bool loadUrls ( ) {
static bool s_loaded = false;
if ( s_loaded ) return true;
s_loaded = true;
// use injectme3 file
s_ubuf1.load("./injectme3");
// scan for +++URL: xxxxx
char *s = s_ubuf1.getBufStart();
for ( ; *s ; s++ ) {
if ( strncmp(s,"+++URL: ",8) ) continue;
// got one
// \0 term it for s_contentPtrs below
*s = '\0';
// find end of it
s += 8;
char *e = s;
for ( ; *e && ! is_wspace_a(*e); e++ );
// null term it
if ( *e ) *e = '\0';
// store ptr
s_ubuf2.pushLong((long)s);
// skip past that
s = e;
// point to content
s_cbuf2.pushLong((long)(s+1));
}
// make array of url ptrs
s_urlPtrs = (char **)s_ubuf2.getBufStart();
s_contentPtrs= (char **)s_cbuf2.getBufStart();
return true;
}
bool Msg3a::gotAllSplitReplies ( ) {
// if any of the split requests had an error, give up and set m_errno
// but don't set if for non critical errors like query truncation
if ( m_errno ) {
g_errno = m_errno;
return true;
}
// also reset the finalbuf and the oldNumTopDocIds
if ( m_finalBuf ) {
mfree ( m_finalBuf, m_finalBufSize, "Msg3aF" );
m_finalBuf = NULL;
m_finalBufSize = 0;
}
// update our estimated total hits
m_numTotalEstimatedHits = 0;
for ( long i = 0; i < m_numHosts ; i++ ) {
// get that host that gave us the reply
//Host *h = g_hostdb.getHost(i);
// . get the reply from multicast
// . multicast should have destroyed all slots, but saved reply
// . we are responsible for freeing the reply
// . we need to call this even if g_errno or m_errno is
// set so we can free the replies in Msg3a::reset()
// . if we don't call getBestReply() on it multicast should
// free it, because Multicast::m_ownReadBuf is still true
Multicast *m = &m_mcast[i];
bool freeit = false;
long replySize = 0;
long replyMaxSize;
char *rbuf;
Msg39Reply *mr;
// . only get it if the reply not already full
// . if reply already processed, skip
// . perhaps it had no more docids to give us or all termlists
// were exhausted on its disk and this is a re-call
// . we have to re-process it for count m_numTotalEstHits, etc.
rbuf = m->getBestReply ( &replySize ,
&replyMaxSize ,
&freeit ,
true ); //stealIt?
// cast it
mr = (Msg39Reply *)rbuf;
// in case of mem leak, re-label from "mcast" to this so we
// can determine where it came from, "Msg3a-GBR"
relabel( rbuf, replyMaxSize , "Msg3a-GBR" );
// . we must be able to free it... we must own it
// . this is true if we should free it, but we should not have
// to free it since it is owned by the slot?
if ( freeit ) {
log(LOG_LOGIC,"query: msg3a: Steal failed.");
char *xx = NULL;
*xx=0;
}
// bad reply?
if ( ! mr ) {
log(LOG_LOGIC,"query: msg3a: Bad NULL reply.");
m_reply [i] = NULL;
m_replyMaxSize[i] = 0;
// it might have been timd out, just ignore it!!
continue;
// if size is 0 it can be Msg39 giving us an error!
g_errno = EBADREPLYSIZE;
m_errno = EBADREPLYSIZE;
// all reply buffers should be freed on reset()
return true;
}
// how did this happen?
if ( replySize < 29 && ! mr->m_errno ) {
// if size is 0 it can be Msg39 giving us an error!
g_errno = EBADREPLYSIZE;
m_errno = EBADREPLYSIZE;
log(LOG_LOGIC,"query: msg3a: Bad reply size of %li.",
replySize);
// all reply buffers should be freed on reset()
return true;
}
// can this be non-null? we shouldn't be overwriting one
// without freeing it...
if ( m_reply[i] )
// note the mem leak now
log("query: mem leaking a 0x39 reply");
// cast it and set it
m_reply [i] = mr;
m_replyMaxSize[i] = replyMaxSize;
// deserialize it (just sets the ptr_ and size_ member vars)
//mr->deserialize ( );
deserializeMsg ( sizeof(Msg39Reply) ,
&mr->size_docIds,
&mr->size_clusterRecs,
&mr->ptr_docIds,
mr->m_buf );
// sanity check
if ( mr->m_nqt != m_q->getNumTerms() ) {
g_errno = EBADREPLY;
m_errno = EBADREPLY;
log("query: msg3a: Split reply qterms=%li != %li.",
(long)mr->m_nqt,(long)m_q->getNumTerms() );
return true;
}
// return if split had an error, but not for a non-critical
// error like query truncation
if ( mr->m_errno && mr->m_errno != EQUERYTRUNCATED ) {
g_errno = mr->m_errno;
m_errno = mr->m_errno;
log("query: msg3a: Split had error: %s",
mstrerror(g_errno));
return true;
}
// skip down here if reply was already set
//skip:
// add of the total hits from each split, this is how many
// total results the lastest split is estimated to be able to
// return
// . THIS should now be exact since we read all termlists
// of posdb...
m_numTotalEstimatedHits += mr->m_estimatedHits;
// debug log stuff
if ( ! m_debug ) continue;
// cast these for printing out
long long *docIds = (long long *)mr->ptr_docIds;
score_t *scores = (score_t *)mr->ptr_scores;
// print out every docid in this split reply
for ( long j = 0; j < mr->m_numDocIds ; j++ ) {
// print out score_t
logf( LOG_DEBUG,
"query: msg3a: [%lu] %03li) "
"split=%li docId=%012llu domHash=0x%02lx "
"score=%lu" ,
(unsigned long)this ,
j ,
i ,
docIds [j] ,
(long)g_titledb.getDomHash8FromDocId(docIds[j]),
(long)scores[j] );
}
}
// this seems to always return true!
mergeLists ( );
if ( ! m_r->m_useSeoResultsCache ) return true;
// now cache the reply
SafeBuf cr;
long dataSize = 4 + 4 + 4 + m_numDocIds * (8+4+4);
long need = sizeof(key_t) + 4 + dataSize;
bool status = cr.reserve ( need );
// sanity
if ( ( m_ckey.n0 & 0x01 ) == 0x00 ) {
char *xx=NULL;
*xx=0;
}
// ignore errors
g_errno = 0;
// return on error with g_errno cleared if cache add failed
if ( ! status ) return true;
// add to buf otherwise
cr.safeMemcpy ( &m_ckey , sizeof(key_t) );
cr.safeMemcpy ( &dataSize , 4 );
long now = getTimeGlobal();
cr.pushLong ( now );
cr.pushLong ( m_numDocIds );
cr.pushLong ( m_numTotalEstimatedHits );//Results );
long max = m_numDocIds;
// then the docids
for ( long i = 0 ; i < max ; i++ )
cr.pushLongLong(m_docIds[i] );
for ( long i = 0 ; i < max ; i++ )
cr.pushFloat(m_scores[i]);
for ( long i = 0 ; i < max ; i++ )
cr.pushLong(getSiteHash26(i));
// sanity
if ( cr.length() != need ) {
char *xx=NULL;
*xx=0;
}
// make these
key_t startKey;
key_t endKey;
startKey = m_ckey;
// clear delbit
startKey.n0 &= 0xfffffffffffffffeLL;
// end key is us
endKey = m_ckey;
// that is the single record
m_seoCacheList.set ( cr.getBufStart() ,
cr.length(),
cr.getBufStart(), // alloc
cr.getCapacity(), // alloc size
(char *)&startKey,
(char *)&endKey,
-1, // fixeddatasize
true, // owndata?
false,// use half keys?
sizeof(key_t) );
// do not allow cr to free it, msg1 will
cr.detachBuf();
// note it
//log("seopipe: storing ckey=%s q=%s"
// ,KEYSTR(&m_ckey,12)
// ,m_r->ptr_query
// );
//log("msg1: sending niceness=%li",(long)m_r->m_niceness);
// this will often block, but who cares!? it just sends a request off
if ( ! m_msg1.addList ( &m_seoCacheList ,
RDB_SERPDB,//RDB_CACHEDB,
m_r->ptr_coll,
this, // state
gotSerpdbReplyWrapper, // callback
false, // forcelocal?
m_r->m_niceness ) ) {
//log("blocked");
return false;
}
// we can safely delete m_msg17... just return true
return true;
}
//
// . 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;
}