Biao Liu created FLINK-13635:
--------------------------------
Summary: Unexpectedly interrupted in AsyncFunction#timeout
Key: FLINK-13635
URL: https://issues.apache.org/jira/browse/FLINK-13635
Project: Flink
Issue Type: Improvement
Components: API / DataStream
Affects Versions: 1.9.0
Reporter: Biao Liu
Fix For: 1.10.0
Currently the way of handling {{AsyncFunction#timeout}} is a bit weird in
{{AsyncWaitOperator#processElement}}.
There are two methods in {{AsyncFunction}}, {{asyncInvoke}} and {{timeout}}.
The {{asyncInvoke}} is executed in task thread, while the {{timeout}} is
executed in system time service. When the {{asyncInvoke}} finished, it might
complete the {{ResultFuture}}. Then it cancels the registered timer of
{{timeout}}. However there is no any synchronization between the
{{asyncFunction}}, {{timeout}} and the cancelation. Moreover this cancelation
is with interruption enabled.
The {{timeout}} must be implemented very carefully. Because when the
{{timeout}} is executing, there might be an interruption triggered at the same
time (due to a completion of {{ResultFuture}}). That means the {{timeout}} must
handle {{InterruptedException}} well everywhere if there is any operation
reacting with this exception.
My proposals are described below.
1. It should be written down in document that the {{asyncInvoke}} and
{{timeout}} might be invoked at the same time.
2. This interruption of {{timeout}} should be avoided. There should be a
synchronization between cancelation and {{timeout}}. If the {{timeout}} is
executing, the cancelation should be avoided. If the cancelation has been
invoked, this {{timeout}} should not be invoked anymore. Or we could simply
cancel the timer without an interruption.
CC [~kkl0u], [~till.rohrmann]
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
