[PATCH] replication: fix exit with ER_NO_SUCH_USER during bootstrap

[Vladimir Davydov]

On Thu, Aug 23, 2018 at 03:57:43PM +0300, Serge Petrenko wrote:
> When replication is configured via some user created in box.once()
> function and box.once() takes more than replication_timeout seconds
> to execute, appliers recieve ER_NO_SUCH_USER error, which they don't
> handle. This leads to occasional test failures in replication suite.
> Fix this by handling the aforementioned case in applier_f().
> 
> Closes #3637
> ---
> https://github.com/tarantool/tarantool/issues/3637
> https://github.com/tarantool/tarantool/tree/sp/gh-3637-replication-tests-fix
> 
>  src/box/applier.cc | 8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/src/box/applier.cc b/src/box/applier.cc
> index dbb4d05f9..740778e80 100644
> --- a/src/box/applier.cc
> +++ b/src/box/applier.cc
> @@ -607,6 +607,14 @@ applier_f(va_list ap)
>  				applier_log_error(applier, e);
>  				applier_disconnect(applier, APPLIER_DISCONNECTED);
>  				goto reconnect;
> +			} else if (e->errcode() == ER_NO_SUCH_USER) {
> +				/*
> +				 * Probably box.once() hasn't finished
> +				 * on bootstrap leader yet.
> +				 */
> +				applier_log_error(applier, e);
> +				applier_disconnect(applier, APPLIER_DISCONNECTED);
> +				goto reconnect;

This piece of code isn't covered by any test, see

https://coveralls.io/builds/18630789/source?filename=src/box/applier.cc#L610

Please add a test case. I think it should be pretty easy to do: start a
replica with a small value of box.cfg.replication_timeout which tries to
connect to the default instance as a non-existent user, then wait a bit,
create the user, and make sure it finally connects.

Also, I think that you should share the code with ER_ACCESS_DENIED
case, because these two errors have the same nature - missing or invalid
configuration:

	if (e->errcode() == ER_ACCESS_DENIED ||
	    e->errcode() == ER_NO_SUCH_USER) {
		/* Invalid configuration */
		applier_log_error(applier, e);
		applier_disconnect(applier, APPLIER_DISCONNECTED);
		goto reconnect;
	}