bug#29099 - slow backup for disk data

- implement read ahead during disk data scan


storage/ndb/src/kernel/blocks/dbtup/Dbtup.hpp:
  maximum read ahead
storage/ndb/src/kernel/blocks/dbtup/DbtupGen.cpp:
  maximum read ahead should not be larger than page buffer size
  maximum aslo set to 32 pages
storage/ndb/src/kernel/blocks/dbtup/DbtupScan.cpp:
  read ahead for disk data scan
storage/ndb/src/kernel/blocks/pgman.cpp:
  set higher max_io_waits for read ahead
  round off upwards in page_buffer size
storage/ndb/src/kernel/vm/SimulatedBlock.cpp:
  add null callback to be used for read ahead (no callback needed)
storage/ndb/src/kernel/vm/SimulatedBlock.hpp:
  add null callback to be used for read ahead (no callback needed)
storage/ndb/src/ndbapi/TransporterFacade.cpp:
  remove the debug trace, not needed any more
storage/ndb/test/tools/hugoFill.cpp:
  allow setting database in hugoFill

8 files changed, 92 insertions, 7 deletions

diff --git a/storage/ndb/src/kernel/blocks/dbtup/Dbtup.hpp b/storage/ndb/src/kernel/blocks/dbtup/Dbtup.hpp
index 7845305da6c..05f993dcece 100644
--- a/storage/ndb/src/kernel/blocks/dbtup/Dbtup.hpp
+++ b/storage/ndb/src/kernel/blocks/dbtup/Dbtup.hpp
@@ -2619,6 +2619,9 @@ private:
   ArrayPool<Page> c_page_pool;
   Uint32 cnoOfAllocatedPages;
   Uint32 m_max_allocate_pages;
+
+  /* read ahead in pages during disk order scan */
+  Uint32 m_max_page_read_ahead;
   
   Tablerec *tablerec;
   Uint32 cnoOfTablerec;
diff --git a/storage/ndb/src/kernel/blocks/dbtup/DbtupGen.cpp b/storage/ndb/src/kernel/blocks/dbtup/DbtupGen.cpp
index 3a8e996d435..3ea3deda04f 100644
--- a/storage/ndb/src/kernel/blocks/dbtup/DbtupGen.cpp
+++ b/storage/ndb/src/kernel/blocks/dbtup/DbtupGen.cpp
@@ -347,6 +347,18 @@ void Dbtup::execREAD_CONFIG_REQ(Signal* signal)
   ndbrequire(!ndb_mgm_get_int_parameter(p, CFG_DB_BATCH_SIZE, &nScanBatch));
   c_scanLockPool.setSize(nScanOp * nScanBatch);
 
+
+  /* read ahead for disk scan can not be more that disk page buffer */
+  {
+    Uint64 tmp = 64*1024*1024;
+    ndb_mgm_get_int64_parameter(p, CFG_DB_DISK_PAGE_BUFFER_MEMORY, &tmp);
+    m_max_page_read_ahead = (tmp  + GLOBAL_PAGE_SIZE - 1) / GLOBAL_PAGE_SIZE; // in pages
+    // never read ahead more than 32 pages
+    if (m_max_page_read_ahead > 32)
+      m_max_page_read_ahead = 32;
+  }
+
+
   ScanOpPtr lcp;
   ndbrequire(c_scanOpPool.seize(lcp));
   new (lcp.p) ScanOp();
diff --git a/storage/ndb/src/kernel/blocks/dbtup/DbtupScan.cpp b/storage/ndb/src/kernel/blocks/dbtup/DbtupScan.cpp
index d7ee08f29dd..bb3d8cc0626 100644
--- a/storage/ndb/src/kernel/blocks/dbtup/DbtupScan.cpp
+++ b/storage/ndb/src/kernel/blocks/dbtup/DbtupScan.cpp
@@ -686,13 +686,74 @@ Dbtup::scanNext(Signal* signal, ScanOpPtr scanPtr)
             // move to next extent
             jam();
             pos.m_extent_info_ptr_i = ext_ptr.i;
-            Extent_info* ext = c_extent_pool.getPtr(pos.m_extent_info_ptr_i);
+            ext = c_extent_pool.getPtr(pos.m_extent_info_ptr_i);
             key.m_file_no = ext->m_key.m_file_no;
             key.m_page_no = ext->m_first_page_no;
           }
         }
         key.m_page_idx = 0;
         pos.m_get = ScanPos::Get_page_dd;
+        /*
+          read ahead for scan in disk order
+          do read ahead every 8:th page
+        */
+        if ((bits & ScanOp::SCAN_DD) &&
+            (((key.m_page_no - ext->m_first_page_no) & 7) == 0))
+        {
+          jam();
+          // initialize PGMAN request
+          Page_cache_client::Request preq;
+          preq.m_page = pos.m_key;
+          preq.m_callback = TheNULLCallback;
+
+          // set maximum read ahead
+          Uint32 read_ahead = m_max_page_read_ahead;
+
+          while (true)
+          {
+            // prepare page read ahead in current extent
+            Uint32 page_no = preq.m_page.m_page_no;
+            Uint32 page_no_limit = page_no + read_ahead;
+            Uint32 limit = ext->m_first_page_no + alloc.m_extent_size;
+            if (page_no_limit > limit)
+            {
+              jam();
+              // read ahead crosses extent, set limit for this extent
+              read_ahead = page_no_limit - limit;
+              page_no_limit = limit;
+              // and make sure we only read one extra extent next time around
+              if (read_ahead > alloc.m_extent_size)
+                read_ahead = alloc.m_extent_size;
+            }
+            else
+            {
+              jam();
+              read_ahead = 0; // no more to read ahead after this
+            }
+            // do read ahead pages for this extent
+            while (page_no < page_no_limit)
+            {
+              // page request to PGMAN
+              jam();
+              preq.m_page.m_page_no = page_no;
+              int flags = 0;
+              // ignore result
+              m_pgman.get_page(signal, preq, flags);
+              jamEntry();
+              page_no++;
+            }
+            if (!read_ahead || !list.next(ext_ptr))
+            {
+              // no more extents after this or read ahead done
+              jam();
+              break;
+            }
+            // move to next extent and initialize PGMAN request accordingly
+            Extent_info* ext = c_extent_pool.getPtr(ext_ptr.i);
+            preq.m_page.m_file_no = ext->m_key.m_file_no;
+            preq.m_page.m_page_no = ext->m_first_page_no;
+          }
+        } // if ScanOp::SCAN_DD read ahead
       }
       /*FALLTHRU*/
     case ScanPos::Get_page_dd:
@@ -725,6 +786,7 @@ Dbtup::scanNext(Signal* signal, ScanOpPtr scanPtr)
           safe_cast(&Dbtup::disk_page_tup_scan_callback);
         int flags = 0;
         int res = m_pgman.get_page(signal, preq, flags);
+        jamEntry();
         if (res == 0) {
           jam();
           // request queued
diff --git a/storage/ndb/src/kernel/blocks/pgman.cpp b/storage/ndb/src/kernel/blocks/pgman.cpp
index 78bc70427a9..3709748bae4 100644
--- a/storage/ndb/src/kernel/blocks/pgman.cpp
+++ b/storage/ndb/src/kernel/blocks/pgman.cpp
@@ -122,7 +122,7 @@ Pgman::execREAD_CONFIG_REQ(Signal* signal)
   
   if (page_buffer > 0)
   {
-    page_buffer /= GLOBAL_PAGE_SIZE; // in pages
+    page_buffer = (page_buffer + GLOBAL_PAGE_SIZE - 1) / GLOBAL_PAGE_SIZE; // in pages
     m_param.m_max_pages = page_buffer;
     m_page_entry_pool.setSize(m_param.m_lirs_stack_mult * page_buffer);
     m_param.m_max_hot_pages = (page_buffer * 9) / 10;
@@ -144,7 +144,7 @@ Pgman::Param::Param() :
   m_lirs_stack_mult(10),
   m_max_hot_pages(56),
   m_max_loop_count(256),
-  m_max_io_waits(64),
+  m_max_io_waits(256),
   m_stats_loop_delay(1000),
   m_cleanup_loop_delay(200),
   m_lcp_loop_delay(0)
diff --git a/storage/ndb/src/kernel/vm/SimulatedBlock.cpp b/storage/ndb/src/kernel/vm/SimulatedBlock.cpp
index 7ad1d486a02..bc16b9f364e 100644
--- a/storage/ndb/src/kernel/vm/SimulatedBlock.cpp
+++ b/storage/ndb/src/kernel/vm/SimulatedBlock.cpp
@@ -1658,6 +1658,11 @@ SimulatedBlock::sendFragmentedSignal(NodeReceiverGroup rg,
 }
 
 SimulatedBlock::Callback SimulatedBlock::TheEmptyCallback = {0, 0};
+void
+SimulatedBlock::TheNULLCallbackFunction(class Signal*, Uint32, Uint32)
+{ abort(); /* should never be called */ }
+SimulatedBlock::Callback SimulatedBlock::TheNULLCallback =
+{ &SimulatedBlock::TheNULLCallbackFunction, 0 };
 
 void
 SimulatedBlock::sendFragmentedSignal(BlockReference ref, 
diff --git a/storage/ndb/src/kernel/vm/SimulatedBlock.hpp b/storage/ndb/src/kernel/vm/SimulatedBlock.hpp
index 86e26986f93..a78ee21fb8f 100644
--- a/storage/ndb/src/kernel/vm/SimulatedBlock.hpp
+++ b/storage/ndb/src/kernel/vm/SimulatedBlock.hpp
@@ -131,6 +131,8 @@ public:
   virtual const char* get_filename(Uint32 fd) const { return "";}
 protected:
   static Callback TheEmptyCallback;
+  void TheNULLCallbackFunction(class Signal*, Uint32, Uint32);
+  static Callback TheNULLCallback;
   void execute(Signal* signal, Callback & c, Uint32 returnCode);
   
   
@@ -599,6 +601,8 @@ inline
 void
 SimulatedBlock::execute(Signal* signal, Callback & c, Uint32 returnCode){
   CallbackFunction fun = c.m_callbackFunction; 
+  if (fun == TheNULLCallback.m_callbackFunction)
+    return;
   ndbrequire(fun != 0);
   c.m_callbackFunction = NULL;
   (this->*fun)(signal, c.m_callbackData, returnCode);
diff --git a/storage/ndb/src/ndbapi/TransporterFacade.cpp b/storage/ndb/src/ndbapi/TransporterFacade.cpp
index f982c36cba1..19b384d8dc2 100644
--- a/storage/ndb/src/ndbapi/TransporterFacade.cpp
+++ b/storage/ndb/src/ndbapi/TransporterFacade.cpp
@@ -1403,9 +1403,6 @@ int PollGuard::wait_for_input_in_loop(int wait_time, bool forceSend)
     }
     if (wait_time == -1)
     {
-#ifdef VM_TRACE
-      ndbout << "Waited WAITFOR_RESPONSE_TIMEOUT, continuing wait" << endl;
-#endif
       continue;
     }
     wait_time= max_time - NdbTick_CurrentMillisecond();
diff --git a/storage/ndb/test/tools/hugoFill.cpp b/storage/ndb/test/tools/hugoFill.cpp
index 713c2ca5152..20ceb61b066 100644
--- a/storage/ndb/test/tools/hugoFill.cpp
+++ b/storage/ndb/test/tools/hugoFill.cpp
@@ -30,9 +30,11 @@ int main(int argc, const char** argv){
   const char* _tabname = NULL;
   int _help = 0;
   int _batch = 512;
+  const char* db = "TEST_DB";
   
   struct getargs args[] = {
     { "batch", 'b', arg_integer, &_batch, "Number of operations in each transaction", "batch" },
+    { "database", 'd', arg_string, &db, "Database", "" },
     { "usage", '?', arg_flag, &_help, "Print help", "" }
   };
   int num_args = sizeof(args) / sizeof(args[0]);
@@ -55,7 +57,7 @@ int main(int argc, const char** argv){
   {
     return NDBT_ProgramExit(NDBT_FAILED);
   }
-  Ndb MyNdb(&con, "TEST_DB" );
+  Ndb MyNdb(&con, db);
 
   if(MyNdb.init() != 0){
     ERR(MyNdb.getNdbError());