commit | 2f957f7afec3a6a67c3f786e6a88c505ac395fb3 | [log] [tgz] |
---|---|---|
author | Igal Shilman <igalshilman@gmail.com> | Sat Mar 06 21:34:30 2021 +0100 |
committer | Tzu-Li (Gordon) Tai <tzulitai@apache.org> | Mon Mar 08 16:50:39 2021 +0800 |
tree | 6778b1a718e119dbaf9f9238a940fb618bc94638 | |
parent | 1bea5976fbf6316fb65ad8e2fd3e4705179e4ee6 [diff] |
[FLINK-21642] Make a retry batch on recovery According to the request-reply protocol, on recovery we re-send the uncompleted (in-flight during a failure) messages. But we shouldn't assume that the state definitions are the same as in the previous attempt. We create a retry batch and let the SDK to reply with the correct state specs This closes #208.