[Spread-users] "Should be a hole" error

[Daniel Rall]

David Shaw wrote:
> When starting a 8-daemon Spread cluster, I am pretty reliably getting:
> 
>  Finished configuration file.
>  Conf_init: My name: 4, id: 10.1.1.4, port: 5678
>  Read_form2: seq 0 should be a hole, but is 3
>  Exit caused by Alarm(EXIT)
> 
> Any ideas?  This happens on startup, so the daemon never really gets
> going.  It isn't always daemon #4 - the one with the problem moves
> around among the 8 machines.

Don't know off hand, but here's the surrounding code from the 
Read_form2(sys_scatter *) function of membership.c:

         Highest_seq = my_rg_info->highest_seq;
         Aru         = my_rg_info->aru;
         Discard_packets();

         for( i=0; i < my_rg_info->num_holes; i++ )
         {
                 /* create dummy messages */
                 pack_entry = *my_holes_procs_ptr & PACKET_MASK;
                 Alarm( MEMB , "EXTRACT HOLE IS %d\n",*my_holes_procs_ptr);
                 if( Packets[pack_entry].exist != 0 )
                     Alarm( EXIT, "Read_form2: seq %d should be a hole, but is 
%d\n",
                         *my_holes_procs_ptr, Packets[pack_entry].exist );
                 Packets[pack_entry].exist = 3;
                 my_holes_procs_ptr++;
         }