Bug #33523991 Ndb online reorg : Scan count verification

frazerclement · frazerclement · commit 07436eb73b9c · 2021-11-10T15:35:28.000Z
Online table reorganisation increases the number of fragments of a table,
and moves rows between them.
This is done in a number of steps :
 1 Copying - rows are duplicated to the new fragments
 2 Commit - The distribution info is changed (hashmap count and total fragments)
 3 Commit-wait - Wait for scan activity using the old distribution to stop
 4 Deleting - Delete rows which have moved-out of existing partitions
 5 Complete - Remove the reference to the old hashmap
 6 Complete-wait - Wait for scan activity started since step 2 to stop

Due to a counting error, the reorg can get stuck in step 6 as it does not
correctly decrement the scan reference count, which therefore never reaches
zero.

This is fixed, and an MTR testcase is extended to cover the scenario.

Approved by : Mauritz Sundell &lt;mauritz.sundell@oracle.com&gt;
diff --git a/mysql-test/suite/ndb/r/ndb_add_partition.result b/mysql-test/suite/ndb/r/ndb_add_partition.result
@@ -2728,6 +2728,46 @@ PRIMARY KEY(c) - UniqueHashIndex
 NDBT_ProgramExit: 0 - OK
 
 drop table t1,t2,t3,t5;
+Bug 33523991 Ndb online reorg : Scan count verification
+Scan count issue on complete phase of reorg
+create table t1 (a int primary key, b int, c int, d int, unique(b), unique(c)) engine=ndb;
+Populate some data
+insert into t1 values (1,1,1,1);
+insert into t1 select a+1, b+1, c+1, d+1 from t1;
+insert into t1 select a+2, b+2, c+2, d+2 from t1;
+insert into t1 select a+4, b+4, c+4, d+4 from t1;
+insert into t1 select a+8, b+8, c+8, d+8 from t1;
+insert into t1 select a+16, b+16, c+16, d+16 from t1;
+insert into t1 select a+32, b+32, c+32, d+32 from t1;
+insert into t1 select a+64, b+64, c+64, d+64 from t1;
+insert into t1 select a+128, b+128, c+128, d+128 from t1;
+insert into t1 select a+256, b+256, c+256, d+256 from t1;
+insert into t1 select a+512, b+512, c+512, d+512 from t1;
+insert into t1 select a+1024, b+1024, c+1024, d+1024 from t1;
+insert into t1 select a+2048, b+2048, c+2048, d+2048 from t1;
+create procedure scans (max_iterations int)
+sp: begin
+DECLARE sum_d int;
+set @x=0;
+repeat
+select sum(d) into sum_d from t1;
+set @x = @x + 1;
+if sum_d = 0
+then
+leave sp;
+end if;
+until @x=max_iterations
+end repeat;
+end%
+call scans(20000);;
+alter table t1 algorithm=inplace, add partition partitions 1;
+Alter ok, run again
+alter table t1 algorithm=inplace, add partition partitions 1;
+Second alter ok, finish scans
+select sum(d) into @sum_d from t1;
+insert into t1 values (20000,20000,20000, 0 - @sum_d);
+drop procedure scans;
+drop table t1;
 alter tablespace ts1 drop datafile 'datafile.dat' engine = ndb;
 drop tablespace ts1 engine = ndb;
 drop logfile group lg1 engine = ndb;
diff --git a/mysql-test/suite/ndb/t/ndb_add_partition.test b/mysql-test/suite/ndb/t/ndb_add_partition.test
@@ -3,7 +3,7 @@
 #but needs to be kept for tests that would need MyISAM in future.
 --source include/force_myisam_default.inc
 
---source include/have_ndb.inc
+--source include/have_multi_ndb.inc
 --source include/not_embedded.inc
 
 --disable_warnings
@@ -458,6 +458,69 @@ let $ndb_desc_opts= -b -i -d test t5;
 source suite/ndb/include/ndb_desc_print.inc;
 
 drop table t1,t2,t3,t5;
+
+--echo Bug 33523991 Ndb online reorg : Scan count verification
+--echo Scan count issue on complete phase of reorg
+
+create table t1 (a int primary key, b int, c int, d int, unique(b), unique(c)) engine=ndb;
+
+--echo Populate some data
+insert into t1 values (1,1,1,1);
+insert into t1 select a+1, b+1, c+1, d+1 from t1;
+insert into t1 select a+2, b+2, c+2, d+2 from t1;
+insert into t1 select a+4, b+4, c+4, d+4 from t1;
+insert into t1 select a+8, b+8, c+8, d+8 from t1;
+insert into t1 select a+16, b+16, c+16, d+16 from t1;
+insert into t1 select a+32, b+32, c+32, d+32 from t1;
+insert into t1 select a+64, b+64, c+64, d+64 from t1;
+insert into t1 select a+128, b+128, c+128, d+128 from t1;
+insert into t1 select a+256, b+256, c+256, d+256 from t1;
+insert into t1 select a+512, b+512, c+512, d+512 from t1;
+insert into t1 select a+1024, b+1024, c+1024, d+1024 from t1;
+insert into t1 select a+2048, b+2048, c+2048, d+2048 from t1;
+
+--connection server2
+
+delimiter %;
+create procedure scans (max_iterations int)
+sp: begin
+  DECLARE sum_d int;
+  set @x=0;
+  repeat
+    select sum(d) into sum_d from t1;
+    set @x = @x + 1;
+    if sum_d = 0
+    then
+      leave sp;
+    end if;
+  until @x=max_iterations
+  end repeat;
+end%
+
+delimiter ;%
+
+--send call scans(20000);
+
+--connection server1
+
+alter table t1 algorithm=inplace, add partition partitions 1;
+
+--echo Alter ok, run again
+
+alter table t1 algorithm=inplace, add partition partitions 1;
+
+--echo Second alter ok, finish scans
+select sum(d) into @sum_d from t1;
+insert into t1 values (20000,20000,20000, 0 - @sum_d);
+
+--connection server2
+--reap
+
+drop procedure scans;
+
+--connection server1
+drop table t1;
+
 alter tablespace ts1 drop datafile 'datafile.dat' engine = ndb;
 drop tablespace ts1 engine = ndb;
 drop logfile group lg1 engine = ndb;
diff --git a/storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp b/storage/ndb/src/kernel/blocks/dbdih/DbdihMain.cpp
@@ -15560,7 +15560,6 @@ Dbdih::start_scan_on_table(TabRecordPtr tabPtr,
   }
 
   tabPtr.p->m_scan_count[0]++;
-  ndbrequire(tabPtr.p->m_map_ptr_i != DihScanTabConf::InvalidCookie);
   {
     DihScanTabConf* conf = (DihScanTabConf*)signal->getDataPtrSend();
     conf->tableId = tabPtr.i;
@@ -15579,7 +15578,7 @@ Dbdih::start_scan_on_table(TabRecordPtr tabPtr,
     conf->fragmentCount = tabPtr.p->partitionCount;
 
     conf->noOfBackups = tabPtr.p->noOfBackups;
-    conf->scanCookie = tabPtr.p->m_map_ptr_i;
+    conf->scanCookie = tabPtr.p->m_scan_reorg_flag;
     conf->reorgFlag = tabPtr.p->m_scan_reorg_flag;
     NdbMutex_Unlock(&tabPtr.p->theMutex);
     return;
@@ -15598,7 +15597,7 @@ Dbdih::start_scan_on_table(TabRecordPtr tabPtr,
 
 void
 Dbdih::complete_scan_on_table(TabRecordPtr tabPtr,
-                              Uint32 map_ptr_i,
+                              Uint32 scanCookie,
                               EmulatedJamBuffer *jambuf)
 {
   /**
@@ -15610,16 +15609,18 @@ Dbdih::complete_scan_on_table(TabRecordPtr tabPtr,
    */
 
   NdbMutex_Lock(&tabPtr.p->theMutex);
-  if (map_ptr_i == tabPtr.p->m_map_ptr_i)
+  if (scanCookie == tabPtr.p->m_scan_reorg_flag)
   {
+    /* Scan started + completed in same reorg state */
     thrjam(jambuf);
-    ndbassert(tabPtr.p->m_scan_count[0]);
+    ndbrequire(tabPtr.p->m_scan_count[0]);
     tabPtr.p->m_scan_count[0]--;
   }
   else
   {
+    /* Scan started + completed in diff reorg states */
     thrjam(jambuf);
-    ndbassert(tabPtr.p->m_scan_count[1]);
+    ndbrequire(tabPtr.p->m_scan_count[1]);
     tabPtr.p->m_scan_count[1]--;
   }
   NdbMutex_Unlock(&tabPtr.p->theMutex);
@@ -15872,7 +15873,7 @@ Dbdih::make_new_table_read_and_writeable(TabRecordPtr tabPtr,
     DIH_TAB_WRITE_UNLOCK(tabPtr.p);
 
     /* These variables are only protected by mutex. */
-    ndbassert(tabPtr.p->m_scan_count[1] == 0);
+    ndbrequire(tabPtr.p->m_scan_count[1] == 0);
     tabPtr.p->m_scan_count[1] = tabPtr.p->m_scan_count[0];
     tabPtr.p->m_scan_count[0] = 0;
     tabPtr.p->m_scan_reorg_flag = 1;
@@ -15919,7 +15920,7 @@ Dbdih::make_old_table_non_writeable(TabRecordPtr tabPtr,
      * REORG_NOT_MOVED flag set we also start waiting for those
      * scans to complete here.
      */
-    ndbassert(tabPtr.p->m_scan_count[1] == 0);
+    ndbrequire(tabPtr.p->m_scan_count[1] == 0);
     tabPtr.p->m_scan_count[1] = tabPtr.p->m_scan_count[0];
     tabPtr.p->m_scan_count[0] = 0;
     wait_flag = true;
