Changeset 1278

Timestamp:

11/15/08 21:56:36 (22 months ago)

Author:

ahu

Message:

skip remote masters that result in too much timeouts during domain checking cycle. Tested in real life by Tyler Hall on 50000 domains.

Location:

trunk/pdns/pdns

Files:

• communicator.cc (modified) (3 diffs)
• resolver.cc (modified) (1 diff)

trunk/pdns/pdns/communicator.cc

@@ -247 +247 @@
       (sdomains.size()>1 ? "" : "s")<<
       " checking"<<endl;
-
+  map<string, int> skipMasters;
   for(vector<DomainInfo>::iterator i=sdomains.begin();i!=sdomains.end();++i) {
     Resolver resolver;   
@@ -262 +262 @@
     for(vector<string>::const_iterator iter = i->masters.begin(); iter != i->masters.end(); ++iter) {
       try {
+        if(skipMasters[*iter] > 5)
+          throw AhuException("Skipping query to '"+*iter+"' because of previous timeouts in this cycle");
+        
         resolver.getSoaSerial(*iter, i->zone, &theirserial);
-        
+        skipMasters[*iter]=0;   
         if(theirserial<i->serial) {
           L<<Logger::Error<<"Domain "<<i->zone<<" more recent than master, our serial "<<ourserial<<" > their serial "<<theirserial<<endl;
@@ -279 +282 @@
       }
       catch(ResolverException &re) {
+        if(re.reason.find("Timeout") != string::npos)
+          skipMasters[*iter]++;
+
         L<<Logger::Error<<"Error trying to retrieve/refresh '"+i->zone+"': "+re.reason<<endl;
         if(next(iter) != i->masters.end()) 

trunk/pdns/pdns/resolver.cc

@@ -185 +185 @@
 
   struct timeval timeout;
-  timeout.tv_sec=1;
-  timeout.tv_usec=500000;
+  timeout.tv_sec=0;
+  timeout.tv_usec=750000;
 
   int res=select(d_sock+1,&rd,0,0,&timeout);