[tarantool-patches] [PATCH] [replication] make join stage more informative

Vladimir Davydov vdavydov.dev at gmail.com
Thu Oct 18 15:43:13 MSK 2018


On Tue, Oct 16, 2018 at 01:10:53PM +0300, Serge Petrenko wrote:
> This patch adds logging amount of rows recieved by applier during the
> join stage, the same way that recovery has it. It also adds
> 
>     downstream:
>       status: joining
> 
> message to master's box.info.replication for a replica that is joining.

Please add a test case for the new box.info.replication status.

> 
> Closes #3165
> ---
> https://github.com/tarantool/tarantool/issues/3165
> https://github.com/tarantool/tarantool/tree/sp/gh-3165-join-progress-report
> 
>  src/box/applier.cc | 7 +++++++
>  src/box/lua/info.c | 9 +++++++--
>  2 files changed, 14 insertions(+), 2 deletions(-)
> 
> diff --git a/src/box/applier.cc b/src/box/applier.cc
> index 7da278e68..ba68c3d07 100644
> --- a/src/box/applier.cc
> +++ b/src/box/applier.cc
> @@ -309,11 +309,15 @@ applier_join(struct applier *applier)
>  	 * Receive initial data.
>  	 */
>  	assert(applier->join_stream != NULL);
> +	uint64_t row_count = 0;
>  	while (true) {
>  		coio_read_xrow(coio, ibuf, &row);
>  		applier->last_row_time = ev_monotonic_now(loop());
>  		if (iproto_type_is_dml(row.type)) {
>  			xstream_write_xc(applier->join_stream, &row);
> +			++row_count;
> +			if (row_count % 100000 == 0)
> +				say_info("%.1fM rows recieved", row_count / 1e6);

Without knowing the total number of rows to be received, this isn't very
informative IMO.

>  		} else if (row.type == IPROTO_OK) {
>  			if (applier->version_id < version_id(1, 7, 0)) {
>  				/*
> @@ -354,6 +358,9 @@ applier_join(struct applier *applier)
>  		if (iproto_type_is_dml(row.type)) {
>  			vclock_follow_xrow(&replicaset.vclock, &row);
>  			xstream_write_xc(applier->subscribe_stream, &row);
> +			++row_count;
> +			if (row_count % 100000 == 0)
> +				say_info("%.1fM rows recieved", row_count / 1e6);

When bootstrapping from Tarantool < 1.7, final join won't end here, it
will end in applier_subscribe(). Not sure if we want to handle this case
though.

>  		} else if (row.type == IPROTO_OK) {
>  			/*
>  			 * Current vclock. This is not used now,
> diff --git a/src/box/lua/info.c b/src/box/lua/info.c
> index 655768ec4..4c25255c0 100644
> --- a/src/box/lua/info.c
> +++ b/src/box/lua/info.c
> @@ -119,8 +119,13 @@ static void
>  lbox_pushrelay(lua_State *L, struct relay *relay)
>  {
>  	lua_newtable(L);
> -	lua_pushstring(L, "vclock");
> -	lbox_pushvclock(L, relay_vclock(relay));
> +	if (relay_vclock(relay)->map == 0) {
> +		lua_pushstring(L, "status");
> +		lua_pushstring(L, "joining");
> +	} else {
> +		lua_pushstring(L, "vclock");
> +		lbox_pushvclock(L, relay_vclock(relay));
> +	}

This wouldn't work, because struct replica doesn't exist before initial
join is complete.

>  	lua_settable(L, -3);
>  }



More information about the Tarantool-patches mailing list