Flink sent/received bytes related metrics: how are they calculated?

classic Classic list List threaded Threaded
5 messages Options
Reply | Threaded
Open this post in threaded view
|

Flink sent/received bytes related metrics: how are they calculated?

Andrea Spina
Dear Community, I'd like to ask for some details about bytes related metrics in Flink. Precisely, I'm looking at bytes sent and bytes received metrics: what I am recording is the following:

I am reading from a Kafka topic A records with schema K using a source Sin belonging to a pipeline P1. The topic A is filled with K data using a separated pipeline p2 and a sink Sout.

Basically what I register is that the bytes sent metric of Sin - available through the Flink UI -measure twice or more respect of the bytes received of Sout. For instance, if my sink Sout records 10GB of bytes received inbound, then my source Sin emits between 20-25GB bytes sent.

Screenshot 2019-06-14 at 09.40.08.png

Is someone able to detail how these two metrics are calculated? 

Thank you,

--
Andrea Spina
Software Engineer @ Radicalbit Srl
Via Borsieri 41, 20159, Milano - IT
Reply | Threaded
Open this post in threaded view
|

Re: Flink sent/received bytes related metrics: how are they calculated?

Chesnay Schepler
Which version of Flink are you using? There were some issues at some point about double-counting.

On 14/06/2019 09:49, Andrea Spina wrote:
Dear Community, I'd like to ask for some details about bytes related metrics in Flink. Precisely, I'm looking at bytes sent and bytes received metrics: what I am recording is the following:

I am reading from a Kafka topic A records with schema K using a source Sin belonging to a pipeline P1. The topic A is filled with K data using a separated pipeline p2 and a sink Sout.

Basically what I register is that the bytes sent metric of Sin - available through the Flink UI -measure twice or more respect of the bytes received of Sout. For instance, if my sink Sout records 10GB of bytes received inbound, then my source Sin emits between 20-25GB bytes sent.

Screenshot 2019-06-14 at 09.40.08.png

Is someone able to detail how these two metrics are calculated? 

Thank you,

--
Andrea Spina
Software Engineer @ Radicalbit Srl
Via Borsieri 41, 20159, Milano - IT


Reply | Threaded
Open this post in threaded view
|

Re: Flink sent/received bytes related metrics: how are they calculated?

Andrea Spina
Sorry, I totally missed the version: flink-1.6.4, Streaming API

Il giorno ven 14 giu 2019 alle ore 11:08 Chesnay Schepler <[hidden email]> ha scritto:
Which version of Flink are you using? There were some issues at some point about double-counting.

On 14/06/2019 09:49, Andrea Spina wrote:
Dear Community, I'd like to ask for some details about bytes related metrics in Flink. Precisely, I'm looking at bytes sent and bytes received metrics: what I am recording is the following:

I am reading from a Kafka topic A records with schema K using a source Sin belonging to a pipeline P1. The topic A is filled with K data using a separated pipeline p2 and a sink Sout.

Basically what I register is that the bytes sent metric of Sin - available through the Flink UI -measure twice or more respect of the bytes received of Sout. For instance, if my sink Sout records 10GB of bytes received inbound, then my source Sin emits between 20-25GB bytes sent.

Screenshot 2019-06-14 at 09.40.08.png

Is someone able to detail how these two metrics are calculated? 

Thank you,

--
Andrea Spina
Software Engineer @ Radicalbit Srl
Via Borsieri 41, 20159, Milano - IT




--
Andrea Spina
Head of R&D @ Radicalbit Srl
Via Giovanni Battista Pirelli 11, 20124, Milano - IT
Reply | Threaded
Open this post in threaded view
|

Re: Flink sent/received bytes related metrics: how are they calculated?

Chesnay Schepler
How does the P1 pipeline look like? Are there 2 downstream operators reading from Sin (in this case the number of bytes would be measured twice)?

On 14/06/2019 12:09, Andrea Spina wrote:
Sorry, I totally missed the version: flink-1.6.4, Streaming API

Il giorno ven 14 giu 2019 alle ore 11:08 Chesnay Schepler <[hidden email]> ha scritto:
Which version of Flink are you using? There were some issues at some point about double-counting.

On 14/06/2019 09:49, Andrea Spina wrote:
Dear Community, I'd like to ask for some details about bytes related metrics in Flink. Precisely, I'm looking at bytes sent and bytes received metrics: what I am recording is the following:

I am reading from a Kafka topic A records with schema K using a source Sin belonging to a pipeline P1. The topic A is filled with K data using a separated pipeline p2 and a sink Sout.

Basically what I register is that the bytes sent metric of Sin - available through the Flink UI -measure twice or more respect of the bytes received of Sout. For instance, if my sink Sout records 10GB of bytes received inbound, then my source Sin emits between 20-25GB bytes sent.

Screenshot 2019-06-14 at 09.40.08.png

Is someone able to detail how these two metrics are calculated? 

Thank you,

--
Andrea Spina
Software Engineer @ Radicalbit Srl
Via Borsieri 41, 20159, Milano - IT




--
Andrea Spina
Head of R&D @ Radicalbit Srl
Via Giovanni Battista Pirelli 11, 20124, Milano - IT


Reply | Threaded
Open this post in threaded view
|

Re: Flink sent/received bytes related metrics: how are they calculated?

Andrea Spina
Hi Chesnay, just one downstream: Sin (Source: Enriched Code) outcome is the right part of the following operator as in the figure; this operator is the exclusive downstream of Sin.
Thanks,

Screenshot 2019-06-14 at 13.52.52.png

Il giorno ven 14 giu 2019 alle ore 12:23 Chesnay Schepler <[hidden email]> ha scritto:
How does the P1 pipeline look like? Are there 2 downstream operators reading from Sin (in this case the number of bytes would be measured twice)?

On 14/06/2019 12:09, Andrea Spina wrote:
Sorry, I totally missed the version: flink-1.6.4, Streaming API

Il giorno ven 14 giu 2019 alle ore 11:08 Chesnay Schepler <[hidden email]> ha scritto:
Which version of Flink are you using? There were some issues at some point about double-counting.

On 14/06/2019 09:49, Andrea Spina wrote:
Dear Community, I'd like to ask for some details about bytes related metrics in Flink. Precisely, I'm looking at bytes sent and bytes received metrics: what I am recording is the following:

I am reading from a Kafka topic A records with schema K using a source Sin belonging to a pipeline P1. The topic A is filled with K data using a separated pipeline p2 and a sink Sout.

Basically what I register is that the bytes sent metric of Sin - available through the Flink UI -measure twice or more respect of the bytes received of Sout. For instance, if my sink Sout records 10GB of bytes received inbound, then my source Sin emits between 20-25GB bytes sent.

Screenshot 2019-06-14 at 09.40.08.png

Is someone able to detail how these two metrics are calculated? 

Thank you,

--
Andrea Spina
Software Engineer @ Radicalbit Srl
Via Borsieri 41, 20159, Milano - IT




--
Andrea Spina
Head of R&D @ Radicalbit Srl
Via Giovanni Battista Pirelli 11, 20124, Milano - IT




--
Andrea Spina
Head of R&D @ Radicalbit Srl
Via Giovanni Battista Pirelli 11, 20124, Milano - IT