Skip to main content

Dynamic Parameter Updates

Dynamically update selected runtime parameters of a running deployment so changes take effect in seconds instead of minutes. This capability reduces downtime from full restarts and enables quick, safe scaling for TaskManagers as well as checkpoint‑related tuning.

Overview​

When you apply a traditional configuration change, the deployment must stop and restart, which can interrupt traffic, trigger state backfills, and cause short‑term resource spikes. With dynamic parameter updates, VERA applies supported changes to a running job via an in‑place update. The JobManager and TaskManagers are reused; depending on the change, VERA may perform a lightweight operator restart or apply the change without any restart.

Before you begin​

  • Product/Version: Supported on vera-xxx or later.
  • Deployment state: The target deployment must be Running.
  • Feature maturity: This is an experimental capability. Brief service interruptions can still occur while changes are applied.

Supported parameters​

You can dynamically update only the parameters listed below. Changes to any other parameter require a standard restart to take effect.

  • Parallelism Used for dynamic scaling of TaskManagers/operators.

  • Checkpointing interval The target interval between checkpoints.

  • Checkpoint timeout Maximum time allowed for a checkpoint to complete.

  • Minimum pause between checkpoints Minimum spacing between successive checkpoints.

Important

  • Expert (fine‑grained) resource mode does not support dynamic updates of Parallelism.
  • You can only scale operators whose parallelism is not explicitly fixed in code or configuration. Operators that require specific parallelism (for example, a global operator that must be 1, or a Kafka source bounded by partition count) will not change dynamically if set explicitly. To allow dynamic scaling, avoid DataStream#setParallelism(...) or hard‑coding source/sink parallelism unless required by your design.

How it works (at a glance)​

  1. You propose a change to one or more supported parameters in the deployment configuration.

  2. VERA validates the change for dynamic applicability.

  3. If eligible, VERA performs an in‑place update:

    • No replacement of the JobManager/TaskManagers.
    • Either a targeted operator restart (in‑place) or no restart when safe.
    • Can leverage resource pre‑application and lazy state loading to minimize interruption.

The net effect is a shorter pause for data processing compared to a full redeploy.

Usage notes​

  • Typical interruption is ~5 seconds to 1 minute, but depends on your job graph and state size.
  • Dynamic updates are evaluated per change. If a change set mixes supported and unsupported parameters, only the supported ones can be applied dynamically; the rest require a restart.
  • For dynamic scaling, ensure sources/sinks can handle the new parallelism (for example, Kafka partitions ≥ desired source parallelism).

Apply a dynamic update (Console)​

  1. Open Deployments

    Navigate to Operation Center → Deployments and open the deployment that is Running.

  2. Edit configuration

    On the Configuration tab, choose Resources or Parameters, then click Edit.

  3. Change supported values

    Modify any of: Parallelism, Checkpointing interval, Checkpoint timeout, Minimum pause between checkpoints. Click Save.

  4. Apply without restart

    In the upper‑right corner, click Dynamic update (may also appear as Hot update). Confirm the operation.

  5. Verify

    Watch the deployment status and metrics. A dynamic‑update indicator appears while the change is applied.

If you changed other, unsupported parameters: perform a standard restart for those changes to take effect.

Rollback​

If the change degrades performance or stability:

  • Revert the parameter to its previous value and perform Dynamic update again.
  • Perform a standard restart with the prior configuration to return to a known good state.

Limitations & compatibility​

  • VERA: placeholder
  • Expert mode: Parallelism cannot be updated dynamically.
  • Operator‑level constraints: Nodes with explicitly specified parallelism will not scale dynamically. Certain operators have fixed requirements (for example, global operators at parallelism 1).

Troubleshooting​

Dynamic update button is not visible

  • Confirm you edited only supported parameters.
  • Ensure the deployment is Running and your engine is vera-placeholder.

Update takes longer than expected

  • Large keyed state or complex topologies increase pause time. Consider reducing state size, enabling/increasing incremental checkpoints, or applying changes during low‑traffic windows.

Parallelism change not applied to all operators

  • Check for operators with explicit parallelism in code or connector configuration (e.g., a fixed source/sink parallelism or a global operator). Remove/relax those constraints if dynamic scaling is desired and safe.
On this page