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

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

[Serge Petrenko]

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.

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);
 		} 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);
 		} 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));
+	}
 	lua_settable(L, -3);
 }
 
-- 
2.17.1 (Apple Git-112)

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

[Vladimir Davydov]

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);
>  }

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

[Serge Petrenko]

[-- Attachment #1: Type: text/plain, Size: 3274 bytes --]

HI! Thank you for review.
The new diff is below.

> 18 окт. 2018 г., в 15:43, Vladimir Davydov <vdavydov.dev@gmail.com> написал(а):
> 
> On Tue, Oct 16, 2018 at 01:10:53PM +0300, Serge Petrenko wrote:
>> 
>> +			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.

The same is done in logging during recovery. Here it at least lets you make a rough
time estimation.

> 
>> 		} 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.

I suggest not adding logging to the subscribe stage, it is not needed IMO.

> 
>> 		} 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.

Yes, you’re correct. Removed this.

---
 src/box/applier.cc | 7 +++++++
 1 file changed, 7 insertions(+)

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);
 		} 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);
 		} else if (row.type == IPROTO_OK) {
 			/*
 			 * Current vclock. This is not used now,
-- 
2.17.1 (Apple Git-112)



[-- Attachment #2: Type: text/html, Size: 11459 bytes --]

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

[Vladimir Davydov]

Pushed to 1.10-features