Since the introduction of the concept of multi-threading, there has been a drastic change in the way we code. Now multiple threads run in parallel, hence many tasks can be performed at the same time.
Salesforce too adhered to this concept taking a step forward in its endeavor and provided a multi-threaded environment.
But everything has its pros and cons and so did parallelism. Though it completely changed the way programs are executed but brought with it a new concurrency bug what we call as a race condition.

When does it actually occur?

Race condition occurs when two thread operates on the same object without proper synchronization and their operation interleaves on each other.

To prevent itself from becoming a prey to this predator, Salesforce introduced locks as they are one of the best practices to avoid a race condition.
But folks, locks are associated with the ever famous deadlocking issues since their inception. To prevent deadlock from occurring, Salesforce gave some hands-on approaches.
In the optimistic approach, a transaction waits for 10 seconds to obtain the lock. If the lock is not granted during this time period to the transaction, an exception stating unable to obtain a lock will be thrown. In a pessimistic approach, a transaction needs the lock immediately, failing which an exception would be thrown.
But deadlock may occur in salesforce in case of faulty designs i.e. it is totally dependent on the developer’s design.

Now let’s have a scenario-based discussion on whether race condition exists in Salesforce or not

Consider a situation where you load a humongous amount of data using the workbench. Each batch of the humongous chunk has a trigger associated with it which affects the field of another object. A trigger is so written that it is associated with large processing time and the same field of another object is affected by each batch. So, can the field be affected in a sequential manner concurrently? It seems to be difficult, but there are some approaches you can try: (All the described approaches are explained taking into consideration the above scenario)

Using Future Methods

Future methods are asynchronous methods that run in the background, and we can use them to obtain parallelism. If you have a small and dicey job, you may get results that depict that the task is running sequentially (it seems to be a single-threaded environment) but unfortunately, that’s misleading, and as soon as the jobs assigned to these methods become tedious you will see unexpected results. If you get the expected results in the scenario then it’s a day where you can think of yourself as being born with a silver spoon in your mouth.

Using Queueable Interface

Isn’t the Queueable method asynchronous? If yes, then won’t they show the same characteristics as future methods? If so why are we talking about them?

These are the questions that arise in the moment of heat and the only point that makes queueable interfaces part of the discussion is that they provide a great advantage over future methods through chaining.
One queueable job can be chained to another queueable job by enqueuing the first job in the execute () method of the second job. Not only this, but a queueable job can also be enqueued by itself. This provides a recursive approach.
Every enqueued job, a process in a new transaction, and multiple enqueued jobs run in parallel. As a result, one enqueued job would not be able to determine whether other enqueued jobs have been completed or not.
But today everything has workarounds, and workarounds to this problem are:

Note:

First thought that strikes everyone’s mind is to use a static Boolean variable to obtain mutual exclusion just like a semaphore to maintain data integrity. If one job is being processed within a semaphore block then all the other jobs trying to act on the field will enqueue themselves again till the block or critical section is free i.e. no other job is acting on it.
But this only seems satisfying, as a static variable can only persist values for a single transaction and they are only static within the scope of the request and not throughout the organization. So this is not an approach we are looking for.

Use of custom settings as flags or locks

As list custom setting contains static data that can be reused and accessed easily without repeated queries on the database, it can be considered for developing a locking environment. This kind of environment can be provided by creating a field in the list custom setting which will act as lock as a static variable did but this time it will be static throughout the organization. This approach may or may not work as there may be conditions where the field may have been read and used by one or more jobs and during the same period it may be updated by some other job, so it leads to the possibility of complete distortion of the control we wanted to obtain.

Using FOR UPDATE to obtain locks.

It’s quite shocking that all the approaches talk about locks, and none of them have considered the FOR UPDATE lock provided by Salesforce itself, it’s because for update does not provide a failure-safe environment i.e. it provides DML and QUERY Exception interfaces to work on but doesn’t provide an interface to handle them. Using FOR UPDATE with a queueable interface can serve as a solution we are looking for. As queueable interface can be used to obtain reclusiveness and every time an exception is raised by FOR UPDATE, it can be handled by enqueuing the same enqueue job again.

Use Queue Data Structure

Queue work in FIFO (First in First out) manner. The result of every job can be pushed(enqueued) into the queue and then a schedulable job can be used to dequeue the job at some point in time.
The persistent queue can be designed and developed through a custom object (with fields as per need). In this custom object, a record with the result of a job can be created as soon as a job is completed (enqueuing of the queue). A job can then be scheduled for processing out the results (dequeuing of the queue) within the custom object.

Note: Queue data structure can also be developed using the Apex List Collection class. But persistence comes into play and we have already discussed that making anything static won’t help.

Conclusion

Race condition is not a myth in Salesforce. It can occur in some situations depending on the jobs that are onboard. But there are few ways and approaches that can be tried to avoid a race condition. Efficient designing is the key to avoid deadlocks for developers. As most of the deadlock avoidance tricks can be applied during the designing phase itself.