Ververica Platform 2.3.2¶
Release Date: 2020-12-01
- The memory configuration for Flink containers is translated to base 2 before being passed to Flink. Flink interprets the
munit as base 2 whereas Ververica Platform and Kubernetes interpret
mas base 10. This mismatch could result in containers being OOM killed, because Flink was allowed to allocate more memory than available to the container.
- For low memory configurations with less than 1000m (base 2) of total process memory for TaskManagers, we reduce the managed memory fraction
taskmanager.memory.managed.fractionto 0.2 (default: 0.4). This is required in order to account for the harmonized unit computation with our default configuration of 1G (base 10) for TaskManager memory. If the resulting total process memory is greater or equal to 1000m or the managed memory fraction has been configured manually, we do not set it.
- We now set the Kubernetes resources for the ResultFetcher service. The ResultFetcher is deployed as a sidecar container for SessionClusters and it is responsible to fetch the results of your SQL preview queries. Before this change, no Kubernetes resources were set which could lead to rejected SessionClusters due to Kubernetes resource quotas.
vvp: resultFetcher: resources: limits: cpu: 500m memory: 500Mi requests: cpu: 100m memory: 300Mi
Artifacts bundled as part of custom Flink Docker images can now be referenced in the artifact section of Deployments via the
file://scheme. During startup, referenced dependencies will be linked to Flink’s usrlib directory (/flink/usrlib). It is possible to mix bundled and external dependencies.
This is only supported with the newly released Flink images mentioned above and future versions.
Note that the bundled artifact(s) must be placed under a different location than /flink/usrlib, because this directory is the mount point for dynamically fetched artifacts.
Environment variables referenced in the Flink configuration section are now substituted during runtime of the Flink container.
We have added the following two labels to all Kubernetes resources created for Deployments and SessionClusters:
- The SessionCluster form now has feature parity with Deployment forms, including the YAML editor and additional Flink configuration validation.
We recommend upgrading via Helm using the following commands:
$ helm repo add ververica https://charts.ververica.com $ helm upgrade [RELEASE] ververica/ververica-platform --version 4.3.2 --values custom-values.yaml