Robert Stoll created FLINK-12525:
------------------------------------
Summary: Impose invariant
StreamExecutionEnvironment.setBufferTimeout > 0
Key: FLINK-12525
URL: https://issues.apache.org/jira/browse/FLINK-12525
Project: Flink
Issue Type: Improvement
Reporter: Robert Stoll
The documentation for the [DataStream
API|https://github.com/apache/flink/blob/8674b69964eae50cad024f2c5caf92a71bf21a09/docs/dev/datastream_api.md#controlling-latency]
states:
{quote}buffer timeout of 0 should be avoided, because it can cause severe
performance degradation.
{quote}
I don't know if the documentation is not appropriate and there are valid cases
where a timeout of 0 makes sense. But if not, then the invariant should not be
{color:#c586c0}if{color}{color:#d4d4d4} (timeoutMillis
{color}{color:#d4d4d4}<{color}
{color:#d4d4d4}-{color}{color:#b5cea8}1{color}{color:#d4d4d4}) {{color}
{color:#d4d4d4} {color}{color:#c586c0}throw{color} {color:#c586c0}new{color}
{color:#dcdcaa}IllegalArgumentException{color}{color:#d4d4d4}({color}{color:#ce9178}"Timeout
of buffer must be non-negative or -1"{color}{color:#d4d4d4});{color}
{color:#d4d4d4}}{color}
But {{timeoutMillis < 0}} (can also be a second invariant)
IMO it is bad practice to state it only in the documentation. The API should
guide the user in this case (in this sense a second invariant stating the quote
above would make more sense).
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
