How to ensure that a Scheduled Job runs only on one instance out of 20 in a cluster?

To ensure that a Scheduled Job runs only on one instance out of 20 in a cluster, a distributed locking mechanism can be used. Main approaches:

1. Using a database with a lock flag:

   • When starting the task, the instance attempts to set a flag in the database (for example, in a table with a single record).
   • If the flag is free, the instance sets it and executes the task.
   • After completion, it releases the flag.
   • Other instances see that the task is already running and skip the start.

2. Using a distributed cache with lock support (Redis, ZooKeeper, etcd):

   • For example, with Redis, you can use Redlock — a distributed lock algorithm.
   • The instance attempts to acquire a lock with TTL.
   • If successful, it executes the task.
   • Upon completion, it releases the lock.

3. Using specialized libraries:

   • Spring Boot with @Scheduled and libraries supporting distributed tasks, such as Quartz with clustering.

Example with Redis (pseudocode):

String lockKey = "scheduledJobLock";
boolean acquired = redisClient.setIfAbsent(lockKey, instanceId, ttlSeconds);
if (acquired) {
    try {
        // execute task
    } finally {
        redisClient.delete(lockKey);
    }
}

Thus, only one instance can acquire the lock and execute the task, while others will skip the start.