HPCC-9397 Roxie remote file resolution issues

Fix the remote group resolution code, so that the cache works better.

Signed-off-by: Richard Chapman <rchapman@hpccsystems.com>

dali/base/dadfs.cpp

@@ -6096,20 +6096,6 @@ public:
             return NULL;
         gname.toLowerCase();
         logicalgroupname = gname.str();
-        if ((gname.length()>9)&&(memcmp(logicalgroupname,"foreign::",9)==0)) {
-            StringBuffer eps;
-            const char *s = logicalgroupname+9;
-            while (*s&&((*s!=':')||(s[1]!=':')))
-                eps.append(*(s++));
-            if (*s) {
-                s+=2;
-                if (*s) {
-                    Owned<INode> dali = createINode(eps.str());
-                    if (dali) 
-                        return getRemoteGroup(dali,s,FOREIGN_DALI_TIMEOUT,dirret);
-                }
-            }
-        }
         bool isiprange = (*logicalgroupname!=0);
         for (const char *s1=logicalgroupname;*s1;s1++)
             if (isalpha(*s1)) {
@@ -6173,7 +6159,21 @@ public:
                 }
             }
         }
-        if (epa.ordinality()==0) {
+        if ((gname.length()>9)&&(memcmp(logicalgroupname,"foreign::",9)==0)) {
+            StringBuffer eps;
+            const char *s = logicalgroupname+9;
+            while (*s&&((*s!=':')||(s[1]!=':')))
+                eps.append(*(s++));
+            if (*s) {
+                s+=2;
+                if (*s) {
+                    Owned<INode> dali = createINode(eps.str());
+                    if (!dali || !getRemoteGroup(epa, dali, s, FOREIGN_DALI_TIMEOUT, groupdir))
+                        return NULL;
+                }
+            }
+        }
+        else if (epa.ordinality()==0) {
             struct sLock
             {
                 sLock()  { lock = NULL; };
@@ -6413,7 +6413,15 @@ public:
         cachedgroupdir.clear();
     }
 
-    IGroup *getRemoteGroup(const INode *foreigndali, const char *gname, unsigned foreigndalitimeout, StringBuffer *dirret)
+    unsigned setDefaultTimeout(unsigned timems)
+    {
+        unsigned ret = defaultTimeout;
+        defaultTimeout = timems;
+        return ret;
+    }
+
+private:
+    bool getRemoteGroup(SocketEndpointArray &epa, const INode *foreigndali, const char *gname, unsigned foreigndalitimeout, StringAttr &groupdir)
     {
         StringBuffer lcname(gname);
         gname = lcname.trim().toLowerCase().str();
@@ -6423,7 +6431,7 @@ public:
         foreignDaliSendRecv(foreigndali,mb,foreigndalitimeout);
         checkDfsReplyException(mb);
         if (mb.length()==0)
-            return NULL;
+            return false;
         byte ok;
         mb.read(ok);
         if (ok!=1) {
@@ -6432,39 +6440,21 @@ public:
                 mb.skip(mbsz-1);
                 mb.read(ok);
                 if (ok!=1) 
-                    return NULL;
+                    return false;
             }
             else
-                return NULL;
+                return false;
         }
         Owned<IPropertyTree> pt = createPTree(mb);
         Owned<IPropertyTreeIterator> pe = pt->getElements("Node");
-        SocketEndpointArray epa;
+        groupdir.set(pt->queryProp("@dir"));
         ForEach(*pe) {
             SocketEndpoint ep(pe->query().queryProp("@ip"));
             epa.append(ep);
         }
-        IGroup *ret = createIGroup(epa);
-        {
-            CriticalBlock block(cachesect);
-            cachedgroup.set(ret);
-            cachedname.set(gname);
-            cachedgroupdir.set(pt->queryProp("@dir"));
-            if (dirret)
-                dirret->append(cachedgroupdir);
-            cachedtime = msTick();
-        }
-        return ret;
-    }
-
-    unsigned setDefaultTimeout(unsigned timems)
-    {
-        unsigned ret = defaultTimeout;
-        defaultTimeout = timems;
-        return ret;
+        return epa.ordinality() > 0;
     }
 
-
 };
 
 static CNamedGroupStore *groupStore = NULL;

dali/base/dadfs.hpp

@@ -594,7 +594,6 @@ interface INamedGroupStore: implements IGroupResolver
     virtual bool find(IGroup *grp, StringBuffer &lname, bool add=false) = 0;
     virtual void addUnique(IGroup *group,StringBuffer &lname,const char *dir=NULL) = 0;
     virtual void swapNode(const IpAddress &from, const IpAddress &to) = 0;
-    virtual IGroup *getRemoteGroup(const INode *foreigndali, const char *gname, unsigned foreigndalitimeout=FOREIGN_DALI_TIMEOUT, StringBuffer *dir=NULL) = 0;
     virtual IGroup *lookup(const char *logicalgroupname, StringBuffer &dir) = 0;
     virtual unsigned setDefaultTimeout(unsigned timems) = 0;                                    // sets default timeout for SDS connections and locking                                                                                         // returns previous value
 

dali/dfu/dfuutil.cpp

@@ -402,8 +402,6 @@ class CFileCloner
             throw afor2.exc.getClear();
     }
 
-
-
     void cloneSubFile(IPropertyTree *ftree,const char *destfilename, INode *srcdali)   // name already has prefix added
     {
         Owned<IFileDescriptor> srcfdesc = deserializeFileDescriptorTree(ftree,&queryNamedGroupStore(),0);
@@ -437,11 +435,6 @@ class CFileCloner
         dstfdesc->addCluster(cluster1,grp1,spec);
         if (iskey&&!cluster2.isEmpty())
             dstfdesc->addCluster(cluster2,grp2,spec2);
-#ifdef _TESTING
-//      LOGFDESC("createFileClone",dstfdesc);
-#endif
-
-
 
         for (unsigned pn=0;pn<srcfdesc->numParts();pn++) {
             offset_t sz = srcfdesc->queryPart(pn)->queryProperties().getPropInt64("@size",-1);
@@ -461,6 +454,17 @@ class CFileCloner
         {
             StringBuffer s;
             dstfdesc->queryProperties().setProp("@cloneFrom", srcdali->endpoint().getUrlStr(s).str());
+            unsigned numClusters = srcfdesc->numClusters();
+            for (unsigned clusterNum = 0; clusterNum < numClusters; clusterNum++)
+            {
+                StringBuffer sourceGroup;
+                srcfdesc->getClusterGroupName(clusterNum, sourceGroup, NULL);
+                Owned<IPropertyTree> groupInfo = createPTree("cloneFromGroup");
+                groupInfo->setProp("@groupName", sourceGroup);
+                ClusterPartDiskMapSpec &spec = srcfdesc->queryPartDiskMapping(clusterNum);
+                spec.toProp(groupInfo);
+                dstfdesc->queryProperties().addPropTree("cloneFromGroup", groupInfo.getClear());
+            }
         }
 
         Owned<IDistributedFile> dstfile = fdir->createNew(dstfdesc);
@@ -552,7 +556,6 @@ public:
         }
     }
 
-
     void cloneSuperFile(const char *filename, CDfsLogicalFileName &dlfn)
     {
         level++;

dali/dfu/dfuutil.hpp

@@ -75,7 +75,6 @@ interface IDFUhelper: extends IInterface
         IPropertyTree *relationships,   // if not NULL, tree will have all relationships filled in
         IUserDescriptor *user
     ) = 0;
-
 };
 
 IDFUhelper *createIDFUhelper();

roxie/ccd/ccddali.cpp

@@ -274,6 +274,56 @@ private:
         return localTree.getClear();
     }
 
+    IFileDescriptor *recreateCloneSource(IFileDescriptor *srcfdesc, const char *destfilename)
+    {
+        const char * kind = srcfdesc->queryProperties().queryProp("@kind");
+        bool iskey = kind&&(strcmp(kind,"key")==0);
+
+        Owned<IFileDescriptor> dstfdesc = createFileDescriptor(srcfdesc->getProperties());
+        // calculate dest dir
+
+        CDfsLogicalFileName dstlfn;
+        if (!dstlfn.setValidate(destfilename,true))
+            throw MakeStringException(-1,"Logical name %s invalid",destfilename);
+
+        StringBuffer dstpartmask;
+        getPartMask(dstpartmask,destfilename,srcfdesc->numParts());
+        dstfdesc->setPartMask(dstpartmask.str());
+        unsigned np = srcfdesc->numParts();
+        dstfdesc->setNumParts(srcfdesc->numParts());
+        DFD_OS os = (getPathSepChar(srcfdesc->queryDefaultDir())=='\\')?DFD_OSwindows:DFD_OSunix;
+        StringBuffer dir;
+        StringBuffer dstdir;
+        makePhysicalPartName(dstlfn.get(),0,0,dstdir,false,os,NULL);
+        dstfdesc->setDefaultDir(dstdir.str());
+
+        for (unsigned pn=0;pn<srcfdesc->numParts();pn++) {
+            offset_t sz = srcfdesc->queryPart(pn)->queryProperties().getPropInt64("@size",-1);
+            if (sz!=(offset_t)-1)
+                dstfdesc->queryPart(pn)->queryProperties().setPropInt64("@size",sz);
+            StringBuffer dates;
+            if (srcfdesc->queryPart(pn)->queryProperties().getProp("@modified",dates))
+                dstfdesc->queryPart(pn)->queryProperties().setProp("@modified",dates.str());
+        }
+
+        const char *cloneFrom = srcfdesc->queryProperties().queryProp("@cloneFrom");
+        Owned<IPropertyTreeIterator> groups = srcfdesc->queryProperties().getElements("cloneFromGroup");
+        ForEach(*groups)
+        {
+            IPropertyTree &elem = groups->query();
+            const char *groupName = elem.queryProp("@groupName");
+            StringBuffer foreignGroup("foreign::");
+            foreignGroup.append(cloneFrom).append("::").append(groupName);
+            Owned<IGroup> group = queryNamedGroupStore().lookup(foreignGroup);  // NOTE - this is cached by the named group store
+            ClusterPartDiskMapSpec dmSpec;
+            dmSpec.fromProp(&elem);
+            dstfdesc->addCluster(groupName, group, dmSpec);
+        }
+
+        return dstfdesc.getClear();
+    }
+
+
 public:
 
     IMPLEMENT_IINTERFACE;
@@ -351,24 +401,31 @@ public:
             return NULL;
         if (!fdesc || !fdesc->queryProperties().hasProp("@cloneFrom"))
             return NULL;
-        SocketEndpoint cloneFrom;
-        cloneFrom.set(fdesc->queryProperties().queryProp("@cloneFrom"));
-        if (cloneFrom.isNull())
-            return NULL;
-        CDfsLogicalFileName lfn;
-        lfn.set(_lfn);
-        lfn.setForeign(cloneFrom, false);
-        if (!connected())
-            return resolveCachedLFN(lfn.get());
-        Owned<IDistributedFile> cloneFile = resolveLFN(lfn.get(), cacheIt, false);
-        if (cloneFile)
+        if (fdesc->queryProperties().hasProp("cloneFromGroup"))
+        {
+            return recreateCloneSource(fdesc, _lfn);
+        }
+        else // Legacy mode - recently cloned files should have the extra info
         {
-            Owned<IFileDescriptor> cloneFDesc = cloneFile->getFileDescriptor();
-            if (cloneFDesc->numParts()==fdesc->numParts())
-                return cloneFDesc.getClear();
+            SocketEndpoint cloneFrom;
+            cloneFrom.set(fdesc->queryProperties().queryProp("@cloneFrom"));
+            if (cloneFrom.isNull())
+                return NULL;
+            CDfsLogicalFileName lfn;
+            lfn.set(_lfn);
+            lfn.setForeign(cloneFrom, false);
+            if (!connected())
+                return resolveCachedLFN(lfn.get());
+            Owned<IDistributedFile> cloneFile = resolveLFN(lfn.get(), cacheIt, false);
+            if (cloneFile)
+            {
+                Owned<IFileDescriptor> cloneFDesc = cloneFile->getFileDescriptor();
+                if (cloneFDesc->numParts()==fdesc->numParts())
+                    return cloneFDesc.getClear();
 
-            StringBuffer s;
-            DBGLOG(ROXIE_MISMATCH, "File %s cloneFrom(%s) mismatch", _lfn, cloneFrom.getIpText(s).str());
+                StringBuffer s;
+                DBGLOG(ROXIE_MISMATCH, "File %s cloneFrom(%s) mismatch", _lfn, cloneFrom.getIpText(s).str());
+            }
         }
         return NULL;
     }