Flink is a distributed system and requires effective allocation and management of compute resources in order to execute streaming applications. It integrates with all common cluster resource managers such as Hadoop YARN, Apache Mesos and Kubernetes, but can also be set up to run as a standalone cluster or even as a library.
This section contains an overview of Flink’s architecture and describes how its main components interact to execute applications and recover from failures.
The Flink runtime consists of two types of processes: a JobManager and one or more TaskManagers.
The Client is not part of the runtime and program execution, but is used to
prepare and send a dataflow to the JobManager. After that, the client can
disconnect (detached mode), or stay connected to receive progress reports
(attached mode). The client runs either as part of the Java/Scala program
that triggers the execution, or in the command line process
The JobManager and TaskManagers can be started in various ways: directly on the machines as a standalone cluster, in containers, or managed by resource frameworks like YARN or Mesos. TaskManagers connect to JobManagers, announcing themselves as available, and are assigned work.
The JobManager has a number of responsibilities related to coordinating the distributed execution of Flink Applications: it decides when to schedule the next task (or set of tasks), reacts to finished tasks or execution failures, coordinates checkpoints, and coordinates recovery on failures, among others. This process consists of three different components:
The ResourceManager is responsible for resource de-/allocation and provisioning in a Flink cluster — it manages task slots, which are the unit of resource scheduling in a Flink cluster (see TaskManagers). Flink implements multiple ResourceManagers for different environments and resource providers such as YARN, Mesos, Kubernetes and standalone deployments. In a standalone setup, the ResourceManager can only distribute the slots of available TaskManagers and cannot start new TaskManagers on its own.
The Dispatcher provides a REST interface to submit Flink applications for execution and starts a new JobMaster for each submitted job. It also runs the Flink WebUI to provide information about job executions.
A JobMaster is responsible for managing the execution of a single JobGraph. Multiple jobs can run simultaneously in a Flink cluster, each having its own JobMaster.
There is always at least one JobManager. A high-availability setup might have multiple JobManagers, one of which is always the leader, and the others are standby (see High Availability (HA)).
The TaskManagers (also called workers) execute the tasks of a dataflow, and buffer and exchange the data streams.
There must always be at least one TaskManager. The smallest unit of resource scheduling in a TaskManager is a task slot. The number of task slots in a TaskManager indicates the number of concurrent processing tasks. Note that multiple operators may execute in a task slot (see Tasks and Operator Chains).
For distributed execution, Flink chains operator subtasks together into tasks. Each task is executed by one thread. Chaining operators together into tasks is a useful optimization: it reduces the overhead of thread-to-thread handover and buffering, and increases overall throughput while decreasing latency. The chaining behavior can be configured; see the chaining docs for details.
The sample dataflow in the figure below is executed with five subtasks, and hence with five parallel threads.
Each worker (TaskManager) is a JVM process, and may execute one or more subtasks in separate threads. To control how many tasks a TaskManager accepts, it has so called task slots (at least one).
Each task slot represents a fixed subset of resources of the TaskManager. A TaskManager with three slots, for example, will dedicate 1/3 of its managed memory to each slot. Slotting the resources means that a subtask will not compete with subtasks from other jobs for managed memory, but instead has a certain amount of reserved managed memory. Note that no CPU isolation happens here; currently slots only separate the managed memory of tasks.
By adjusting the number of task slots, users can define how subtasks are isolated from each other. Having one slot per TaskManager means that each task group runs in a separate JVM (which can be started in a separate container, for example). Having multiple slots means more subtasks share the same JVM. Tasks in the same JVM share TCP connections (via multiplexing) and heartbeat messages. They may also share data sets and data structures, thus reducing the per-task overhead.
By default, Flink allows subtasks to share slots even if they are subtasks of different tasks, so long as they are from the same job. The result is that one slot may hold an entire pipeline of the job. Allowing this slot sharing has two main benefits:
A Flink cluster needs exactly as many task slots as the highest parallelism used in the job. No need to calculate how many tasks (with varying parallelism) a program contains in total.
It is easier to get better resource utilization. Without slot sharing, the non-intensive source/map() subtasks would block as many resources as the resource intensive window subtasks. With slot sharing, increasing the base parallelism in our example from two to six yields full utilization of the slotted resources, while making sure that the heavy subtasks are fairly distributed among the TaskManagers.
A Flink Application is any user program that spawns one or multiple Flink
jobs from its
main() method. The execution of these jobs can happen in a
local JVM (
LocalEnvironment) or on a remote setup of clusters with multiple
RemoteEnvironment). For each program, the
ExecutionEnvironment provides methods to
control the job execution (e.g. setting the parallelism) and to interact with
the outside world (see Anatomy of a Flink Program).
The jobs of a Flink Application can either be submitted to a long-running Flink Session Cluster, a dedicated Flink Job Cluster, or a Flink Application Cluster. The difference between these options is mainly related to the cluster’s lifecycle and to resource isolation guarantees.
Cluster Lifecycle: in a Flink Session Cluster, the client connects to a pre-existing, long-running cluster that can accept multiple job submissions. Even after all jobs are finished, the cluster (and the JobManager) will keep running until the session is manually stopped. The lifetime of a Flink Session Cluster is therefore not bound to the lifetime of any Flink Job.
Resource Isolation: TaskManager slots are allocated by the ResourceManager on job submission and released once the job is finished. Because all jobs are sharing the same cluster, there is some competition for cluster resources — like network bandwidth in the submit-job phase. One limitation of this shared setup is that if one TaskManager crashes, then all jobs that have tasks running on this TaskManager will fail; in a similar way, if some fatal error occurs on the JobManager, it will affect all jobs running in the cluster.
Other considerations: having a pre-existing cluster saves a considerable amount of time applying for resources and starting TaskManagers. This is important in scenarios where the execution time of jobs is very short and a high startup time would negatively impact the end-to-end user experience — as is the case with interactive analysis of short queries, where it is desirable that jobs can quickly perform computations using existing resources.
Cluster Lifecycle: in a Flink Job Cluster, the available cluster manager (like YARN or Kubernetes) is used to spin up a cluster for each submitted job and this cluster is available to that job only. Here, the client first requests resources from the cluster manager to start the JobManager and submits the job to the Dispatcher running inside this process. TaskManagers are then lazily allocated based on the resource requirements of the job. Once the job is finished, the Flink Job Cluster is torn down.
Resource Isolation: a fatal error in the JobManager only affects the one job running in that Flink Job Cluster.
Other considerations: because the ResourceManager has to apply and wait for external resource management components to start the TaskManager processes and allocate resources, Flink Job Clusters are more suited to large jobs that are long-running, have high-stability requirements and are not sensitive to longer startup times.
Cluster Lifecycle: a Flink Application Cluster is a dedicated Flink
cluster that only executes jobs from one Flink Application and where the
main() method runs on the cluster rather than the client. The job
submission is a one-step process: you don’t need to start a Flink cluster
first and then submit a job to the existing cluster session; instead, you
package your application logic and dependencies into a executable job JAR and
the cluster entrypoint (
is responsible for calling the
main() method to extract the JobGraph.
This allows you to deploy a Flink Application like any other application on
Kubernetes, for example. The lifetime of a Flink Application Cluster is
therefore bound to the lifetime of the Flink Application.
Resource Isolation: in a Flink Application Cluster, the ResourceManager and Dispatcher are scoped to a single Flink Application, which provides a better separation of concerns than the Flink Session Cluster.