[FLINK-3556] [runtime] Remove false check in HA blob store configuration

[uce]

If you use the RocksDBStateBackend and configure HA w/o setting the state backend to filesystem, restoring your job will fail, because the BLOBs are not persisted. The really ugly part is that this does not log a warning or anything. I'm leaning towards making this a release blocker, but it can also be postponed to 1.0.1, if that follows very soon.

[tillrohrmann]

Changes look good to me. Good catch @uce :-) +1 for merging.

One minor thing: We check that the configuration contains the ZOOKEEPER_RECOVERY_PATH and then give the configuration to the FileSystemBlobStore. Wouldn't it make more sense to directly pass in the value of ZOOKEEPER_RECOVERY_PATH or do the checking of the correctness of the value in the FileSystemBlobStore? If we pass the configuration, then we don't actually know, which values will be used internally. This would untie the BlobServer and the FileSystemBlobStore a bit.

[uce]

Thanks Till! The check was duplicated in BlobServer and FileSystemBlobStore. Now only config is passed and it's the responsibility of FileSystemBlobStore to construct itself correctly or fail.

[uce]

PS I had a local Travis build with this (1st commit), which succeeded https://travis-ci.org/uce/flink/builds/112830163 and the blob and recovery tests succeed on my local laptop with the latest commit.

[tillrohrmann]

+1 for merging

[uce]

OK, going to merge to master and release-1.0.