test/fio: fix global CephContext life cycle

[ifed01]

It looks like that's a bad idea to omit add_ref when returning intrusive_ptr from global_init. Assigning temporary resulting pointer to another var doesn't increment ref counter. From the boost manual - this is equivalent to swap call instead. Hence we might have either lack of CephContext release or its' premature release. The latter case I observed in fio plugin.

Consider the following code snippet:

#include <iostream>
#include <boost/intrusive_ptr.hpp>
using namespace std;
struct  A
{
  int nref = 0;
  A() { cout<<"A"<<this<<std::endl;}
  ~A() { cout<<"~A"<<this<<std::endl;}

  void put() { if (--nref == 0 ) delete this;}
  void get() { nref++;}
};

void intrusive_ptr_add_ref(A* a)
{
  a->get();
  cout<<"add "<<a<<" "<<a->nref<<std::endl;
}
void intrusive_ptr_release(A* a)
{
  cout<<"release "<<a<<" "<<a->nref-1<<std::endl;
  a->put();
}

typedef boost::intrusive_ptr<A> APtr;
 APtr createAPtr() {
  A* res = new A;
  return APtr(res, false);
}

int main()
{
  auto aptr = createAPtr();
  APtr aptr2;
  aptr2 = createAPtr();
  APtr aptr3;
  aptr3 = createAPtr();
  aptr3->get(); //OK
  APtr aptr4(createAPtr());
  APtr aptr5;
  aptr5 = createAPtr();
  aptr5->get(); //OK
  APtr aptr6;
  aptr6 = aptr5; //OK
  createAPtr();
  cout<<"-----------------------"<<std::endl;
}

Corresponding output:

A0x9d8c20
A0x9d9050
A0x9d9070
A0x9d9090
A0x9d90b0
add 0x9d90b0 2
A0x19cd0d0
release 0x19cd0d0 -1
-----------------------
release 0x9d90b0 1
release 0x9d90b0 0
~A0x9d90b0
release 0x9d9090 -1
release 0x9d9070 0
~A0x9d9070
release 0x9d9050 -1
release 0x9d8c20 -1

[cbodley]

the difference between CephContext and struct A in your example is that the CephContext ctor initializes nref(1)

your change to the fio engine is correct though

[tchaikov]

agreed with @cbodley , f6ea4d5 looks good to me. but we might want to drop c139e9b from this PR.

[ifed01]

@cbodley @tchaikov yeah, missed that "minor" thing... Fixed.

[ifed01]

oops, dropped another commit..

[ifed01]

fixed.

[ifed01]

@tchaikov would you approve?

review comments addressed

[tchaikov]

lgtm =)