maria-developers team mailing list archive

[Sergei Golubchik <serg@xxxxxxxxxxx>]

Hi, Andrei,

I'm sorry, but I don't understand that fix at all.
Why did you move do_accept_own_server_id into Master_info if you only
use it in one static function queue_event() ?
Why did you change it under some complex condition that you moved
somewhere up in the queue_event() when you could've done simply

  && opt_gtid_strict_mode

?

On Feb 17, Andrei wrote:
> revision-id: 592f57e25a5 (mariadb-10.6.5-82-g592f57e25a5)
> parent(s): 8d742fe4acb
> author: Andrei
> committer: Andrei
> timestamp: 2022-02-14 16:31:25 +0200
> message:
> 
> MDEV-27760 event may non stop replicate in circular semisync setup
> 
> diff --git a/sql/slave.cc b/sql/slave.cc
> index 5ff40d8fb25..d5275d17460 100644
> --- a/sql/slave.cc
> +++ b/sql/slave.cc
> @@ -5037,6 +5037,7 @@ log space");
>    mi->abort_slave= 0;
>    mi->slave_running= MYSQL_SLAVE_NOT_RUN;
>    mi->io_thd= 0;
> +  mi->do_accept_own_server_id= false;
>    /*
>      Note: the order of the two following calls (first broadcast, then unlock)
>      is important. Otherwise a killer_thread can execute between the calls and
> @@ -6175,15 +6176,6 @@ static int queue_event(Master_info* mi, const uchar *buf, ulong event_len)
>    uchar new_buf_arr[4096];
>    bool is_malloc = false;
>    bool is_rows_event= false;
> -  /*
> -    The flag has replicate_same_server_id semantics and is raised to accept
> -    a same-server-id event on the semisync slave, for both the gtid and legacy
> -    connection modes.
> -    Such events can appear as result of this server recovery so the event
> -    was created there and replicated elsewhere right before the crash. At recovery
> -    it could be evicted from the server's binlog.
> -  */
> -  bool do_accept_own_server_id= false;
>    /*
>      FD_q must have been prepared for the first R_a event
>      inside get_master_version_and_clock()
> @@ -6272,6 +6264,8 @@ static int queue_event(Master_info* mi, const uchar *buf, ulong event_len)
>                      dbug_rows_event_count = 0;
>                    };);
>  #endif
> +  s_id= uint4korr(buf + SERVER_ID_OFFSET);
> +
>    mysql_mutex_lock(&mi->data_lock);
>  
>    switch (buf[EVENT_TYPE_OFFSET]) {
> @@ -6713,6 +6707,20 @@ static int queue_event(Master_info* mi, const uchar *buf, ulong event_len)
>  
>      ++mi->events_queued_since_last_gtid;
>      inc_pos= event_len;
> +
> +    if (s_id == global_system_variables.server_id &&
> +	rpl_semi_sync_slave_enabled &&
> +	mi->using_gtid != Master_info::USE_GTID_NO &&
> +	opt_gtid_strict_mode)
> +    {
> +      rpl_gtid *state_gtid= mi->gtid_current_pos.find(event_gtid.domain_id);
> +      mi->do_accept_own_server_id=
> +	!state_gtid || state_gtid->seq_no < event_gtid.seq_no;
> +    }
> +    else
> +    {
> +      mi->do_accept_own_server_id= false;
> +    }
>    }
>    break;
>    /*
> @@ -6900,7 +6908,6 @@ static int queue_event(Master_info* mi, const uchar *buf, ulong event_len)
>    */
>  
>    mysql_mutex_lock(log_lock);
> -  s_id= uint4korr(buf + SERVER_ID_OFFSET);
>    /*
>      Write the event to the relay log, unless we reconnected in the middle
>      of an event group and now need to skip the initial part of the group that
> @@ -6946,7 +6953,7 @@ static int queue_event(Master_info* mi, const uchar *buf, ulong event_len)
>    else
>    if ((s_id == global_system_variables.server_id &&
>         !(mi->rli.replicate_same_server_id ||
> -         (do_accept_own_server_id= rpl_semi_sync_slave_enabled))) ||
> +         mi->do_accept_own_server_id)) ||
>        event_that_should_be_ignored(buf) ||
>        /*
>          the following conjunction deals with IGNORE_SERVER_IDS, if set
> @@ -7006,7 +7013,7 @@ static int queue_event(Master_info* mi, const uchar *buf, ulong event_len)
>    }
>    else
>    {
> -    if (do_accept_own_server_id)
> +    if (mi->do_accept_own_server_id)
>      {
>        int2store(const_cast<uchar*>(buf + FLAGS_OFFSET),
>                  uint2korr(buf + FLAGS_OFFSET) | LOG_EVENT_ACCEPT_OWN_F);
> 
Regards,
Sergei
VP of MariaDB Server Engineering
and security@xxxxxxxxxxx