datarace

A datarace, or data race, is a condition in concurrent programming where two or more threads access the same memory location concurrently, and at least one of the accesses is a write, with no proper synchronization or ordering guarantees between the accesses. When a datarace can occur, the program’s behavior becomes unpredictable and often undefined depending on the language and memory model.

Dataraces arise from shared mutable state that is accessed without appropriate synchronization. Causes include omissions of mutexes or locks, using non-thread-safe data structures, improper use of atomic operations, or complex interactions between threads where one thread updates data while another reads or writes it at the same time. The consequences can include corrupted data, lost updates, stale reads, crashes, or security vulnerabilities, and problems may be intermittent and hard to reproduce.

The formal treatment of dataraces varies by language. In C and C++11 and later, a data race is defined as concurrent conflicting accesses to the same atomic non-atomic object without a happens-before relationship, yielding undefined behavior. In Java, a data race occurs when unsynchronized reads and writes to shared variables happen without a happens-before ordering, potentially leading to visibility and ordering issues; the Java Memory Model provides mechanisms (synchronized blocks, volatile fields, and concurrent data structures) to prevent races.

Detection and mitigation efforts include using race detectors (such as ThreadSanitizer or Helgrind), static analysis, and careful design practices. Mitigations focus on proper synchronization (mutexes, locks, atomic operations), using thread-safe or immutable data, reducing shared state, and avoiding data-dependent race conditions.