Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/6147#discussion_r195065351
--- Diff:
flink-core/src/main/java/org/apache/flink/api/common/cache/DistributedCache.java
---
@@ -40,6 +41,14 @@
@Public
public class DistributedCache {
+ /**
+ * An entry for a single file or directory that should be cached.
+ *
+ * <p>Entries have different semantics for local directories depending
on where we are in the job-submission process.
+ * After registration through the API {@code filePath} denotes the
original directory.
+ * Before the job is submitted to the cluster directories are zipped,
at which point {@code filePath} denotes the path to the local zip.
+ * After the upload to the cluster, {@code filePath} denotes the
(server-side) copy of the zip.
+ */
public static class DistributedCacheEntry implements Serializable {
--- End diff --
It might be out of scope of this PR but I think the `DistributedCacheEntry`
mixes too many responsibilities. On the one hand it is used to transport cache
entry information like `isZipped`, `blobKey` and `isExecutable` which is only
relevant for the job submission. On the other hand, it also contains
information about which files to transmit to the cluster at the job creation
time. I think it would be a good idea to separate these responsibilities. As a
side effect, we would not have `nullable` fields such as the `blobKey` in this
class.
---
