
// it will totally use all of our memory
for ( int32_t i = 0 ; i < m_numChunks ; i++ )
m_lists[i].destructor();
// count retries
m_retryNum++;
// backoff scheme, wait 100ms more each time
int32_t wait ;
if ( m_retryNum == 1 ) wait = 10;
else wait = 200 * m_retryNum;
// . don't wait more than 10 secs between tries
// . i've seen gf0 and gf16 get mega saturated
if ( wait > 10000 ) wait = 10000;
// wait 500 ms
if ( g_loop.registerSleepCallback ( wait , // ms
this ,
doneSleepingWrapper3,
m_niceness))
return false;
// otherwise, registration failed
log(
"net: Failed to register sleep callback for retry. "
"Abandoning read. This is bad.");
// return, g_errno should be set
g_errno = EBUFTOOSMALL;
m_errno = EBUFTOOSMALL;
return true;
}
// if we got an error and should not retry any more then give up
if ( g_errno ) {
log(
"net: Had error reading %s: %s. Giving up after %" PRId32" "
"retries.",
base->m_dbname,mstrerror(g_errno) , m_retryNum );
return true;
}
// note it if the retry finally worked
if ( m_retryNum > 0 )
log(LOG_INFO,"disk: Read succeeded after retrying %" PRId32" times.",
(int32_t)m_retryNum);
// count total bytes for logging
int32_t count = 0;
// . constrain all lists to make merging easier
// . if we have only one list, then that's nice cuz the constrain
// will allow us to send it right away w/ zero copying
// . if we have only 1 list, it won't be merged into a final list,
// that is, we'll just set m_list = &m_lists[i]
for ( int32_t i = 0 ; i < m_numFileNums ; i++ ) {
QUICKPOLL(m_niceness);
// count total bytes for logging
count += m_lists[i].getListSize();
// . hint offset is relative to the offset of first key we read
// . if that key was only 6 bytes RdbScan shift the list buf
// down 6 bytes to make the first key 12 bytes... a
// requirement for all RdbLists
// . don't inc it, though, if it was 0, pointing to the start
// of the list because our shift won't affect that
if ( m_scans[i].m_shifted == 6 && m_hintOffsets[i] > 0 )
m_hintOffsets[i] += 6;
// posdb double compression
if ( m_scans[i].m_shifted == 12 && m_hintOffsets[i] > 0 )
m_hintOffsets[i] += 12;
// . don't constrain on minRecSizes here because it may
// make our endKey smaller, which will cause problems
// when Msg5 merges these lists.
// . If all lists have different endKeys RdbList's merge
// chooses the min and will merge in recs beyond that
// causing a bad list BECAUSE we don't check to make
// sure that recs we are adding are below the endKey
// . if we only read from one file then constrain based
// on minRecSizes so we can send the list back w/o merging
// OR if just merging with RdbTree's list
int32_t mrs ;
// . constrain to m_minRecSizesOrig, not m_minRecSizes cuz
// that could be adjusted by compensateForNegativeRecs()
// . but, really, they should be the same if we only read from
// the root file
if ( m_numFileNums == 1 ) mrs = m_minRecSizesOrig;
else mrs = -1;
// . this returns false and sets g_errno on error
// . like if data is corrupt
BigFile *ff = base->getFile(m_fileNums[i]);
// if we did a merge really quick and delete one of the
// files we were reading, i've seen 'ff' be NULL
char *filename = "lostfilename";
if ( ff ) filename = ff->getFilename();
// compute cache info
RdbCache *rpc = getDiskPageCache ( m_rdbId );
if ( ! m_allowPageCache ) rpc = NULL;
int64_t vfd ;
if ( ff ) vfd = ff->getVfd();
key192_t ck ;
if ( ff )
ck = makeCacheKey ( vfd ,
m_scans[i].m_offset ,
m_scans[i].m_bytesToRead );
if ( m_validateCache && ff && rpc && vfd != -1 ) {
bool inCache;
char *rec; int32_t recSize;
inCache = rpc->getRecord ( (collnum_t)0 , // collnum
(char *)&ck ,
&rec ,
&recSize ,
true , // copy?
-1 , // maxAge, none
true ); // inccounts?
if ( inCache &&
// 1st byte is RdbScan::m_shifted
( m_lists[i].m_listSize != recSize-1 ||
memcmp ( m_lists[i].m_list , rec+1,recSize-1) ||
*rec != m_scans[i].m_shifted ) ) {
log("msg3: cache did not validate");
char *xx=NULL;*xx=0;
}
mfree ( rec , recSize , "vca" );
}
///////
//
// STORE IN PAGE CACHE
//
///////
// store what we read in the cache. don't bother storing
// if it was a retry, just in case something strange happened.
// store pre-constrain call is more efficient.
if ( m_retryNum<=0 && ff && rpc && vfd != -1 &&
! m_scans[i].m_inPageCache )
rpc->addRecord ( (collnum_t)0 , // collnum
(char *)&ck ,
// rec1 is this little thingy
&m_scans[i].m_shifted,
1,
// rec2
m_lists[i].getList() ,
m_lists[i].getListSize() ,
0 ); // timestamp. 0 = now
QUICKPOLL(m_niceness);
// if from our 'page' cache, no need to constrain
if ( ! m_lists[i].constrain ( m_startKey ,
m_constrainKey , // m_endKey
mrs , // m_minRecSizes
m_hintOffsets[i] ,
//m_hintKeys [i] ,
&m_hintKeys [i*m_ks] ,
filename,//ff->getFilename() ,
m_niceness ) ) {
log("net: Had error while constraining list read from "
"%s: %s/%s. vfd=%" PRId32" parts=%" PRId32". "
"This is likely caused by corrupted "
"data on disk.",
mstrerror(g_errno), ff->getDir(),
ff->getFilename(), ff->m_vfd ,
(int32_t)ff->m_numParts );
continue;
}
}
// print the time
if ( g_conf.m_logTimingDb ) {
int64_t now = gettimeofdayInMilliseconds();
int64_t took = now - m_startTime;
log(LOG_TIMING,
"net: Took %" PRId64" ms to read %" PRId32" lists of %" PRId32" bytes total"
" from %s (niceness=%" PRId32").",
took,m_numFileNums,count,base->m_dbname,m_niceness);
}
return true;
}
// . return false if blocked, true otherwise
// . set g_errno on error
// . read list of keys in [startKey,endKey] range
// . read at least "minRecSizes" bytes of keys in that range
// . the "m_endKey" of resulting, merged list may have a smaller endKey
// than the argument, "endKey" due to limitation by "minRecSizes"
// . resulting list will contain ALL keys between ITS [m_startKey,m_endKey]
// . final merged list "should" try to have a size of at least "minRecSizes"
// but due to negative/postive rec elimination may be less
// . the endKey of the lists we read may be <= "endKey" provided
// . we try to shrink the endKey if minRecSizes is >= 0 in order to
// avoid excessive reading
// . by shrinking the endKey we cannot take into account the size of deleted
// records, so therefore we may fall short of "minRecSizes" in actuality,
// in fact, the returned list may even be empty with a shrunken endKey
// . we merge all lists read from disk into the provided "list"
// . caller should call Msg3.getList(int32_t i) and Msg3:getNumLists() to retrieve
// . this makes the query engine faster since we don't need to merge the docIds
// and can just send them across the network separately and they will be
// hashed into IndexTable's table w/o having to do time-wasting merging.
// . caller can specify array of filenums to read from so incremental syncing
// in Sync class can just read from titledb*.dat files that were formed
// since the last sync point.
bool Msg3::readList ( char rdbId ,
collnum_t collnum ,
const char *startKeyArg ,
const char *endKeyArg ,
int32_t minRecSizes , // max size of scan
int32_t startFileNum , // first file to scan
int32_t numFiles , // rel. to startFileNum
void *state , // for callback
void (* callback ) ( void *state ) ,
int32_t niceness ,
int32_t retryNum ,
int32_t maxRetries ,
bool compensateForMerge ,
bool justGetEndKey ,
bool allowPageCache ,
bool hitDisk ) {
// set this to true to validate
m_validateCache = false;//true;
// clear, this MUST be done so if we return true g_errno is correct
g_errno = 0;
// assume lists are not checked for corruption
m_listsChecked = false;
// warn
if ( minRecSizes < -1 ) {
log(LOG_LOGIC,"db: Msg3 got minRecSizes of %" PRId32", changing "
"to -1.",minRecSizes);
minRecSizes = -1;
}
// reset m_alloc and data in all lists in case we are a re-call
reset();
// warning
if ( collnum < 0 ) log(LOG_LOGIC,"net: NULL collection. msg3.");
// remember the callback
m_rdbId = rdbId;
m_collnum = collnum;
m_callback = callback;
m_state = state;
m_niceness = niceness;
m_numScansCompleted = 0;
m_retryNum = retryNum;
m_maxRetries = maxRetries;
m_compensateForMerge = compensateForMerge;
m_allowPageCache = allowPageCache;
m_hitDisk = hitDisk;
m_hadCorruption = false;
// get keySize of rdb
m_ks = getKeySizeFromRdbId ( m_rdbId );
// reset the group error
m_errno = 0;
// . reset all our lists
// . these are reset in call the RdbScan::setRead() below
//for ( int32_t i = 0 ; i < MAX_RDB_FILES ; i++ ) m_lists[i].reset();
// . ensure startKey last bit clear, endKey last bit set
// . no! this warning is now only in Msg5
// . if RdbMerge is merging some files, not involving the root
// file, then we can expect to get a lot of unmatched negative recs.
// . as a consequence, our endKeys may often be negative. This means
// it may not annihilate with the positive key, but we should only
// miss like this at the boundaries of the lists we fetch.
// . so in that case RdbList::merge will stop merging once the
// minRecSizes limit is reached even if it means ending on a negative
// rec key
//if ( (startKey.n0 & 0x01) == 0x01 )
if ( !KEYNEG(startKeyArg) )
log(LOG_REMIND,"net: msg3: StartKey lastbit set.");
if ( KEYNEG(endKeyArg) )
log(LOG_REMIND,"net: msg3: EndKey lastbit clear.");
// declare vars here becaues of 'goto skip' below
int32_t mergeFileNum = -1 ;
int32_t max ;
// get base, returns NULL and sets g_errno to ENOCOLLREC on error
RdbBase *base = getRdbBase( m_rdbId, m_collnum );
if ( ! base ) {
return true;
}
// store the file numbers in the array, these are the files we read
m_numFileNums = 0;
// save startFileNum here, just for recall
m_startFileNum = startFileNum;
m_numFiles = numFiles;
// . if we have a merge going on, we may have to change startFileNum
// . if some files get unlinked because merge completes then our
// reads will detect the error and loop back here
// . we launch are reads right after this without giving up the cpu
// and we use file descriptors, so any changes to Rdb::m_files[]
// should not hurt us
// . WARNING: just make sure you don't lose control of cpu until after
// you call RdbScan::set()
// . we use hasMergeFile() instead of isMerging() because he may not
// be merging cuz he got suspended or he restarted and
// hasn't called attemptMerge() yet, but he may still contain it
if ( g_conf.m_logDebugQuery )
log(LOG_DEBUG,
"net: msg3: "
"c=%" PRId32" hmf=%" PRId32" sfn=%" PRId32" msfn=%" PRId32" nf=%" PRId32" db=%s.",
(int32_t)compensateForMerge,(int32_t)base->hasMergeFile(),
(int32_t)startFileNum,(int32_t)base->m_mergeStartFileNum-1,
(int32_t)numFiles,base->m_dbname);
int32_t pre = -10;
if ( compensateForMerge && base->hasMergeFile() &&
startFileNum >= base->m_mergeStartFileNum - 1 &&
(startFileNum > 0 || numFiles != -1) ) {
// now also include the file being merged into, but only
// if we are reading from a file being merged...
if ( startFileNum < base->m_mergeStartFileNum +
base->m_numFilesToMerge - 1 )
//m_fileNums [ m_numFileNums++ ] =
// base->m_mergeStartFileNum - 1;
pre = base->m_mergeStartFileNum - 1;
// debug msg
if ( g_conf.m_logDebugQuery )
log(LOG_DEBUG,
"net: msg3: startFileNum from %" PRId32" to %" PRId32" (mfn=%" PRId32")",
startFileNum,startFileNum+1,mergeFileNum);
// if merge file was inserted before us, inc our file number
startFileNum++;
}
// adjust num files if we need to, as well
if ( compensateForMerge && base->hasMergeFile() &&
startFileNum < base->m_mergeStartFileNum - 1 &&
numFiles != -1 &&
startFileNum + numFiles - 1 >= base->m_mergeStartFileNum - 1 ) {
// debug msg
if ( g_conf.m_logDebugQuery )
log(LOG_DEBUG,"net: msg3: numFiles up one.");
// if merge file was inserted before us, inc our file number
numFiles++;
}
// . how many rdb files does this base have?
// . IMPORTANT: this can change since files are unstable because they
// might have all got merged into one!
// . so do this check to make sure we're safe... especially if
// there was an error before and we called readList() on ourselves
max = base->getNumFiles();
// -1 means we should scan ALL the files in the base
if ( numFiles == -1 ) numFiles = max;
// limit it by startFileNum, however
if ( numFiles > max - startFileNum ) numFiles = max - startFileNum;
// set g_errno and return true if it is < 0
if ( numFiles < 0 ) {
log(LOG_LOGIC,
"net: msg3: readList: numFiles = %" PRId32" < 0 (max=%" PRId32")(sf=%" PRId32")",
numFiles , max , startFileNum );
g_errno = EBADENGINEER;
// force core dump
char *xx=NULL;*xx=0;
return true;
}
// . allocate buffer space
// . m_scans, m_startpg, m_endpg, m_hintKeys, m_hintOffsets,
// m_fileNums, m_lists
int32_t chunk = sizeof(RdbScan) + // m_scans
4 + // m_startpg
4 + // m_endpg
//sizeof(key_t) + // m_hintKeys
m_ks + // m_hintKeys
4 + // m_hintOffsets
4 + // m_fileNums
sizeof(RdbList) ; // m_lists
int32_t nn = numFiles;
if ( pre != -10 ) nn++;
m_numChunks = nn;
int32_t need = nn * (chunk);
m_alloc = m_buf;
if ( need > (int32_t)MSG3_BUF_SIZE ) {
m_allocSize = need;
m_alloc = (char *)mcalloc ( need , "Msg3" );
if ( ! m_alloc ) {
log("disk: Could not allocate %" PRId32" bytes read "
"structures to read %s.",need,base->m_dbname);
return true;
}
}
char *p = m_alloc;
m_scans = (RdbScan *)p; p += nn * sizeof(RdbScan);
m_startpg = (int32_t *)p; p += nn * 4;
m_endpg = (int32_t *)p; p += nn * 4;
//m_hintKeys = (key_t *)p; p += nn * sizeof(key_t);
m_hintKeys = (char *)p; p += nn * m_ks;
m_hintOffsets = (int32_t *)p; p += nn * 4;
m_fileNums = (int32_t *)p; p += nn * 4;
m_lists = (RdbList *)p; p += nn * sizeof(RdbList);
// sanity check
if ( p - m_alloc != need ) {
log(LOG_LOGIC,"disk: Bad malloc in Msg3.cpp.");
char *xx = NULL; *xx = 0;
}
// call constructors
for ( int32_t i = 0 ; i < nn ; i++ ) m_lists[i].constructor();
// make fix from up top
if ( pre != -10 ) m_fileNums [ m_numFileNums++ ] = pre;
// store them all
for ( int32_t i = startFileNum ; i < startFileNum + numFiles ; i++ )
m_fileNums [ m_numFileNums++ ] = i;
// . remove file nums that are being unlinked after a merge now
// . keep it here (below skip: label) so sync point reads can use it
int32_t n = 0;
for ( int32_t i = 0 ; i < m_numFileNums ; i++ ) {
// skip those that are being unlinked after the merge
if ( base->m_isUnlinking &&
m_fileNums[i] >= base->m_mergeStartFileNum &&
m_fileNums[i] < base->m_mergeStartFileNum +
base->m_numFilesToMerge )
continue;
// otherwise, keep it
m_fileNums[n++] = m_fileNums[i];
}
m_numFileNums = n;
// . if root file is being merged, he's file #0, & root file is file #1
// . this is a hack so caller gets what he wants
//if ( startFileNum == 0 && base->getFileId(0) == 0 && numFiles == 1 )
// numFiles = 2;
// remember the file range we should scan
m_numScansStarted = 0;
m_numScansCompleted = 0;
//m_startKey = startKey;
//m_endKey = endKey;
//m_constrainKey = endKey; // set in case justGetEndKey is true
KEYSET(m_startKey,startKeyArg,m_ks);
KEYSET(m_endKey,endKeyArg,m_ks);
KEYSET(m_constrainKey,endKeyArg,m_ks);//set incase justGetEndKey istrue
m_minRecSizes = minRecSizes;
m_compensateForMerge = compensateForMerge;
// bail if 0 files to scan -- no! need to set startKey/endKey
if ( numFiles == 0 ) return true;
// don't read anything if endKey < startKey
//if ( m_startKey > m_endKey ) return true;
if ( KEYCMP(m_startKey,m_endKey,m_ks)>0 ) return true;
// keep the original in tact in case g_errno == ETRYAGAIN
//m_endKeyOrig = endKey;
KEYSET(m_endKeyOrig,endKeyArg,m_ks);
m_minRecSizesOrig = minRecSizes;
// start reading at this key
m_fileStartKey = startKeyArg;
// start the timer, keep it fast for clusterdb though
if ( g_conf.m_logTimingDb ) m_startTime = gettimeofdayInMilliseconds();
// translate base to an id, for the sake of m_msg0
//char baseId = m_msg0->getRdbId ( base );
// map ptrs
RdbMap **maps = base->getMaps();
// . we now boost m_minRecSizes to account for negative recs
// . but not if only reading one list, cuz it won't get merged and
// it will be too big to send back
if ( m_numFileNums > 1 ) compensateForNegativeRecs ( base );
// . often endKey is too big for an efficient read of minRecSizes bytes
// because we end up reading too much from all the files
// . this will set m_startpg[i], m_endpg[i] for each RdbScan/RdbFile
// to ensure we read "minRecSizes" worth of records, not much more
// . returns the new endKey for all ranges
// . now this just overwrites m_endKey
//m_endKey = setPageRanges ( base ,
setPageRanges ( base ,
m_fileNums ,
m_numFileNums ,
m_fileStartKey , // start reading @ key
m_endKey , // stop reading @ key
m_minRecSizes );
// . NEVER let m_endKey be a negative key, because it will
// always be unmatched, since delbit is cleared
// . adjusting it here ensures our generated hints are valid
// . we will use this key to call constrain() with
//m_constrainKey = m_endKey;
//if ( ( m_constrainKey.n0 & 0x01) == 0x00 )
// m_constrainKey -= (uint32_t)1;
KEYSET(m_constrainKey,m_endKey,m_ks);
if ( KEYNEG(m_constrainKey) )
KEYSUB(m_constrainKey,m_ks);
// Msg5 likes to get the endkey for getting the list from the tree
if ( justGetEndKey ) return true;
// sanity check
if ( m_numFileNums > nn ) {
log(LOG_LOGIC,"disk: Failed sanity check in Msg3.");
char *xx = NULL; *xx = 0;
}
// debug msg
//log("msg3 getting list (msg5=%" PRIu32")",m_state);
// . MDW removed this -- go ahead an end on a delete key
// . RdbMerge might not pick it up this round, but oh well
// . so we can have both positive and negative co-existing in same file
// make sure the last bit is set so we don't end on a delete key
//m_endKey.n0 |= 0x01LL;
// . now start reading/scanning the files
// . our m_scans array starts at 0
for ( int32_t i = 0 ; i < m_numFileNums ; i++ ) {
// get the page range
//int32_t p1 = m_startpg [ i ];
//int32_t p2 = m_endpg [ i ];
//#ifdef GBSANITYCHECK
int32_t fn = m_fileNums[i];
// this can happen somehow!
if ( fn < 0 ) {
log(LOG_LOGIC,"net: msg3: fn=%" PRId32". Bad engineer.",fn);
continue;
}
// sanity check
if ( i > 0 && m_fileNums[i-1] >= fn ) {
log(LOG_LOGIC,
"net: msg3: files must be read in order "
"from oldest to newest so RdbList::indexMerge_r "
"works properly. Otherwise, corruption will "
"result. ");
char *xx = NULL; *xx = 0;
return true;
}
// . sanity check?
// . no, we must get again since we turn on endKey's last bit
int32_t p1 , p2;
maps[fn]->getPageRange ( m_fileStartKey ,
m_endKey ,
&p1 ,
&p2 ,
NULL );
//if ( p1 != p1c || p2 != p2c ) {
// fprintf(stderr,"Msg3::bad page range\n");
// sleep(50000);
//}
// sanity check, each endpg's key should be > endKey
//if ( p2 < maps[fn]->getNumPages() &&
// maps[fn]->getKey ( p2 ) <= m_endKey ) {
// fprintf(stderr,"Msg3::bad page range 2\n");
// sleep(50000);
//}
//#endif
//int32_t p1 , p2;
//maps[fn]->getPageRange (startKey,endKey,minRecSizes,&p1,&p2);
// now get some read info
int64_t offset = maps[fn]->getAbsoluteOffset ( p1 );
int32_t bytesToRead = maps[fn]->getRecSizes ( p1, p2, false);
// max out the endkey for this list
// debug msg
//#ifdef _DEBUG_
//if ( minRecSizes == 2000000 )
//log("Msg3:: reading %" PRId32" bytes from file #%" PRId32,bytesToRead,i);
//#endif
// inc our m_numScans
m_numScansStarted++;
// . keep stats on our disk accesses
// . count disk seeks (assuming no fragmentation)
// . count disk bytes read
if ( bytesToRead > 0 ) {
base->m_rdb->didSeek ( );
base->m_rdb->didRead ( bytesToRead );
}
// . the startKey may be different for each RdbScan class
// . RdbLists must have all keys within their [startKey,endKey]
// . therefore set startKey individually from first page in map
// . this endKey must be >= m_endKey
// . this startKey must be < m_startKey
//key_t startKey = maps[fn]->getKey ( p1 );
//key_t endKey = maps[fn]->getKey ( p2 );
char startKey2 [ MAX_KEY_BYTES ];
char endKey2 [ MAX_KEY_BYTES ];
maps[fn]->getKey ( p1 , startKey2 );
maps[fn]->getKey ( p2 , endKey2 );
//char *startKey = maps[fn]->getKeyPtr ( p1 );
//char *endKey = maps[fn]->getKeyPtr ( p2 );
// store in here
m_startpg [ i ] = p1;
m_endpg [ i ] = p2;
// . we read UP TO that endKey, so reduce by 1
// . but iff p2 is NOT the last page in the map/file
// . maps[fn]->getKey(lastPage) will return the LAST KEY
// and maps[fn]->getOffset(lastPage) the length of the file
//if ( maps[fn]->getNumPages()!=p2) endKey -=(uint32_t)1;
if ( maps[fn]->getNumPages() != p2 ) KEYSUB(endKey2,m_ks);
// otherwise, if we're reading all pages, then force the
// endKey to virtual inifinite
//else endKey.setMax();
else KEYMAX(endKey2,m_ks);
// . set up the hints
// . these are only used if we are only reading from 1 file
// . these are used to call constrain() so we can constrain
// the end of the list w/o looping through all the recs
// in the list
int32_t h2 = p2 ;
// decrease by one page if we're on the last page
if ( h2 > p1 && maps[fn]->getNumPages() == h2 ) h2--;
// . decrease hint page until key is <= endKey on that page
// AND offset is NOT -1 because the old way would give
// us hints passed the endkey
// . also decrease so we can constrain on minRecSizes in
// case we're the only list being read
// . use >= m_minRecSizes instead of >, otherwise we may
// never be able to set "size" in RdbList::constrain()
// because "p" could equal "maxPtr" right away
while ( h2 > p1 &&
//( maps[fn]->getKey (h2) > m_constrainKey ||
(KEYCMP(maps[fn]->getKeyPtr(h2),m_constrainKey,m_ks)>0||
maps[fn]->getOffset(h2) == -1 ||
maps[fn]->getAbsoluteOffset(h2) - offset >=
m_minRecSizes ) )
h2--;
// now set the hint
m_hintOffsets [ i ] = maps[fn]->getAbsoluteOffset ( h2 ) -
maps[fn]->getAbsoluteOffset ( p1 ) ;
//m_hintKeys [ i ] = maps[fn]->getKey ( h2 );
KEYSET(&m_hintKeys[i*m_ks],maps[fn]->getKeyPtr(h2),m_ks);
// reset g_errno before calling setRead()
g_errno = 0;
// . this fix is now in RdbList::checklist_r()
// . we can now have dup keys, so, we may read in
// a rec with key "lastMinKey" even though we don't read
// in the first key on the end page, so don't subtract 1...
//if ( endKey != m_endKeyOrig )
// endKey += (uint32_t) 1;
// timing debug
if ( g_conf.m_logTimingDb )
log(LOG_TIMING,
"net: msg: reading %" PRId32" bytes from %s file #%" PRId32" "
"(niceness=%" PRId32")",
bytesToRead,base->m_dbname,i,m_niceness);
// log huge reads, those hurt us
if ( bytesToRead > 150000000 ) {
logf(LOG_INFO,"disk: Reading %" PRId32" bytes at offset %" PRId64" "
"from %s.",
bytesToRead,offset,base->m_dbname);
}
// if any keys in the map are the same report corruption
char tmpKey [16];
char lastTmpKey[16];
int32_t ccount = 0;
if ( bytesToRead > 10000000 &&
bytesToRead / 2 > m_minRecSizes &&
base->m_fixedDataSize >= 0 ) {
for ( int32_t pn = p1 ; pn <= p2 ; pn++ ) {
maps[fn]->getKey ( pn , tmpKey );
if ( KEYCMP(tmpKey,lastTmpKey,m_ks) == 0 )
ccount++;
gbmemcpy(lastTmpKey,tmpKey,m_ks);
}
}
if ( ccount > 10 ) {
logf(LOG_INFO,"disk: Reading %" PRId32" bytes from %s file #"
"%" PRId32" when min "
"required is %" PRId32". Map is corrupt and has %" PRId32" "
"identical consecutive page keys because the "
"map was \"repaired\" because out of order keys "
"in the index.",
(int32_t)bytesToRead,
base->m_dbname,fn,
(int32_t)m_minRecSizes,
(int32_t)ccount);
m_numScansCompleted++;
m_errno = ECORRUPTDATA;
m_hadCorruption = true;
//m_maxRetries = 0;
break;
}
////////
//
// try to get from PAGE CACHE
//
////////
BigFile *ff = base->getFile(m_fileNums[i]);
RdbCache *rpc = getDiskPageCache ( m_rdbId );
if ( ! m_allowPageCache ) rpc = NULL;
// . vfd is unique 64 bit file id
// . if file is opened vfd is -1, only set in call to open()
int64_t vfd = ff->getVfd();
key192_t ck = makeCacheKey ( vfd , offset, bytesToRead);
char *rec; int32_t recSize;
bool inCache = false;
if ( rpc && vfd != -1 && ! m_validateCache )
inCache = rpc->getRecord ( (collnum_t)0 , // collnum
(char *)&ck ,
&rec ,
&recSize ,
true , // copy?
-1 , // maxAge, none
true ); // inccounts?
m_scans[i].m_inPageCache = false;
if ( inCache ) {
m_scans[i].m_inPageCache = true;
m_numScansCompleted++;
// now we have to store this value, 6 or 12 so
// we can modify the hint appropriately
m_scans[i].m_shifted = *rec;
m_lists[i].set ( rec +1,
recSize-1 ,
rec , // alloc
recSize , // allocSize
startKey2 ,
endKey2 ,
base->m_fixedDataSize ,
true , // owndata
base->useHalfKeys() ,
getKeySizeFromRdbId ( m_rdbId ) );
continue;
}
// . do the scan/read of file #i
// . this returns false if blocked, true otherwise
// . this will set g_errno on error
bool done = m_scans[i].setRead (base->getFile(m_fileNums[i]),
base->m_fixedDataSize ,
offset ,
bytesToRead ,
startKey2 ,
endKey2 ,
m_ks ,
&m_lists[i] ,
this ,
doneScanningWrapper ,
base->useHalfKeys() ,
m_rdbId,
m_niceness ,
m_allowPageCache ,
m_hitDisk ) ;
// . damn, usually the above will indirectly launch a thread
// to do the reading, but it sets g_errno to EINTR,
// "interrupted system call"!
// . i guess the thread does the read w/o blocking and then
// queues the signal on g_loop's queue before it exits
// . try ignoring, and keep going
if ( g_errno == EINTR ) {
log("net: Interrupted system call while reading file. "
"Ignoring.");
g_errno = 0;
}
// debug msg
//fprintf(stderr,"Msg3:: reading %" PRId32" bytes from file #%" PRId32","
// "done=%" PRId32",offset=%" PRId64",g_errno=%s,"
// "startKey=n1=%" PRIu32",n0=%" PRIu64", "
// "endKey=n1=%" PRIu32",n0=%" PRIu64"\n",
// bytesToRead,i,(int32_t)done,offset,mstrerror(g_errno),
// m_startKey,m_endKey);
//if ( bytesToRead == 0 )
// fprintf(stderr,"shit\n");
// if it did not block then it completed, so count it
if ( done ) m_numScansCompleted++;
// break on an error, and remember g_errno in case we block
if ( g_errno && g_errno != ENOTHREADSLOTS ) {
int32_t tt = LOG_WARN;
if ( g_errno == EFILECLOSED ) tt = LOG_INFO;
log(tt,"disk: Reading %s had error: %s.",
base->m_dbname, mstrerror(g_errno));
m_errno = g_errno;
break;
}
}
// debug test
//if ( rand() % 100 <= 10 ) m_errno = EIO;
// if we blocked, return false
if ( m_numScansCompleted < m_numScansStarted ) return false;
// . if all scans completed without blocking then wrap it up & ret true
// . doneScanning may now block if it finds data corruption and must
// get the list remotely
return doneScanning();
}
// . return false if blocked, true otherwise
// . sets g_errno on error
bool RdbMerge::getNextList ( ) {
// return true if g_errno is set
if ( g_errno || m_doneMerging ) return true;
// it's suspended so we count this as blocking
if ( m_isSuspended ) {
m_isReadyToSave = true;
return false;
}
// if the power is off, suspend the merging
if ( ! g_process.m_powerIsOn ) {
m_isReadyToSave = true;
doSleep();
return false;
}
// no chop threads
m_numThreads = 0;
// get base, returns NULL and sets g_errno to ENOCOLLREC on error
RdbBase *base; if (!(base=getRdbBase(m_rdbId,m_collnum))) return true;
// . if a contributor has just surpassed a "part" in his BigFile
// then we can delete that part from the BigFile and the map
for ( long i = m_startFileNum ; i < m_startFileNum + m_numFiles; i++ ){
RdbMap *map = base->m_maps[i];
long page = map->getPage ( m_startKey );
long long offset = map->getAbsoluteOffset ( page );
BigFile *file = base->m_files[i];
long part = file->getPartNum ( offset ) ;
if ( part == 0 ) continue;
// i've seen this bug happen if we chop a part off on our
// last dump and the merge never completes for some reason...
// so if we're in the last part then don't chop the part b4 us
if ( part >= file->m_maxParts - 1 ) continue;
// if we already unlinked part # (part-1) then continue
if ( ! file->doesPartExist ( part - 1 ) ) continue;
// . otherwise, excise from the map
// . we must be able to chop the mapped segments corresponding
// EXACTLY to the part file
// . therefore, PAGES_PER_SEGMENT define'd in RdbMap.h must
// evenly divide MAX_PART_SIZE in BigFile.h
// . i do this check in RdbMap.cpp
if ( ! map->chopHead ( MAX_PART_SIZE ) ) {
// we had an error!
log("db: Failed to remove data from map for "
"%s.part%li.",
file->getFilename(),part);
return true;
}
// . also, unlink any part files BELOW part # "part"
// . this returns false if it blocked, true otherwise
// . this sets g_errno on error
// . now we just unlink part file #(part-1) explicitly
if ( ! file->chopHead ( part - 1 , chopWrapper , this ) )
m_numThreads++;
if ( ! g_errno ) continue;
log("db: Failed to unlink file %s.part%li.",
file->getFilename(),part);
return true;
}
// wait for file to be unlinked before getting list
if ( m_numThreads > 0 ) return false;
// otherwise, get it now
return getAnotherList ( );
}
bool Syncdb::init ( ) {
// reset
loopReset();
m_doRcp = false;
m_rcpStarted = false;
// setup quick tree
if ( ! m_qt.set ( 0 , // fixedDataSize
300000 , // 300k nodes
true , // balance?
-1 , // maxmem, no max
false , // ownData?
"tresyncdb" ,
false , // dataInPtrs?
"quicktree" , // dbname
16 , // keySize
false ))// useProtection?
return false;
BigFile f;
f.set ( g_hostdb.m_dir , "quicktree.dat" );
// only load if it exists
bool exists = f.doesExist();
// load it
if ( exists && ! m_qt.fastLoad( &f , &m_stack ) )
return log("sync: quicktree.dat load failed: %s",
mstrerror(g_errno));
// done
f.close();
// assume permanently out of sync
long val = 2;
// load the insync.dat file
f.set ( g_hostdb.m_dir , "insync.dat" );
// fail on open failure
if ( ! f.open ( O_RDONLY ) ) return false;
// if not there, permanently out of sync
if ( ! f.doesExist() )
log("sync: insync.dat does not exist. Assuming host is "
"unrecoverable.");
else {
// get the value
char buf[20];
long n = f.read ( &buf , 10 , 0 ) ;
if ( n <= 0 )
return log("sync: read insync.dat: %s",
mstrerror(g_errno));
// must be digit
if ( ! is_digit ( buf[0] ) )
return log("sync: insync.dat has no number in it.");
// unlink it
if ( ! f.unlink() )
return log("sync: failed to unlink insync.dat: %s",
mstrerror(g_errno));
// get the value
val = atol ( buf );
}
// bad val?
if ( val < 0 || val > 2 )
return log("sync: insync.dat had bad value of %li",val);
// report if in sync or not
if ( val == 0 ) log("sync: insync.dat says out of sync");
if ( val == 1 ) log("sync: insync.dat says in sync");
if ( val == 2 ) log("sync: insync.dat says PERMANENTLY out of sync");
// set it
Host *h = g_hostdb.m_myHost;
if ( val == 1 ) h->m_inSync = 1;
if ( val == 2 ) h->m_isPermanentOutOfSync = 1;
// call this once per second
if ( ! g_loop.registerSleepCallback ( 1000 , NULL , sleepWrapper ) )
return false;
// 10 MB
long maxTreeMem = 10000000;
// . what's max # of tree nodes?
// . key+4+left+right+parents+dataPtr = 12+4 +4+4+4+4 = 32
// . 28 bytes per record when in the tree
long maxTreeNodes = maxTreeMem / ( 16 + 1000 );
// . initialize our own internal rdb
// . records are actual msg4 requests received from Msg4
// . the key is formed calling Syncdb::makeKey() which is based on
// the tid, sid and zid of the msg4 request, where tid is the
// twin hostid we are chatting with in our group, sid is the
// ORIGINAL sending hostid of the msg4 request, and zid is the
// kinda transaction #, and is unique.
if ( ! m_rdb.init ( g_hostdb.m_dir ,
"syncdb" ,
true , // dedup
-1 , // dataSize is variable
50 , // min files to merge
maxTreeMem ,
maxTreeNodes , // maxTreeNodes ,
true , // balance tree?
50000 , // maxCacheMem ,
100 , // maxCacheNodes ,
false , // half keys?
false , // save cache?
NULL , // page cache
false , // is titledb
false , // preload disk page cache
16 , // key size
false , // bias disk page cache?
true ))// is collectionless?
return false;
// add the coll
//if ( ! g_syncdb.m_rdb.addColl ( "dummy" ) ) return true;
// reset quick tree?
if ( ! h->m_isPermanentOutOfSync ) return true;
// clear it all!
m_qt.clear();
// add the base since it is a collectionless rdb
return m_rdb.addRdbBase1 ( NULL );
}
// . sets m_fileOffset and m_bf
// . returns false and sets g_errno on error
// . returns false if nothing to read too... but does not set g_errno
bool ImportState::setCurrentTitleFileAndOffset ( ) {
// leave m_bf and m_fileOffset alone if there is more to read
if ( m_fileOffset < m_bfFileSize )
return true;
CollectionRec *cr = g_collectiondb.getRec ( m_collnum );
if ( ! cr ) return false;
log("import: import finding next file");
// if ( m_offIsValid ) {
// //*off = m_fileOffset;
// return &m_bf;
// }
//m_offIsValid = true;
// look for titledb0001.dat etc. files in the
// workingDir/inject/ subdir
SafeBuf ddd;
ddd.safePrintf("%sinject",cr->m_importDir.getBufStart());
// now use the one provided. we should also provide the # of threads
if ( cr->m_importDir.getBufStart() &&
cr->m_importDir.getBufStart()[0] ) {
ddd.reset();
ddd.safeStrcpy ( cr->m_importDir.getBufStart() );
}
//
// assume we are the first filename
// set s_fileId to the minimum
//
Dir dir;
dir.set(ddd.getBufStart());
if ( ! dir.open() ) return false;
// assume none
long minFileId = -1;
// getNextFilename() writes into this
char pattern[64]; strcpy ( pattern , "titledb*" );
char *filename;
while ( ( filename = dir.getNextFilename ( pattern ) ) ) {
// filename must be a certain length
long filenameLen = gbstrlen(filename);
// we need at least "titledb0001.dat"
if ( filenameLen < 15 ) continue;
// ensure filename starts w/ our m_dbname
if ( strncmp ( filename , "titledb", 7 ) != 0 )
continue;
// skip if not .dat file
if ( ! strstr ( filename , ".dat" ) )
continue;
// then a 4 digit number should follow
char *s = filename + 7;
if ( ! isdigit(*(s+0)) ) continue;
if ( ! isdigit(*(s+1)) ) continue;
if ( ! isdigit(*(s+2)) ) continue;
if ( ! isdigit(*(s+3)) ) continue;
// convert digit to id
long id = atol(s);
// . do not accept files we've already processed
// . -1 means we haven't processed any yet
if ( m_bfFileId >= 0 && id <= m_bfFileId ) continue;
// the min of those we haven't yet processed/injected
if ( id < minFileId || minFileId < 0 ) minFileId = id;
}
// get where we left off
if ( ! m_loadedPlaceHolder ) {
// read where we left off from file if possible
char fname[256];
sprintf(fname,"%slasttitledbinjectinfo.dat",g_hostdb.m_dir);
SafeBuf ff;
ff.fillFromFile(fname);
if ( ff.length() > 1 ) {
m_loadedPlaceHolder = true;
// get the placeholder
sscanf ( ff.getBufStart()
, "%llu,%lu"
, &m_fileOffset
, &minFileId
);
}
}
// if no files! return false to indicate we are done
if ( minFileId == -1 ) return false;
// set up s_bf then
//if ( m_bfFileId != minFileId ) {
SafeBuf tmp;
tmp.safePrintf("titledb%04li-000.dat"
//,dir.getDirname()
,minFileId);
m_bf.set ( dir.getDirname() ,tmp.getBufStart() );
if ( ! m_bf.open( O_RDONLY ) ) {
log("inject: import: could not open %s%s for reading",
dir.getDirname(),tmp.getBufStart());
return false;
}
m_bfFileId = minFileId;
// reset ptr into file
//*off = 0;
// and set this
m_bfFileSize = m_bf.getFileSize();
m_fileOffset = 0;
//}
log("import: importing from file %s",m_bf.getFilename());
return true;//&m_bf;
}
//
// . ENTRY POINT FOR IMPORTING TITLEDB RECS FROM ANOTHER CLUSTER
// . when user clicks 'begin' in import page we come here..
// . so when that parm changes in Parms.cpp we sense that and call
// beginImport(CollectionRec *cr)
// . or on startup we call resumeImports to check each coll for
// an import in progress.
// . search for files named titledb*.dat
// . if none found just return
// . when msg7 inject competes it calls this
// . call this from sleep wrapper in Process.cpp
// . returns false if would block (outstanding injects), true otherwise
// . sets g_errno on error
bool ImportState::importLoop ( ) {
CollectionRec *cr = g_collectiondb.getRec ( m_collnum );
if ( ! cr || g_hostdb.m_hostId != 0 ) {
// if coll was deleted!
log("import: collnum %li deleted while importing into",
(long)m_collnum);
//if ( m_numOut > m_numIn ) return true;
// delete the entire import state i guess
// what happens if we have a msg7 reply come back in?
// it should see the collrec is NULL and just fail.
mdelete ( this, sizeof(ImportState) , "impstate");
delete (this);
return true;
}
INJECTLOOP:
// stop if waiting on outstanding injects
long long out = m_numOut - m_numIn;
if ( out >= cr->m_numImportInjects ) {
g_errno = 0;
return false;
}
if ( ! cr->m_importEnabled ) {
// wait for all to return
if ( out > 0 ) return false;
// then delete it
log("import: collnum %li import loop disabled",
(long)m_collnum);
mdelete ( this, sizeof(ImportState) , "impstate");
delete (this);
return true;
}
// scan each titledb file scanning titledb0001.dat first,
// titledb0003.dat second etc.
//long long offset = -1;
// . when offset is too big for current m_bigFile file then
// we go to the next and set offset to 0.
// . sets m_bf and m_fileOffset
if ( ! setCurrentTitleFileAndOffset ( ) ) {//cr , -1 );
log("import: import: no files to read");
//goto INJECTLOOP;
return true;
}
// this is -1 if none remain!
if ( m_fileOffset == -1 ) {
log("import: import fileoffset is -1. done.");
return true;
}
long long saved = m_fileOffset;
//Msg7 *msg7;
//GigablastRequest *gr;
//SafeBuf *sbuf = NULL;
long need = 12;
long dataSize = -1;
//XmlDoc xd;
key_t tkey;
bool status;
SafeBuf tmp;
SafeBuf *sbuf = &tmp;
long long docId;
long shardNum;
long key;
Multicast *mcast;
char *req;
long reqSize;
if ( m_fileOffset >= m_bfFileSize ) {
log("inject: import: done processing file %li %s",
m_bfFileId,m_bf.getFilename());
goto nextFile;
}
// read in title rec key and data size
status = m_bf.read ( &tkey, sizeof(key_t) , m_fileOffset );
//if ( n != 12 ) goto nextFile;
if ( g_errno ) {
log("inject: import: reading file error: %s. advancing "
"to next file",mstrerror(g_errno));
goto nextFile;
}
m_fileOffset += 12;
// if negative key, skip
if ( (tkey.n0 & 0x01) == 0 ) {
goto INJECTLOOP;
}
// if non-negative then read in size
status = m_bf.read ( &dataSize , 4 , m_fileOffset );
if ( g_errno ) {
log("main: failed to read in title rec "
"file. %s. Skipping file %s",
mstrerror(g_errno),m_bf.getFilename());
goto nextFile;
}
m_fileOffset += 4;
need += 4;
need += dataSize;
need += 4; // collnum, first 4 bytes
if ( dataSize < 0 || dataSize > 500000000 ) {
log("main: could not scan in titledb rec of "
"corrupt dataSize of %li. BAILING ENTIRE "
"SCAN of file %s",dataSize,m_bf.getFilename());
goto nextFile;
}
//gr = &msg7->m_gr;
//XmlDoc *xd = getAvailXmlDoc();
//msg7 = getAvailMsg7();
mcast = getAvailMulticast();
// if none, must have to wait for some to come back to us
if ( ! mcast ) {
// restore file offset
//m_fileOffset = saved;
// no, must have been a oom or something
log("import: import no mcast available");
return true;//false;
}
// this is for holding a compressed titlerec
//sbuf = &mcast->m_sbuf;//&gr->m_sbuf;
// point to start of buf
sbuf->reset();
// ensure we have enough room
sbuf->reserve ( need );
// collnum first 4 bytes
sbuf->pushLong( (long)m_collnum );
// store title key
sbuf->safeMemcpy ( &tkey , sizeof(key_t) );
// then datasize if any. neg rec will have -1 datasize
if ( dataSize >= 0 )
sbuf->pushLong ( dataSize );
// then read data rec itself into it, compressed titlerec part
if ( dataSize > 0 ) {
// read in the titlerec after the key/datasize
status = m_bf.read ( sbuf->getBuf() ,
dataSize ,
m_fileOffset );
if ( g_errno ) { // n != dataSize ) {
log("main: failed to read in title rec "
"file. %s. Skipping file %s",
mstrerror(g_errno),m_bf.getFilename());
// essentially free up this msg7 now
//msg7->m_inUse = false;
//msg7->reset();
goto nextFile;
}
// advance
m_fileOffset += dataSize;
// it's good, count it
sbuf->m_length += dataSize;
}
// set xmldoc from the title rec
//xd->set ( sbuf.getBufStart() );
//xd->m_masterState = NULL;
//xd->m_masterCallback ( titledbInjectLoop );
// we use this so we know where the doc we are injecting
// was in the foregien titledb file. so we can update our bookmark
// code.
mcast->m_hackFileOff = saved;//m_fileOffset;
mcast->m_hackFileId = m_bfFileId;
//
// inject a title rec buf this time, we are doing an import
// FROM A TITLEDB FILE!!!
//
//gr->m_titleRecBuf = &sbuf;
// break it down into gw
// xd.set2 ( sbuf.getBufStart() ,
// sbuf.length() , // max size
// cr->m_coll, // use our coll
// NULL , // pbuf for page parser
// 1 , // niceness
// NULL ); //sreq );
// // note it
// log("import: importing %s",xd.m_firstUrl.getUrl());
// now we can set gr for the injection
// TODO: inject the whole "sbuf" so we get sitenuminlinks etc
// all exactly the same...
// gr->m_url = xd.getFirstUrl()->getUrl();
// gr->m_queryToScrape = NULL;
// gr->m_contentDelim = 0;
// gr->m_contentTypeStr = g_contentTypeStrings [xd.m_contentType];
// gr->m_contentFile = NULL;
// gr->m_content = xd.ptr_utf8Content;
// gr->m_diffbotReply = NULL;
// gr->m_injectLinks = false;
// gr->m_spiderLinks = true;
// gr->m_shortReply = false;
// gr->m_newOnly = false;
// gr->m_deleteUrl = false;
// gr->m_recycle = true; // recycle content? or sitelinks?
// gr->m_dedup = false;
// gr->m_hasMime = false;
// gr->m_doConsistencyTesting = false;
// gr->m_getSections = false;
// gr->m_gotSections = false;
// gr->m_charset = xd.m_charset;
// gr->m_hopCount = xd.m_hopCount;
//
// point to next doc in the titledb file
//
//m_fileOffset += need;
// get docid from key
docId = g_titledb.getDocIdFromKey ( &tkey );
// get shard that holds the titlerec for it
shardNum = g_hostdb.getShardNumFromDocId ( docId );
// for selecting which host in the shard receives it
key = (long)docId;
m_numOut++;
// then index it. master callback will be called
//if ( ! xd->index() ) return false;
// TODO: make this forward the request to an appropriate host!!
// . gr->m_sbuf is set to the titlerec so this should handle that
// and use XmlDoc::set4() or whatever
// if ( msg7->injectTitleRec ( msg7 , // state
// gotMsg7ReplyWrapper , // callback
// cr )) {
// // it didn't block somehow...
// msg7->m_inUse = false;
// msg7->gotMsg7Reply();
// }
req = sbuf->getBufStart();
reqSize = sbuf->length();
if ( reqSize != need ) { char *xx=NULL;*xx=0 ; }
// do not free it, let multicast free it after sending it
sbuf->detachBuf();
if ( ! mcast->send ( req ,
reqSize ,
0x07 ,
true , // ownmsg?
shardNum,
false, // send to whole shard?
key , // for selecting host in shard
mcast , // state
NULL , // state2
gotMulticastReplyWrapper ,
999999 ) ) { // total timeout in seconds
log("import: import mcast had error: %s",mstrerror(g_errno));
m_numIn++;
}
goto INJECTLOOP;
nextFile:
// invalidate this flag
//m_offIsValid = false;
// . and call this function. we add one to m_bfFileId so we
// do not re-get the file we just injected.
// . sets m_bf and m_fileOffset
// . returns false if nothing to read
if ( ! setCurrentTitleFileAndOffset ( ) ) { //cr , m_bfFileId+1 );
log("import: import: no files left to read");
//goto INJECTLOOP;
return true;
}
// if it returns NULL we are done!
log("main: titledb injection loop completed. waiting for "
"outstanding injects to return.");
if ( m_numOut > m_numIn )
return false;
log("main: all injects have returned. DONE.");
// dummy return
return true;
}
int main(int argc, char* argv[])
{
int i;
int nums;
FDATA fdata;
Hash AddHash;
BigFile DestBigFile;
BigFile SourceBigFile;
BigFileEntry *bfe;
int numok,numnew,numupdated;
int sourcetype;
bool update;
char *p1=0;
char *p2=0;
char *p3=0;
int tosspath=0;
DataHandle addhandle;
int vargc;
char **vargv;
vargc=argc;
vargv=argv;
/* handle encrypted file */
kGUIProt DestProt;
bool usedestprot;
kGUIProt SourceProt;
bool usesourceprot;
kGUISystemBig sysbig;
kGUI::SetSystem(&sysbig);
signal(SIGINT, sigint_handler);
optcompress=false;
optrecursive=true;
optverify=false;
optdelete=false;
optmissing=false;
usedestprot=false;
usesourceprot=false;
#if 0
p1="/source/kgui/_data.big";
p2="/source/kgui/big";
p3="/source/kgui/big/";
optverify=true;
#endif
for(i=1; i<vargc; ++i)
{
if(vargv[i][0]=='-')
{
switch(vargv[i][1])
{
case 'c':
optcompress=true;
break;
case 'd':
optdelete=true;
break;
case 'm':
optmissing=true; /* delete missing files from subdir */
break;
case 'v':
optverify=true;
break;
case 'r':
optrecursive=false;
break;
case 'k':
if(vargv[i][2]=='d') /* encryption key on destination file */
{
printf("using dest key\n");
if(DestProt.SetKey(vargv[i+1],atoi(vargv[i+2]),atoi(vargv[i+3]),true)==false)
{
printf("Error loading dest keyfile '%s'\n",vargv[i+1]);
return(0);
}
usedestprot=true;
i+=3;
}
else if(vargv[i][2]=='s') /* encryption key on source file */
{
printf("using source key\n");
if(SourceProt.SetKey(vargv[i+1],atoi(vargv[i+2]),atoi(vargv[i+3]),true)==false)
{
printf("Error loading source keyfile '%s'\n",vargv[i+1]);
return(0);
}
usesourceprot=true;
i+=3;
}
optrecursive=false;
break;
default:
printf("Unknown parm '%s'\n",vargv[i]);
return(0);
break;
}
}
else
{
if(!p1)
p1=vargv[i];
else if(!p2)
p2=vargv[i];
else if(!p3)
{
p3=vargv[i];
tosspath=(int)strlen(p3);
}
else
{
printf("Unknown parm '%s'\n",vargv[i]);
return(0);
}
}
}
/* need at least 1 parm */
if(!p1)
{
printf("kguibig: (c) 2005 Kevin Pickelll\n");
printf("usage: kguibig bigfile.big path [root]\n");
printf(" -c = compress\n");
printf(" -v = verify\n");
printf(" -r = don't recurse\n");
printf(" -k[d,s] = filename offset len // source/dest key\n");
return(0);
}
DestBigFile.SetFilename(p1);
if(usedestprot==true)
DestBigFile.SetEncryption(&DestProt);
DestBigFile.Load(true);
if(DestBigFile.IsBad()==true)
{
printf("Dest file exists and is not a bigfile, or decryption key is incorrect!\n");
return(0);
}
/* list, verify or compress ? */
if(!p2)
{
if(optcompress)
{
}
else
{
unsigned long crc;
unsigned long vfsize;
BigFileEntry *sfe;
unsigned char copybuf[65536];
DataHandle checkhandle;
/* verify or list */
nums=DestBigFile.GetNumEntries();
for(i=0; ((i<nums) && (g_userabort==false)); ++i)
{
sfe=(BigFileEntry *)DestBigFile.GetEntry(i);
if(optverify)
{
/* check crc and print if no match */
printf("%d%c",i,13);
vfsize=sfe->m_size;
crc=0;
DestBigFile.CopyArea(&checkhandle,sfe->m_offset,sfe->m_size,sfe->m_time);
checkhandle.Open();
while(vfsize>sizeof(copybuf))
{
checkhandle.Read(©buf,(unsigned long)sizeof(copybuf));
crc=DestBigFile.CrcBuffer(crc,copybuf,sizeof(copybuf));
vfsize-=sizeof(copybuf);
};
/* write remainder */
if(vfsize>0)
{
checkhandle.Read(©buf,vfsize);
crc=DestBigFile.CrcBuffer(crc,copybuf,vfsize);
}
checkhandle.Close();
if(crc!=sfe->m_crc)
printf("CRC Error on file '%s' %06x<>%06x\n",sfe->GetName()->GetString(),(int)crc,sfe->m_crc);
}
else /* assume list if verify is not set */
printf("%s, len=%d,crc=%06x\n",sfe->GetName()->GetString(),sfe->m_size,sfe->m_crc);
}
}
return(0);
}
AddHash.Init(16,sizeof(FDATA));
/* is p2 a bigfile? */
if(strstr(p2,".big"))
{
SourceBigFile.SetFilename(p2);
if(usesourceprot==true)
SourceBigFile.SetEncryption(&SourceProt);
SourceBigFile.Load(true);
if(SourceBigFile.IsBad()==false)
sourcetype=SOURCE_BIG;
else
{
printf("Source file exists and is not a bigfile, or decryption key is incorrect!\n");
return(0);
}
}
else if(kGUI::IsDir(p2)==false)
{
fdata.time=kGUI::SysFileTime(p2);
//fdata.root=p3;
AddHash.Add(p2,&fdata);
sourcetype=SOURCE_FILE;
}
else
{
unsigned int df;
const char *name;
kGUIDir dir;
printf("loading directory!\n");
dir.LoadDir(p2,true,true);
for(df=0; df<dir.GetNumFiles(); ++df)
{
name=dir.GetFilename(df);
fdata.time=kGUI::SysFileTime(name);
AddHash.Add(name,&fdata);
}
// scandir(&AddHash,p2);
sourcetype=SOURCE_DIR;
}
/* now, look for differences between bigfile and files in the addhash list */
numok=0;
numnew=0;
numupdated=0;
/* todo: optdelete function */
/* add from source bigfile to dest bigfile will not work */
/* if source is encrypted so I need to rewrite addfile to use a datahandle */
/* instead of a filestream */
if(sourcetype==SOURCE_BIG)
{
BigFileEntry *sfe;
nums=SourceBigFile.GetNumEntries();
for(i=0; ((i<nums) && (g_userabort==false)); ++i)
{
sfe=(BigFileEntry *)SourceBigFile.GetEntry(i);
bfe=DestBigFile.Locate(sfe->GetName()->GetString());
if(!bfe)
{
// printf("File '%s' not in destination set!\n",she->m_string);
update=true;
++numnew;
}
else
{
if(sfe->m_time==bfe->m_time)
{
update=false;
++numok;
}
else
{
int deltatime;
deltatime=abs(sfe->m_time-bfe->m_time);
if(deltatime==46400 || deltatime==3600)
{
update=false;
++numok;
}
else
{
printf("File '%s' %d,%d times are different!(%d)\n",sfe->GetName()->GetString(),sfe->m_time,bfe->m_time,deltatime);
++numupdated;
update=true;
}
}
}
if(update==true)
{
SourceBigFile.CopyArea(&addhandle,sfe->m_offset,sfe->m_size,sfe->m_time);
// addsize=sfe->m_size;
// addtime=sfe->m_time;
// fseek(addhandle,sfe->m_offset,SEEK_SET);
/* add the file to the bigfile */
DestBigFile.AddFile(sfe->GetName()->GetString(),&addhandle,false);
}
}
DestBigFile.UpdateDir();
}
else
{
HashEntry *she;
FDATA *sfdata;
nums=AddHash.GetNum();
she=AddHash.GetFirst();
for(i=0; ((i<nums) && (g_userabort==false)); ++i)
{
sfdata=(FDATA *)she->m_data;
bfe=DestBigFile.Locate(she->m_string+tosspath);
if(!bfe)
{
// printf("File '%s' not in destination set!\n",she->m_string+tosspath);
++numnew;
update=true;
}
else
{
if(sfdata->time==bfe->m_time)
{
update=false;
++numok;
}
else
{
int deltatime;
deltatime=(abs((int)sfdata->time-bfe->m_time));
if(deltatime==46400 || deltatime==3600)
{
update=false;
++numok;
}
else
{
printf("File '%s' %d,%d times are different!(%d)\n",she->m_string+tosspath,(int)sfdata->time,bfe->m_time,deltatime);
++numupdated;
update=true;
}
}
}
if(update==true)
{
addhandle.SetFilename(she->m_string);
/* add the file to the bigfile */
DestBigFile.AddFile(she->m_string+tosspath,&addhandle,false);
}
she=she->m_next;
}
DestBigFile.UpdateDir();
}
printf("numok=%d,numnew=%d,numupdated=%d\n",numok,numnew,numupdated);
return 0;
}