Re: Fatal error after starting postgres : sys identifiers must be different

Hi!

On 2013-07-19 07:31:07 +0900, Michael Paquier wrote:
> If this behavior is confirmed, I think that this bug should be
> reported directly to Andres and not this mailing list, just because
> logical replication is not integrated into Postgres core as of now.

I think I agree, although I don't know where to report it, but to me
personally, so far. Hackers seems to be the wrong crowd for it, given
most of the people on it haven't even heard of bdr, much less read its
code.
We're definitely planning to propose it for community inclusion in some
form, but there are several rather large preliminary steps (like getting
changeset extraction in) that need to be done first.

Not sure what's best.

On 2013-07-19 07:31:07 +0900, Michael Paquier wrote:
> On Fri, Jul 19, 2013 at 5:02 AM, Indrajit Roychoudhury
> <indrajit.roychoudhury@gmail.com> wrote:
> > Could you please let me know what does the error "system identifiers are
> > same" mean? Below is the snapshot from one of the masters.
> > I am setting up BDR as per the wiki
> >   http://wiki.postgresql.org/wiki/BDR_User_Guide#Initial_setup
> > and source @
> >  git://git.postgresql.org/git/users/andresfreund/postgres.git
> >
> > irc1@ubuntuirc:~/bdr2/postgres-d6ed89e/postgres-bdr-bin/bin$ ./postgres -D
> > ~/bdr2/postgres-d6ed89e/postgres-bdr-bin/bin/data2/
> > LOG:  bgworkers, connection: dbname=testdb2
> > LOG:  option: dbname, val: testdb2
> > LOG:  registering background worker: bdr apply: ubuntuirc2
> > LOG:  loaded library "bdr"
> > LOG:  database system was shut down at 2013-03-17 10:56:52 PDT
> > LOG:  doing logical startup from 0/17B6410
> > LOG:  starting up replication identifier with ckpt at 0/17B6410
> > LOG:  autovacuum launcher started
> > LOG:  starting background worker process "bdr apply: ubuntuirc2"
> > LOG:  database system is ready to accept connections
> > LOG:  bdr apply: ubuntuirc2 initialized on testdb2, remote dbname=testdb2
> > replication=true fallback_application_name=bdr
> > FATAL:  system identifiers must differ between the nodes
> > DETAIL:  Both system identifiers are 5856368744762683487.

> I am not the best specialist about logical replication, but as it
> looks to be a requirement to have 2 nodes with different system
> identifiers, you shouldn't link 1 node to another node whose data
> folder has been created from the base backup of the former (for
> example create the 2nd node based on a base backup of the 1st node).
> The error returned here would mean so.

Yes, that's correct.

> > LOG:  worker process: bdr apply: ubuntuirc2 (PID 16712) exited with exit
> > code 1
> > ^CLOG:  received fast shutdown request
> > LOG:  aborting any active transactions
> > LOG:  autovacuum launcher shutting down
> > LOG:  shutting down
> >
> > pgcontrol_data outputs different database system ids for the two nodes. So
> > don't understand why it says identifiers are same.
> Are you sure? Please re-ckeck.
>
> > postgresql.conf content in one of the masters is like this-
> >
> > /////////////////
> > shared_preload_libraries = 'bdr'
> > bdr.connections = 'ubuntuirc2'
> > bdr.ubuntuirc2.dsn = 'dbname=testdb2'
> > /////////////////
> >
> > Two nodes are ubuntuirc and ubuntuirc2. Above is the output of the
> > postgresql.conf from ubuntuirc.

The problem seems to be that your dsn doesn't include a hostname but
only a database name. So, what probably happens is both your hosts try
to connect to themselves since connecting to the local host is the
default when no hostname is specified. Which is one of the primary
reasons the requirement for differing system identifiers exist...

Greetings,

Andres Freund