MX in Java

(Usage hints for this presentation)

IT Systems, Summer Term 2024
Dr. Jens Lechtenbörger (License Information)

Chair of Data Science: Machine Learning and Data Engineering (Prof. Gieseke)
Dept. of Information Systems
University of Münster, Germany

1. Introduction

1.1. Thread Terminology

1.2. Thread States

1.3. Java Threads

1.4. Races

1.5. Mutual Exclusion

2. Monitors

2.1. Monitor Idea

Monitor ≈ instance of class with methods and attributes
Equip every object (= class instance) with a lock
- Automatically
  - Call lock() when method is entered
    - As usual: Thread is blocked if lock is already locked
    - Thus, automatic MX
    - We say that executing thread entered the monitor or executes inside the monitor when it has passed lock() and executes a method
  - Call unlock() when method is left
    - Thread leaves the monitor

The basic idea of monitors is as follows: Think of a monitor as an instance of a special type of class, where each instance is automatically equipped with its own lock. The run-time system ensures that before a method of such a class is executed on a class instance (which is this in Java), the lock for that class instance needs to be acquired.

We say that a thread that has successfully executed lock(), “entered the monitor” or “executes inside the monitor”.

Thus, monitors automatically provide mutual exclusion for methods of the monitor class: If multiple threads share the same object (with a potential for race conditions), only one of them can execute inside the monitor at any point in time, while others are blocked.

Importantly, each object has its own lock. Thus, two threads that operate on different class instances can both acquire their different locks and execute the same monitor method in parallel (without the danger of races as they do not share resources). Thus, to explicitly point out a frequent misunderstanding: Methods are not locked; instead, objects are protected with locks.

The next slide explains the origin of monitors in terms of an abstract data type (instead of the more modern “class” formulation presented here). On that slide, you also see that monitors not only guarantee MX; in addition, they provide methods for cooperation of threads.

Subsequent slides then discuss how the monitor concept is implemented in Java with the keyword synchronized (which activates locking of the this object as explained here in general terms) and methods for cooperation.

2.2. Monitor Origin

Monitors proposed by Hoare; 1974
Abstract data type with MX guarantee

Monitors were proposed decades ago by famous computer scientist Tony Hoare. You may also know him as developer of quicksort.

Monitors can be understood as abstract data type that guarantees MX.
- Methods encapsulate local variables
  - Just like methods in Java classes
  Importantly, just as with object-oriented programming, monitors encapsulate local variables as embedded data. Thus, all accesses to embedded data require method invocations.
- Thread enters monitor via method
  - Built-in MX: At most one thread in monitor
  Threads executing monitor methods automatically try to acquire a lock for the embedded data. As only one thread can own the lock, monitor methods guarantee MX. As stated on the previous slide, we say that the successful thread “enters” the monitor. While one thread executes inside the monitor, other threads are blocked when they attempt to enter as well.
- In addition: Methods for cooperation
  - cwait(x): Blocks calling thread until csignal(x)
    - Monitor free then
  - csignal(x): Starts at most one thread waiting for x
    - If existing; otherwise, nothing happens

3. MX with Monitors in Java

3.1. Monitors in Java: Overview

In Java, classes and objects come with built-in locks
- Which are ignored by default
You may be surprised to learn that in Java, classes and objects come with built-in locks. However, those locks are ignored by default.

Keyword synchronized activates locks
- Automatic locking of this object during execution of method
  - Automatic MX for method’s body
  - Useful if (large part of) body is a CS
  To activate those locks, the keyword synchronized is necessary. If you declare a method to be synchronized, a thread entering such a method automatically attempts to acquire the lock for the method’s this object. Thus, MX is guaranteed for the object accessed in the method’s body.
  
  This approach is particularly useful if a large part of the method is a critical section.
- E.g., for sample code from (Hailperin 2019) (for which you found races previously):
```
public synchronized void sell() {
	 if (seatsRemaining > 0) {
	   dispenseTicket();
	   seatsRemaining = seatsRemaining - 1;
	 } else displaySorrySoldOut();
}
```

3.1.1. Java, `synchronized`, `this`

Java basics, hopefully clear
- Method sell() from previous slides invoked on some object, say theater
  - Each theater has its own attribute seatsRemaining
  - seatsRemaining is really this.seatsRemaining, which is the same as theater.seatsRemaining
    - Inside the method, the name theater is unknown, theater is the this object, which is used implicitly
Without synchronized, races arise when two threads invoke sell() on the same object theater
- With synchronized, only one of the threads obtains the lock on theater, so races are prevented

3.1.2. Possible Sources of Confusion

With synchronized, locks for objects are activated
- For synchronized methods, thread needs to acquire lock for this object
To some up, with synchronized, locks for objects are activated. When executing a synchronized methods, the thread needs to acquire the lock for the this object. This procedure turns the method into a critical section with MX for the this object.

Methods cannot be locked

Note, again, that methods are not locked.
Individual attributes of the this object (e.g., seatsRemaining) are not locked
- (Which is not a problem as object-orientation recommends to encapsulate attributes, i.e., they cannot be accessed directly but only through synchronized methods)
Also, a lock for the this object is acquired. However, individual attributes of that object are not locked. If accesses to attributes are encapsulated by methods, as they should be, this does not pose a problem for MX.

3.1.3. Self-Study Task

Inspect and understand, compile, and run this sample program, which embeds the code to sell tickets, for which you found races previously.
Change sell() to use the monitor concept, recompile, and run again. Observe the expected outcome.

(Nothing to submit here; maybe ask questions online.)

3.2. Java Monitors in Detail

MX based on monitor concept
- See Java specification if you are interested in details
Java implements the monitor concept, with details specified at the URL given here.

Every Java object (and class) comes with
- Monitor with lock (not activated by default)
  - Keyword synchronized activates lock
  In essence, MX with Java is quite simple, as every Java object is equipped with a lock. By default, however, these locks are not used. Instead, you need to use the keyword synchronized if you want threads to acquire the locks for MX.
  - For method: public synchronized methodAsCS(...) {…}
    - First thread acquires lock for this object upon call (Class object for static methods)
    - Further threads get blocked
    The simplest way to enforce MX is to declare methods operating on shared resources as synchronized. If a thread wants to execute such a synchronized method on some object, then the thread automatically attempts to acquire the lock for that object. If that lock has been taken by another thread, then the locking attempt is not successful but leads to blocking of the thread. Once the thread holding the lock leaves the method, it automatically releases the lock, and other threads blocked on that lock are made runnable by the OS. Thus, locking attempts of previously blocked threads can continue when they are scheduled again.
    
    MX with monitors is really that simple.
  - Or for block: synchronized (syncObj) {…}
    - Thread acquires lock for syncObj
    Besides, you can also use other objects for synchronization if you want to turn blocks of code into critical sections. We will not use this, however.
- Wait set (set of threads; wait() and notify(), explained next)
  
  Finally, the Java monitor concept also includes a mechanism for cooperation based on waiting and signaling. For this purpose, Java adds a wait set to each object, to be explained subsequently.

4. Cooperation with Monitors in Java

4.1. Producer/Consumer problems

Classical synchronization problems
- Producers produce data, to be consumed by consumers
- Data in shared data structure, e.g., Java array
  - Synchronization for data structure necessary
    
    So-called producer/consumer problems are classical examples for mutual exclusion and thread synchronization in OS contexts.
    
    Here, two types of threads cooperate: Producers produce some data, e.g., records, messages, or tasks, which are then consumed by consumers. Producers place the generated data into a shared data structure, from which consumers retrieve data. Thus, all these threads race around the shared data structure, with accessing code in critical sections, for which mutual exclusion is necessary.
- One or more producers
  - Generate data, e.g., records, messages, tasks
  - Place data into buffer (shared resource)
    - Two buffer variants: unbounded or bounded
    - Producer blocks, if bounded buffer is full
- One or more consumers
  - Consume data
    - Take data out of buffer
    - Consumer blocks, if buffer is empty
In general, arbitrary numbers of producers and consumers may exist, and different data structures may be used to manage the data exchange between them. Frequently, buffers are used as abstraction for such data structures. Importantly, buffers may be bounded, i.e., have limited capacity, or they may be unbounded.

If a buffer is bounded, an insert operation by a producer may be a blocking operation: If the buffer is full, the thread is blocked until a consumer freed a slot in the buffer.

Subsequently, you will see a method to insert data into a bounded buffer, which is based on a Java array.

A get or retrieve operation by a consumer may also be blocking: If the buffer is empty, the thread is blocked until a producer filled a slot in the buffer.

4.2. Ideas for Cooperation

Use waiting and signaling of monitors
Threads may work with different roles on shared data structures
- E.g., producer/consumer problems on previous slide
Some may find that they cannot continue before others did their work
- The former call wait() and hope for notify() by the latter
- Cooperation (orthogonal to and not necessary for MX!)
  - Wait set mentioned above and explained subsequently

4.3. `wait()` and `notify()` in Java

Waiting via blocking
- wait(): thread unlocks and leaves monitor, enters wait set
  - Thread enters state blocked
  - Called by thread that cannot continue (without work/help of another thread)
If a thread finds that it cannot make use of the shared resource in the current state, it can invoke wait() to release the lock on that resource and leave its monitor. At that point in time, the thread’s state changes to blocked, and the thread is recorded in a special data structure associated with the object, called wait set. In the wait set, Java keeps track of all threads that have invoked wait() on the object.

Once a thread has executed wait(), the object’s lock is released, and other threads can acquire the object’s lock and modify the object’s state.
Notifications
- notify()
  - Remove one thread from wait set (if such a thread exists)
    - Change its state from blocked to runnable
  - Called by thread whose work may help another thread to continue
- notifyAll()
  - Remove all threads from wait set
    - Only one can lock and enter the monitor, of course
    - Only after the notifying thread has left the monitor, of course
    - Overhead (may be avoidable with appropriate synchronization objects)
If a thread has modified the object’s state in such a way that there is reason to believe that waiting threads might now be able to continue, the thread invokes notify() on the object, which removes one thread from the wait set and makes it runnable. When that runnable thread is scheduled for execution later on, it can again try to enter the monitor by locking the object; once the lock has been acquired, the thread resumes execution after the wait() method.

The method notifyAll() is an alternative to notify() that removes all threads from the wait set, not just one. You may want to think about advantages and disadvantage of notifying all waiting threads yourself.

4.4. Sample `synchronized` Java Method

// Based on Fig. 4.17 of [Hai17]
public synchronized void insert(Object o)
  throws InterruptedException
// Called by producer thread
{
  while(numOccupied == buffer.length)
      // block thread as buffer is full;
      // cooperation from consumer required to unblock
      wait();
  buffer[(firstOccupied + numOccupied) % buffer.length] = o;
  numOccupied++;
  // in case any retrieves are waiting for data, wake/unblock them
  notifyAll();
}

(Part of SynchronizedBoundedBuffer.java)

This slide shows the synchronized method insert() of a bounded buffer, which is called by producers to insert some object. The URL points to the full example, which also includes a method retrieve() for consumers. In the same directory, thread classes are available as well.

As explained earlier, the method is synchronized to guarantee MX, which prevents races on the shared data structure.

Initially, the producer checks whether the buffer is full. If so, it calls wait(), hoping for a notification from a consumer thread. Otherwise, the producer inserts the object at a free position.

This buffer implementation makes use of an array, which has a fixed length. Note how two variables, firstOccupied and numOccupied combined with modulo arithmetic determine a free slot in the array.

Eventually, the producer calls notifyAll(), which unblocks all threads waiting on the buffer. In particular, consumers that were waiting for a new object in the buffer can now try again to retrieve an object.

4.5. Comments on `synchronized`

Previous method in larger program: bb.zip
- SynchronizedBoundedBuffer as shared resource
- Different threads (Producer instances and Consumer instances) call synchronized methods on that bounded buffer
  - Before methods are executed, lock of buffer needs to be acquired
    - This enforces MX for methods insert() and retrieve()
  - In methods, threads call wait() on buffer if unable to continue
    - this object used implicitly as target of wait()
    - Thread enters wait set of buffer
    - Until notifyAll() on same buffer
  - Note that thread classes contain neither synchronized nor wait/notify

Bibliography

Hailperin, Max. 2019. Operating Systems and Middleware – Supporting Controlled Interaction. revised edition 1.3.1. https://gustavus.edu/mcs/max/os-book/.

License Information

Source files are available on GitLab (check out embedded submodules) under free licenses. Icons of custom controls are by @fontawesome, released under CC BY 4.0.