Github user zentol commented on a diff in the pull request:

    https://github.com/apache/flink/pull/4080#discussion_r120343058
  
    --- Diff: docs/monitoring/large_state_tuning.md ---
    @@ -94,21 +94,11 @@ When a savepoint is manually triggered, it may be in 
process concurrently with a
     
     ## Tuning Network Buffers
     
    -The number of network buffers is a parameter that can currently have an 
effect on checkpointing at large scale.
    -The Flink community is working on eliminating that parameter in the next 
versions of Flink.
    -
    -The number of network buffers defines how much data a TaskManager can hold 
in-flight before back-pressure kicks in.
    -A very high number of network buffers means that a lot of data may be in 
the stream network channels when a checkpoint
    -is started. Because the checkpoint barriers travel with that data (see 
[description of how checkpointing 
works](../internals/stream_checkpointing.html)),
    -a lot of in-flight data means that the barriers have to wait for that data 
to be transported/processed before arriving
    -at the target operator.
    -
    -Having a lot of data in-flight also does not speed up the data processing 
as a whole. It only means that data is picked up faster
    -from the data source (log, files, message queue) and buffered longer in 
Flink. Having fewer network buffers means that
    -data is picked up from the source more immediately before it is actually 
being processed, which is generally desirable.
    -The number of network buffers should hence not be set arbitrarily large, 
but to a low multiple (such as 2x) of the
    -minimum number of required buffers.
    -
    +Previously, an increased number of network buffers also caused increased 
checkpointing times since
    --- End diff --
    
    Replace "Previously" with an absolute reference, i.e "Before 1.3".


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---

Reply via email to