add_index() does not use locking

[cc32d9]

I'm using chainbase in Chronicle project in concurrent mode: one process opens the database in read-write mode and is constantly updating it as state history data arrives. Then, multiple read-only processes read from the same shared memory.

I define CHAINBASE_CHECK_LOCKING globally.

I get the following error when read-only client opens the database and defines the indexes:

May 25 01:05:46 sthist chronicle-receiver[5523]: warn  2019-05-25T01:05:46.141 chronicle receiver_plugin.cpp:1201      plugin_initialize    ] 13 N5boost16exception_detail10clone_implINS0_19error_info_injectorISt11logic_errorEEEE: existing index for chronicle::received_block_object has an undo stack (revision range [-1, 23598015]) that is inconsistent with other indices in the database (revision range [23597735, 23597736]); corrupted database?
May 25 01:05:46 sthist chronicle-receiver[5523]: rethrow existing index for chronicle::received_block_object has an undo stack (revision range [-1, 23598015]) that is inconsistent with other indices in the database (revision range [23597735, 23597736]); corrupted database?:
May 25 01:05:46 sthist chronicle-receiver[5523]:     {"what":"existing index for chronicle::received_block_object has an undo stack (revision range [-1, 23598015]) that is inconsistent with other indices in the database (revision range [23597735, 23597736]); corrupted database?"}
May 25 01:05:46 sthist chronicle-receiver[5523]:     chronicle-recei  receiver_plugin.cpp:1201 plugin_initialize

so, there should be a global lock wrapping around add_index(). For now I will wrap the locks around calling add_index in my program, but I think it makes sense to protect it inside chainbase.

[cc32d9]

actually I don't see how this is locking anything. It's not modifying the lock count and not using the read_lock object. Does it work at all?

         void require_lock_fail( const char* method, const char* lock_type, const char* tname )const;

         void require_read_lock( const char* method, const char* tname )const
         {
            if( BOOST_UNLIKELY( _enable_require_locking & _read_only & (_read_lock_count <= 0) ) )
               require_lock_fail(method, "read", tname);
         }

         void require_write_lock( const char* method, const char* tname )
         {
            if( BOOST_UNLIKELY( _enable_require_locking & (_write_lock_count <= 0) ) )
               require_lock_fail(method, "write", tname);
         }

[cc32d9]

as Todd explained in Devs chat, chainbase is not made to be used in concurrent environment. So probably it makes sense to cleanup unused parts in the code to avoid confusion

[datnguyencse]

Look like multi process trying to set_revision on same database.