Home > Back-end >  Is replacing one java object reference with another considered thread safe, or are there potential s
Is replacing one java object reference with another considered thread safe, or are there potential s

Time:06-16

I'm rolling my own simple caching solution to improve the UX of a lengthy lookup. The basic outline is that I have a class handling all accountId lookups, and another class called AccountListCache with two variables: a timestamp indicating when it was created, and a List object containing a whole bunch of accountId objects. When the lookup is invoked, that class will first check if its cache object has a timestamp from within the past X minutes. If so, it serves the cached results, and if not it refreshes the cache and serves the new list.

My plan is to have a scheduled task that regularly refreshes the cache by creating a new instance of the AccountListCache, populating its list/timestamp, and then re-assigning the variable in the lookup class. My boss brought up concerns about thread safety, but I'm thinking this should be thread safe already.

The old list and the new list are separate objects contained within separate instances of the same class, and the actual update call that matters would just be changing where the reference is pointing in memory. That should be an effectively instantaneous/atomic operation, right?

If it's not thread safe, what does a failure/conflict look like, and how might I resolve it?

CodePudding user response:

Yes, since one thread is writing data to be read by other threads, there are concurrency issues here.

While writing and reading object references are atomic, in the sense that "tearing" is guaranteed not to occur (unlike double or long types), there is no guarantee that writes will ever be visible to other threads unless a memory barrier is used.

In this case, at a minimum, I would make the AccountListCache field in your lookup class volatile. This will ensure that any state of the cache modified before it is assigned to the field is visible to any thread that reads the field.

CodePudding user response:

The Answer by erickson is correct.

As an alternative to volatile, I prefer using AtomicReference for two reasons:

  • Many programmers do not fully understand volatile, both in general and specifically because it’s meaning changed when the Java Memory Model was revised.
  • The AtomicReference really screams out to the human readers that we have a concurrency issue to manage.

Example of the atomic reference. Notice the final to guarantee that we have one and only one AtomicReference instance.

final AtomicReference < AccountListCache > cacheRef = new AtomicReference<>( … ) ;

I would use a record to hold your current cache value, a timestamp with a list. A record’s member field references are immutable (has getters but no setters).

record AccountListCache ( Instant whenFetched , List< Account > ) {}

Use List.copyOf to get an unmodifiable list.

You could add a constructor to the record for data validation. You could check that the passed instant is before now. And you could make sure the list is not empty (if that’s intolerable). And you could put the passed list through List.copyOf to ensure it is not modifiable (copyOf is a no-op if already unmodifiable).

Use a ScheduledExecutorService to repeatedly replace the record reference in the AtomicReference with a reference to a new fresh instance of the AccountListCache record class.

Any code using the cache can verify how fresh or stale it may be.

Duration ageOfCache = 
    Duration.between ( 
        cacheRef.get().whenFetched() , 
        Instant.now() 
    ) 
;

Or make that a method on the record.

record AccountListCache ( Instant whenFetched , List< Account > ) 
{
    Duration age () { return Duration.between ( this.whenFetched , Instant.now() ) } 
}

Usage:

Duration ageOfCache = cacheRef.get().age() ;
  • Related