flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From tzulitai <...@git.apache.org>
Subject [GitHub] flink issue #5337: [FLINK-8484][flink-kinesis-connector] Ensure a Kinesis co...
Date Wed, 24 Jan 2018 09:36:58 GMT
Github user tzulitai commented on the issue:

    re @StephanEwen 
    yes, currently, state is still kept indefinitely for closed shards. A special `SHARD_END`
marker sequence number is stored as the sequence number for closed shards, so that the consumer
does not attempt to read them on restore.
    A threshold timestamp could work if AWS API provides shard creation times. A threshold
sequence numbers would also work if sequence numbers are always monotonically increasing across
shards. Will need some investigation to see if this is feasible.
    Either way, I think this is an improvement out-of-scope for the issue at hand, and would
also require some migration path from the old state (where this constant threshold state doesn't
    For 1.4.1 bugfix, I think we should continue with the current approach.
    It might make sense though, to fix this via a constant threshold state in 1.5.0.


View raw message