Status
Current state: Done
Discussion thread: here (<- link to https://mail-archives.apache.org/mod_mbox/flink-dev/)
...
Page properties | ||||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Document the state by adding a label to the FLIP page with one of "discussion", "accepted", "released", "rejected".
|
...
|
Motivation
Currently, Flink processes (JobManager and TaskManager) can only persist information via Flink's HA services. The stored information is usually cluster-scoped, meaning that it is currently not easily possible to store process specific information. Moreover, for storing large blobs Flink currently needs to go to a remote object store from which it needs to download these blobs when restarting. If the Flink processes were restarted on the same node or on a node where the same volume was mounted to, then we could simply use the local disk/mounted volume to store process specific information. For example, we could use this volume to store local state. This would allow us to use local state when recovering from a process failure. Flink could also store the cached blobs on this volume which would avoid having to download them again from the BlobServer.
...