[jira] [Created] (SLIDER-1242) Review uses of double-checked locking

["Billie Rinaldi (JIRA)" <ji...@apache.org>]

Billie Rinaldi created SLIDER-1242:
--------------------------------------

             Summary: Review uses of double-checked locking
                 Key: SLIDER-1242
                 URL: https://issues.apache.org/jira/browse/SLIDER-1242
             Project: Slider
          Issue Type: Bug
            Reporter: Billie Rinaldi
            Assignee: Billie Rinaldi
             Fix For: Slider 1.0.0


There are several places where we perform double-checked locking. Even though the practice is discouraged, I believe it is technically correct when the check is performed on the presence of a ConcurrentHashMap key, which is how we are using it.

However, in two places, AgentProviderService#getCurrentExports and AgentProviderService#getAllocatedPorts, containsKey is used instead of get to perform the check. I am seeing some indication that containsKey is not sufficient, and that get must be used for double-checked locking to be correct. There is a comment in the ConcurrentHashMap#containsKey method that says "same as get() except no need for volatile value read" -- and I think that volatile value read is what we need for correctness.

Also, in the [ConcurrentHashMap api doc|https://docs.oracle.com/javase/8/docs/api/java/util/concurrent/ConcurrentHashMap.html], it specifically mentions get and does not mention containsKey: "Any non-null result returned from get(key) and related access methods bears a happens-before relation with the associated insertion or update" and "an update operation for a given key bears a happens-before relation with any (non-null) retrieval for that key reporting the updated value."



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)