ISPN-14654 MSETNX

[rigazilla]

https://issues.redhat.com/browse/ISPN-14654

[tristantarrant]

Please add a row to the table in documentation/src/main/asciidoc/topics/ref_redis_commands.adoc

[rigazilla]

With the current implementation:

1. entries = getAll(keys)
2. if entries.size()==0 then putAll(newEntries)

if an concurrent put(conflictingEntry) operation happens btween 1 and 2, there's no way to keep the semantic of the operation, I mean:

• order (msetnx then set) must result in conflictEntry to be present, but it has been actually overwritten
• order (set then msetnx) must result in msetnx not be executed since concEntry is present.

Replacing putAll with a loop of putIfAbsent results in a consistent story (msetnx then set), the cost is to replace an atomic operation with a loop.

I can't see any better solution without involving transactions.

any preference/though?

[wburns]

I don't think there is any way to do this without transactions or some type of 2 step operation when in a cluster. We can do simpler things with a single node, but in a cluster we really need something like a distributed transaction, a new command won't remedy that sadly.

[tristantarrant]

Rebase please

[rigazilla]

rebased

[tristantarrant]

@rigazilla needs rebase. @karesti @jabolina can you review?

[wburns]

Seems fine to me, if anyone else wants to double check.

[jabolina]

Thanks, @rigazilla!