[erlang-questions] mnesia recover from netsplit, can't delete node from schema

[Rick Pettit]

On Jun 29, 2012, at 10:46 AM, Daniel Dormont wrote:

> Well, it appears that I may have done just that. How do I determine if the database is in an inconsistent state and what can be done about that? Again, I am ok with completely deleting certain tables and nodes from the schema, "brutally" if need be. Is there any way of doing such a thing short of wiping the entire schema and starting from scratch?

Daniel,

I think you might be in a situation similar (though perhaps not exactly like) one which I believe was solved on the Trap Exit forums:

    > >>> On 24 Oct 2010, at 23:37, Jeffrey Rennie wrote: 
    > >>> 
    > >>>> I seem to be stuck in a state where I can't create a table because it 
    > >>>> exists, but I can't delete the table because it doesn't exist! 

You can view the thread @ http://forum.trapexit.org/viewtopic.php?p=62092&sid=7a78bf70100c90aadea4267c921e662d

Take a quick look and see if that sounds like the problem you are having.

If so, I would pay particular attention to comments from Ulf W.

-Rick


> On Fri, Jun 29, 2012 at 10:41 AM, Rick Pettit <rpettit@REDACTED> wrote:
> 
> On Jun 28, 2012, at 2:40 PM, Daniel Dormont wrote:
> 
> > Here is the scenario that happened to me as best I can tell. I had two nodes in a cluster, let's call them A and B. B became unavailable for a while and got rebooted. When I tried to start it again, things seem to work except that certain tables seem not to exist any more. As far as I can tell, these tables used to be enabled only on B and not A, and are now in some sort of weird hybrid unavailable state.
> >
> > A is still running fine in production even with these tables missing, but I can't seem to get a clean start of my application (Ejabberd) on B. So what I figured I would do would be just start a fresh node on B, start Mnesia, add extra_db_nodes pointing to A and go from there. But the problem is A still thinks these certain tables exist only on B (they are listed as remote on A). Fortunately, Ejabberd is smart enough to create any tables it needs on startup, so I was thinking a clean start on B would do this. So I went into A and ran
> >
> > mnesia:del_table_copy(schema, B).
> >
> > thinking this would make the remote tables sort of go away. But instead it fails with
> >
> > {aborted,{no_exists,vcard_search}}
> >
> > And trying to delete the table directly yields the same result.
> >
> > Is there a way I can force Mnesia on A to completely forget about a set of remote tables (and, for that matter, the node that was supposed to store them) before I bring a new node online?
> 
> You might want to take a look at the documentation for mnesia:set_master_nodes/1,2 and maybe mnesia:force_load_table/1.
> 
> Just make sure you understand exactly how these work before using either in production--if used incorrectly, you could leave the database in an inconsistent state.
> 
> Hope that helps,
> 
> -Rick
> 
>