// . the url being reuqested
// . removes &code= facebook cruft
bool HttpRequest::getCurrentUrl ( SafeBuf &cu ) {
// makre sure we got enough room
if ( ! cu.reserve ( m_hostLen + 64 + m_plen + 1 + 1 ) ) return false;
// need a "Host: "
char *host = m_host;
if ( ! host ) host = APPSUBDOMAIN;
cu.safePrintf("http");
if ( m_isSSL ) cu.pushChar('s');
cu.safePrintf("://%s",host);
char *path = m_path;
long plen = m_plen;
if ( ! path ) {
path = "/";
plen = 1;
}
// . scan path and change \0 back to = or &
// . similar logic in HttpServer.cpp for logging!
char *dst = cu.getBuf();
char *src = path;
char *srcEnd = path + plen;
char dd = '=';
for ( ; src < srcEnd ; src++ , dst++ ) {
*dst = *src;
if ( *src ) continue;
*dst = dd;
if ( dd == '=' ) dd = '&';
else dd = '=';
}
*dst = '\0';
// cut it off at facebook's &code=
char *buf = cu.getBufStart();
char *code = strstr( buf,"&code=");
// fix for eventguru.com/blog.html?code=
if ( ! code ) code = strstr(buf,"?code=");
// hack that off if there
if ( code ) {
*code = '\0';
dst = code;
}
// update length
cu.setLength( dst - cu.getBufStart() );
return true;
}
bool HttpRequest::getCurrentUrlPath ( SafeBuf &cup ) {
// makre sure we got enough room
if ( ! cup.reserve ( m_plen + 1 + 1 ) ) return false;
char *path = m_path;
long plen = m_plen;
if ( ! path ) {
path = "/";
plen = 1;
}
// . scan path and change \0 back to = or &
// . similar logic in HttpServer.cpp for logging!
char *dst = cup.getBuf();
char *start = dst;
char *src = path;
char *srcEnd = path + plen;
// stop if we hit '?'
for ( ; src < srcEnd && *src != '?' ; src++ , dst++ ) {
*dst = *src;
}
cup.incrementLength(dst - start);
*dst = '\0';
return true;
}
void Msg39::estimateHitsAndSendReply ( ) {
// no longer in use
m_inUse = false;
// now this for the query loop on the QueryLogEntries.
m_topDocId50 = 0LL;
m_topScore50 = 0.0;
// a little hack for the seo pipeline in xmldoc.cpp
m_topDocId = 0LL;
m_topScore = 0.0;
m_topDocId2 = 0LL;
m_topScore2 = 0.0;
int32_t ti = m_tt.getHighNode();
if ( ti >= 0 ) {
TopNode *t = &m_tt.m_nodes[ti];
m_topDocId = t->m_docId;
m_topScore = t->m_score;
}
// try the 2nd one too
int32_t ti2 = -1;
if ( ti >= 0 ) ti2 = m_tt.getNext ( ti );
if ( ti2 >= 0 ) {
TopNode *t2 = &m_tt.m_nodes[ti2];
m_topDocId2 = t2->m_docId;
m_topScore2 = t2->m_score;
}
// convenience ptrs. we will store the docids/scores into these arrays
int64_t *topDocIds;
double *topScores;
key_t *topRecs;
// numDocIds counts docs in all tiers when using toptree.
int32_t numDocIds = m_tt.m_numUsedNodes;
// the msg39 reply we send back
int32_t replySize;
char *reply;
//m_numTotalHits = m_posdbTable.m_docIdVoteBuf.length() / 6;
// make the reply?
Msg39Reply mr;
// this is what you want to look at if there is no seo.cpp module...
if ( ! m_callback ) {
// if we got clusterdb recs in here, use 'em
if ( m_gotClusterRecs ) numDocIds = m_numVisible;
// don't send more than the docs that are asked for
if ( numDocIds > m_r->m_docsToGet) numDocIds =m_r->m_docsToGet;
// # of QueryTerms in query
int32_t nqt = m_tmpq.m_numTerms;
// start setting the stuff
mr.m_numDocIds = numDocIds;
// copy # estiamted hits into 8 bytes of reply
//int64_t est = m_posdbTable.m_estimatedTotalHits;
// ensure it has at least as many results as we got
//if ( est < numDocIds ) est = numDocIds;
// or if too big...
//if ( numDocIds < m_r->m_docsToGet ) est = numDocIds;
// . total estimated hits
// . this is now an EXACT count!
mr.m_estimatedHits = m_numTotalHits;
// sanity check
mr.m_nqt = nqt;
// the m_errno if any
mr.m_errno = m_errno;
// int16_tcut
PosdbTable *pt = &m_posdbTable;
// the score info, in no particular order right now
mr.ptr_scoreInfo = pt->m_scoreInfoBuf.getBufStart();
mr.size_scoreInfo = pt->m_scoreInfoBuf.length();
// that has offset references into posdbtable::m_pairScoreBuf
// and m_singleScoreBuf, so we need those too now
mr.ptr_pairScoreBuf = pt->m_pairScoreBuf.getBufStart();
mr.size_pairScoreBuf = pt->m_pairScoreBuf.length();
mr.ptr_singleScoreBuf = pt->m_singleScoreBuf.getBufStart();
mr.size_singleScoreBuf = pt->m_singleScoreBuf.length();
// save some time since seo.cpp gets from posdbtable directly,
// so we can avoid serializing/copying this stuff at least
if ( ! m_r->m_makeReply ) {
mr.size_scoreInfo = 0;
mr.size_pairScoreBuf = 0;
mr.size_singleScoreBuf = 0;
}
//mr.m_sectionStats = pt->m_sectionStats;
// reserve space for these guys, we fill them in below
mr.ptr_docIds = NULL;
mr.ptr_scores = NULL;
mr.ptr_clusterRecs = NULL;
// this is how much space to reserve
mr.size_docIds = 8 * numDocIds; // int64_t
mr.size_scores = sizeof(double) * numDocIds; // float
// if not doing site clustering, we won't have these perhaps...
if ( m_gotClusterRecs )
mr.size_clusterRecs = sizeof(key_t) *numDocIds;
else
mr.size_clusterRecs = 0;
#define MAX_FACETS 20000
/////////////////
//
// FACETS
//
/////////////////
// We can have multiple gbfacet: terms in a query so
// serialize all the QueryTerm::m_facetHashTables into
// Msg39Reply::ptr_facetHashList.
//
// combine the facet hash lists of each query term into
// a list of lists. each lsit is preceeded by the query term
// id of the query term (like gbfacet:xpathsitehash12345)
// followed by a 4 byte length of the following 32-bit
// facet values
int32_t need = 0;
for ( int32_t i = 0 ; i < m_tmpq.m_numTerms; i++ ) {
QueryTerm *qt = &m_tmpq.m_qterms[i];
// skip if not facet
if ( qt->m_fieldCode != FIELD_GBFACETSTR &&
qt->m_fieldCode != FIELD_GBFACETINT &&
qt->m_fieldCode != FIELD_GBFACETFLOAT )
continue;
HashTableX *ft = &qt->m_facetHashTable;
if ( ft->m_numSlotsUsed == 0 ) continue;
int32_t used = ft->m_numSlotsUsed;
// limit for memory
if ( used > (int32_t)MAX_FACETS ) {
log("msg39: truncating facet list to 20000 "
"from %"INT32" for %s",used,qt->m_term);
used = (int32_t)MAX_FACETS;
}
// store query term id 64 bit
need += 8;
// then size
need += 4;
// then buckets. keys and counts
need += (4+sizeof(FacetEntry)) * used;
}
// allocate
SafeBuf tmp;
if ( ! tmp.reserve ( need ) ) {
log("query: Could not allocate memory "
"to hold reply facets");
sendReply(m_slot,this,NULL,0,0,true);
return;
}
// point to there
char *p = tmp.getBufStart();
for ( int32_t i = 0 ; i < m_tmpq.m_numTerms ; i++ ) {
QueryTerm *qt = &m_tmpq.m_qterms[i];
// skip if not facet
if ( qt->m_fieldCode != FIELD_GBFACETSTR &&
qt->m_fieldCode != FIELD_GBFACETINT &&
qt->m_fieldCode != FIELD_GBFACETFLOAT )
continue;
// get all the facet hashes and their counts
HashTableX *ft = &qt->m_facetHashTable;
// skip if none
if ( ft->m_numSlotsUsed == 0 ) continue;
// store query term id 64 bit
*(int64_t *)p = qt->m_termId;
p += 8;
int32_t used = ft->getNumSlotsUsed();
if ( used > (int32_t)MAX_FACETS )
used = (int32_t)MAX_FACETS;
// store count
*(int32_t *)p = used;
p += 4;
int32_t count = 0;
// for sanity check
char *pend = p + (used * (4+sizeof(FacetEntry)));
// serialize the key/val pairs
for ( int32_t k = 0 ; k < ft->m_numSlots ; k++ ) {
// skip empty buckets
if ( ! ft->m_flags[k] ) continue;
// store key. the hash of the facet value.
*(int32_t *)p = ft->getKey32FromSlot(k); p += 4;
// then store count
//*(int32_t *)p = ft->getVal32FromSlot(k); p += 4;
// now this has a docid on it so we can
// lookup the text of the facet in Msg40.cpp
FacetEntry *fe;
fe = (FacetEntry *)ft->getValFromSlot(k);
// sanity
// no, count can be zero if its a range facet
// that was never added to. we add those
// empty FaceEntries only for range facets
// in Posdb.cpp
//if(fe->m_count == 0 ) { char *xx=NULL;*xx=0;}
gbmemcpy ( p , fe , sizeof(FacetEntry) );
p += sizeof(FacetEntry);
// do not breach
if ( ++count >= (int32_t)MAX_FACETS ) break;
}
// sanity check
if ( p != pend ) { char *xx=NULL;*xx=0; }
// do the next query term
}
// now point to that so it can be serialized below
mr.ptr_facetHashList = tmp.getBufStart();
mr.size_facetHashList = p - tmp.getBufStart();//tmp.length();
/////////////
//
// END FACETS
//
/////////////
// . that is pretty much it,so serialize it into buffer,"reply"
// . mr.ptr_docIds, etc., will point into the buffer so we can
// re-serialize into it below from the tree
// . returns NULL and sets g_errno on error
// . "true" means we should make mr.ptr_* reference into the
// newly serialized buffer.
reply = serializeMsg ( sizeof(Msg39Reply), // baseSize
&mr.size_docIds, // firstSizeParm
&mr.size_clusterRecs,//lastSizePrm
&mr.ptr_docIds , // firstStrPtr
&mr , // thisPtr
&replySize ,
NULL ,
0 ,
true ) ;
if ( ! reply ) {
log("query: Could not allocated memory "
"to hold reply of docids to send back.");
sendReply(m_slot,this,NULL,0,0,true);
return;
}
topDocIds = (int64_t *) mr.ptr_docIds;
topScores = (double *) mr.ptr_scores;
topRecs = (key_t *) mr.ptr_clusterRecs;
}
int32_t docCount = 0;
// loop over all results in the TopTree
for ( int32_t ti = m_tt.getHighNode() ; ti >= 0 ;
ti = m_tt.getPrev(ti) ) {
// get the guy
TopNode *t = &m_tt.m_nodes[ti];
// skip if clusterLevel is bad!
if ( m_gotClusterRecs && t->m_clusterLevel != CR_OK )
continue;
// if not sending back a reply... we were called from seo.cpp
// State3f logic to evaluate a QueryLogEntry, etc.
if ( m_callback ) {
// skip results past #50
if ( docCount > 50 ) continue;
// set this
m_topScore50 = t->m_score;
m_topDocId50 = t->m_docId;
// that's it
continue;
}
// get the docid ptr
//char *diptr = t->m_docIdPtr;
//int64_t docId = getDocIdFromPtr(diptr);
// sanity check
if ( t->m_docId < 0 ) { char *xx=NULL; *xx=0; }
//add it to the reply
topDocIds [docCount] = t->m_docId;
topScores [docCount] = t->m_score;
if ( m_tt.m_useIntScores )
topScores[docCount] = (double)t->m_intScore;
// supply clusterdb rec? only for full splits
if ( m_gotClusterRecs )
topRecs [docCount] = t->m_clusterRec;
//topExplicits [docCount] =
// getNumBitsOn(t->m_explicits)
docCount++;
// 50th score? set this for seo.cpp. if less than 50 results
// we want the score of the last doc then.
if ( docCount <= 50 ) m_topScore50 = t->m_score;
if ( m_debug ) {
logf(LOG_DEBUG,"query: msg39: [%"PTRFMT"] "
"%03"INT32") docId=%012"UINT64" sum=%.02f",
(PTRTYPE)this, docCount,
t->m_docId,t->m_score);
}
//don't send more than the docs that are wanted
if ( docCount >= numDocIds ) break;
}
if ( docCount > 300 && m_debug )
log("query: Had %"INT32" nodes in top tree",docCount);
// this is sensitive info
if ( m_debug ) {
log(LOG_DEBUG,
"query: msg39: [%"PTRFMT"] "
"Intersected lists took %"INT64" (%"INT64") "
"ms "
"docIdsToGet=%"INT32" docIdsGot=%"INT32" "
"q=%s",
(PTRTYPE)this ,
m_posdbTable.m_addListsTime ,
gettimeofdayInMilliseconds() - m_startTime ,
m_r->m_docsToGet ,
numDocIds ,
m_tmpq.getQuery() );
}
// if we blocked because we used a thread then call callback if
// summoned from a msg3f handler and not a msg39 handler
if ( m_callback ) {
// if we blocked call user callback
if ( m_blocked ) m_callback ( m_state );
// if not sending back a udp reply, return now
return;
}
// now send back the reply
sendReply(m_slot,this,reply,replySize,replySize,false);
return;
}
// . displays the stats for a username
// . show stats for every day we have them for
// . in a big list
// . if they click the day display all docids evaluated for that day
// . show the accuracy for that day too
// . how many docs they edited
// . how many of those docs were verified by another
// . and if there was consensus
void gotTransdbList ( State60 *st ) {
// get today's time range
time_t now = getTimeGlobal();
// get start of today
time_t dayStart = now / (24*3600);
SafeBuf sb;
// int16_tcut
TcpSocket *s = st->m_s;
// make about 200k of mem to write into
if ( ! sb.reserve ( 200000 ) )
return g_httpServer.sendErrorReply(s,500,mstrerrno(g_errno));
// print description so they can clikc a button to start the turk
sb.safePrintf("<html>\n"
"<title>Event Editor</title>\n"
"<body>\n"
"<table width=\"100%%\" border=\"0\">\n"
"<tr><td style=\"background-color:#0079ba;\">\n"
"<center><font color=#00000>"
"<h2>Event Editor</h2>\n"
"</font></center></td>"
"</tr></table>");
// print the content
sb.safePrintf("<center><font size=4><blink>"
"<b><a href=\"/pageturk?c=%s&edit=1\">"
"Click here to start editing.</a></b></blink>"
"</font><br><i>Please take your "
"time to read the information below before you begin"
"</i><br><font color=\"red\" size=2> Warning: Adult "
"content might be presented to you."
" You should be above 18 years of age to continue."
"</center></font>",st->m_coll);
sb.safePrintf("<font face=arial,sans-serif color=black size=3>"
"<p>By clicking <i>Start Voting</i>, you will be "
"presented with an interface for editing events. "
"The editor will display a modified web page that "
"contains one or more events. Each event's description "
"will be highlight with a blue background. You can "
"toggle whether a particular event is displayed by "
"clicking on that event's ID. You can highlight one or "
"multiple event descriptions at the same time. "
"</p><p>"
"By clicking on the section icons in the web page you "
"can tell the editor that a virtual fence should be "
"erected around that section. The fence will make sure "
"that event descriptions can not span across it. Each "
"event description must be fully contained either "
"inside or outside the fence. However, you can also "
"declare a section as a title section, which means that "
"the text that the title section contains is free to be "
"used by any event description."
"</p>\n"
"<p>When you are done erecting section fences, you "
"submit your changes. The more changes you make the "
"more points you earn. Other users may evaluate "
"your edits for accuracy. You will be paid based on the "
"points you earn as well as your accuracy. All "
"transactions are listed in the table below.</p>"
"<p>You may not change your username or password "
"but you can change your email address. Your email "
"address will be used to pay you with PayPal every "
"Friday. Paypal fees will be deducted on your end. By "
"using this service you agree to all stated Terms & "
"Conditions.</p>"
"</font>\n");
// get the user record
User *uu = g_users.getUser ( username );
// print out their info, like paypal email
sb.safePrintf("<table>\n"
"<tr><td colspan=10><center>Your Info</center>"
"</td></tr>\n"
"<tr>"
"<td>Email</td>"
"<td><input type=text value=%s></td>"
"<td>email address used to pay with paypal</td>"
"</tr>\n"
"<tr><td colspan=10><input type=submit value=update>"
"</td></tr>\n"
"</table>\n" ,
uu->m_payPalEmail );
// print your stats here now
sb.safePrintf("<table>\n"
"<tr><td colspan=10><center>Your Stats</center>"
"</td></tr>\n"
"<tr>"
"<td>date</td>"
"<td>action</td>"
"<td>amount</td>"
"<td>desc</td>"
"</tr>\n");
// int16_tcut
RdbList *list = &st->m_list;
int32_t lastDay = -1;
int32_t totalReceives = 0;
int32_t totalSubmits = 0;
int32_t totalPasses = 0;
int32_t totalFails = 0;
// scan the list
for ( ; ! list->isExhausted() ; ) {
// get rec
char *rec = list->getCurrentRecord();
char *data = list->getCurrentData();
int32_t dataSize = list->getCurrentDataSize();
// skip that
list->skipCurrentRecord();
// skip if negative
if ( (rec[0] & 0x01) == 0x00 ) continue;
// get the time (global time - sync'd with host #0)
time_t tt = g_transdb.getTimeStamp ( rec );
// get day #
int32_t daynum = tt / (24*3600);
// is it today?
bool isToday = ( daynum >= dayStart );
// point to the Transaction
Trans *trans = (Trans *)data;
// if is today, print it out verbatim
if ( isToday ) {
// print it in html row format to match table above
//printTrans ( &sb , rec );
sb.safePrintf("<tr>");
// make it into a nice date
time_t dd = lastDay * 86400;
struct tm *timeStruct = localtime ( &dd );
char ppp[100];
strftime(ppp,100,"%H:%M:%S",timeStruct);
// print last days stats first
sb.safePrintf("<td>%s</td>",ppp);
// then stats
if ( trans->m_actionType == AT_RECEIVE_DOC )
sb.safePrintf("<td>receive</td>"
"<td>%"INT32" pts</td>"
"<td>docid=%"UINT64"</td>",
(int32_t)trans->m_number,
trans->m_docId);
else if ( trans->m_actionType == AT_SUBMIT_DOC )
sb.safePrintf("<td>submit</td>"
"<td>%"INT32" pts</td>"
"<td>docid=%"UINT64"</td>",
(int32_t)trans->m_number,
trans->m_docId);
else if ( trans->m_actionType == AT_PASS_DOC )
sb.safePrintf("<td>verify</td>"
"<td>%"INT32" pts</td>"
"<td>docid=%"UINT64" was verified "
"by user=\"%s\"</td>",
(int32_t)trans->m_number,
trans->m_docId,
trans->m_desc);
else if ( trans->m_actionType == AT_FAIL_DOC )
sb.safePrintf("<td>verify</td>"
"<td>%"INT32" pts</td>"
"<td>docid=%"UINT64" was deemed to "
"be incorrect "
"by user=\"%s\"</td>",
(int32_t)trans->m_number,
trans->m_docId,
trans->m_desc);
else if ( trans->m_actionType == AT_ACCURACY_EVAL)
sb.safePrintf("<td>accuracy eval</td>"
"<td>%.02f</td>"
"<td>docid=%"UINT64"</td>",
trans->m_number,
trans->m_docId);
else if ( trans->m_actionType == AT_CHARGE)
sb.safePrintf("<td>credit</td>"
"<td>%.02f</td>"
"<td>You made money.</td>",
trans->m_number);
else if ( trans->m_actionType == AT_PAYMENT)
sb.safePrintf("<td>payment</td>"
"<td>%.02f</td>"
"<td>We paid you.</td>",
trans->m_number);
else if ( trans->m_actionType == AT_LOGIN)
sb.safePrintf("<td>login</td>"
"<td>-</td>"
"<td>You logged in.</td>");
else if ( trans->m_actionType == AT_LOGOUT)
sb.safePrintf("<td>logout</td>"
"<td>-</td>"
"<td>You logged out.</td>");
else if ( trans->m_actionType == AT_AUTO_LOGOUT)
sb.safePrintf("<td>logout</td>"
"<td>-</td>"
"<td>You were auto "
"logged out.</td>");
else {
char *xx=NULL;*xx=0; }
sb.safePrintf("</tr>\n");
continue;
}
// if does not match last day, print out that last day's stats
// and reset for next guy
if ( daynum != lastDay && lastDay != -1 ) {
// make it into a nice date
time_t dd = lastDay * 86400;
struct tm *timeStruct = localtime ( &dd );
char ppp[100];
strftime(ppp,100,"%b-%d-%Y",timeStruct);
// print last days stats first
sb.safePrintf("<td>%s</td>",ppp);
// then stats
sb.safePrintf("<tr>"
"<td>receive</td>"
"<td>%"INT32"</td>"
"<td>Total received</td>"
"</tr>\n",
totalReceives);
sb.safePrintf("<tr>"
"<td>submit</td>"
"<td>%"INT32"</td>"
"<td>Total submitted</td>"
"</tr>\n",
totalSubmits);
sb.safePrintf("<tr>"
"<td>pass</td>"
"<td>%"INT32"</td>"
"<td>Total accuracy tests passed</td>"
"</tr>\n",
totalPasses);
sb.safePrintf("<tr>"
"<td>fail</td>"
"<td>%"INT32"</td>"
"<td>Total accuracy tests failed</td>"
"</tr>\n",
totalFails);
// reset as well
totalReceived = 0;
totalSubmits = 0;
totalPasses = 0;
totalFails = 0;
}
// remember last day # we processed for accumulating stats
lastDay = daynum;
// accum stats
if ( trans->m_actionType == AT_RECEIVE_DOC )
totalReceives++;
if ( trans->m_actionType == AT_SUBMIT_DOC )
totalSubmits++;
if ( trans->m_actionType == AT_PASS_DOC )
totalPasses++;
if ( trans->m_actionType == AT_FAIL_DOC )
totalFails++;
}
sb.safePrintf("</body></html>\n");
sendReply ( &sb );
}
bool sendReply ( void *state ) {
StateCatdb *st = (StateCatdb*)state;
// check for error
if (g_errno) {
if (st->m_catLookup)
log("PageCatdb: Msg8b had error getting Site Rec: %s",
mstrerror(g_errno));
else
log("PageCatdb: Msg2a had error generating Catdb: %s",
mstrerror(g_errno));
st->m_catLookup = false;
g_errno = 0;
}
long long endTime = gettimeofdayInMilliseconds();
// page buffer
SafeBuf sb;
sb.reserve(64*1024);
// . print standard header
// . do not print big links if only an assassin, just print host ids
g_pages.printAdminTop ( &sb, st->m_socket , &st->m_r );
sb.safePrintf(
"<style>"
".poo { background-color:#%s;}\n"
"</style>\n" ,
LIGHT_BLUE );
sb.safePrintf ( "<table %s>"
"<tr><td colspan=2>"
"<center><font size=+1><b>Catdb</b></font></center>"
"</td></tr>", TABLE_STYLE );
// instructions
sb.safePrintf("<tr bgcolor=#%s>"
"<td colspan=3>"
"<font size=-2>"
"<center>"
"Don't just start using this, you need to follow the "
"instructions in the <i>admin guide</i> for adding "
"DMOZ support."
"</center>"
"</font>"
"</td>"
"</tr>"
,DARK_BLUE
);
// print the generate Catdb link
sb.safePrintf ( "<tr class=poo><td>Update Catdb from DMOZ data.</td>"
"<td><center>"
"<a href=\"/master/catdb?c=%s&gencatdb=2\">"
"Update Catdb</a> "
"</center></td></tr>",
st->m_coll );
sb.safePrintf ( "<tr class=poo>"
"<td>Generate New Catdb from DMOZ data.</td>"
"<td><center>"
"<a href=\"/master/catdb?c=%s&gencatdb=1\">"
"Generate Catdb</a> "
"</center></td></tr>",
st->m_coll );
if (st->m_genCatdb)
sb.safePrintf ( "<tr class=poo>"
"<td> Catdb Generation took %lli ms."
"</td></tr>",
endTime - st->m_startTime );
// print Url Catgory Lookup
sb.safePrintf ( "<tr class=poo><td>Lookup Category of Url.</td>"
"<td><input type=text name=caturl size=80"
" value=\"");
if (st->m_catLookup) {
sb.safeMemcpy(st->m_url.getUrl(), st->m_url.getUrlLen());
}
sb.safePrintf("\"></center></td></tr>" );
// print Url Info if Lookup was done
if (st->m_catLookup) {
sb.safePrintf("<tr><td>");
// print the url
sb.safeMemcpy(st->m_url.getUrl(), st->m_url.getUrlLen());
sb.safePrintf(" (%lli ms)</td><td>",
endTime - st->m_startTime );
// print each category id and path
for (long i = 0; i < st->m_catRec.m_numCatids; i++) {
sb.safePrintf("<b>[%li] ",
st->m_catRec.m_catids[i]);
g_categories->printPathFromId(&sb,
st->m_catRec.m_catids[i]);
sb.safePrintf("</b><br>");
// lookup title and summary
char title[1024];
long titleLen = 0;
char summ[4096];
long summLen = 0;
char anchor[256];
unsigned char anchorLen = 0;
g_categories->getTitleAndSummary(
st->m_url.getUrl(),
st->m_url.getUrlLen(),
st->m_catRec.m_catids[i],
title,
&titleLen,
1023,
summ,
&summLen,
4098,
anchor,
&anchorLen,
255 );
title[titleLen] = '\0';
summ[summLen] = '\0';
anchor[anchorLen] = '\0';
// print title and summary
sb.safePrintf("<b>Title:</b> %s<br>"
"<b>Summary:</b> %s<br>",
title, summ);
if (anchorLen > 0)
sb.safePrintf("<b>Anchor:</b> %s<br>",
anchor);
sb.safePrintf("<br>");
}
sb.safePrintf("<b>Filenum:</b> %li<br>",
st->m_catRec.m_filenum);
// print indirect catids
if (st->m_catRec.m_numIndCatids > 0) {
sb.safePrintf("<hr><b>Indirect Catids [%li]:"
"</b><br>\n",
st->m_catRec.m_numIndCatids );
for (long i = 0;
i < st->m_catRec.m_numIndCatids; i++) {
sb.safePrintf("%lu<br>",
st->m_catRec.m_indCatids[i]);
}
}
sb.safePrintf("</td></tr>");
}
// end it
sb.safePrintf ( "</center></td></tr></table>" );
// print submit button
sb.safePrintf ( "<br><center>"
"<input type=submit value=\"Submit\" border=0>"
"</form></center>" );
// print the final tail
//p += g_httpServer.printTail ( p , pend - p );
// clear g_errno, if any, so our reply send goes through
g_errno = 0;
// extract the socket
TcpSocket *s = st->m_socket;
// clear the state
mdelete ( st, sizeof(StateCatdb), "PageCatdb" );
delete st;
// . send this page
// . encapsulates in html header and tail
// . make a Mime
return g_httpServer.sendDynamicPage(s , sb.getBufStart(), sb.length());
}
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;
}