Limitations

NFS Mount Points

Although Flamenco supports different mount base-file paths for different operating systems (aka platforms), it assumes the same base-file path for a specific platform across all those workstations. So two MacOS Users must have the same pre-defined mount point (eg: /private/nfs/flamenco) so that Flamenco can path-swap MacOS to Linux or MacOS to Windows consistently. (more details)

Flamenco Task Concurrent Count

Each Flamenco Task has a default of 1 (one) frame. This can be increased although this will increase the number of concurrent tasks queued up. This may cause instability or errors with Flamenco Manager depending how large this queue becomes. Bug 104201

Flamenco Task Timeouts

Tasks will timeout if the rendering takes longer than a pre-defined maximum duration. In this project the pre-defined timeout for a Worker is 5minutes, and for a Task is 5 hours. This is configured in the Flamenco Manager configuration file.

You can change these timeouts by editing the flamenco-manager.yaml template file in the build-docker-registry Role that is included in the Flamenco + Blender container.

Blender Project Size

The Blender Add-On will attempt to post the Blender Project file to Flamenco through the Gateway Device (NGINX Proxy) and to the Flamenco Manager in the Cluster.

The NGINX Proxy and NGINX Ingress controller in the cluster both need to be provisioned to accept >1MB (default) HTTP BODY sizes. This can be done in gateway-install and build-cluster-foundation Ansible Roles.

(NGINX Proxy) (NGINX Ingress details)

Worker CPU Assignment

In this project, a Worker is assigned 1 CPU core. (This can be changed by configuring the Worker deployment manifest in kubernetes.)

Worker GPU Assignment

GPUs cannot be shared between Workers. There is a subset of Workers (equal to the number of nodes with GPUs) that have access to 1 GPU per Worker per node.

Worker Memory Exhaustion

In this project, a Worker is assigned a maximum of 2GB RAM that Blender (and the Flamenco Worker) can consume. (This can be changed by configuring the Worker deployment manifest in kubernetes.)

Manager Restarts

If the Flamenco Manager restarts, then the Worker deployment will have to be re-deployed as existing Workers will not automatically re-register.

Blender Versions

Flamenco assumes one version of Flamenco and Blender across all Flamenco Workers. It does not support the notion of a specific Blender project being rendered on a specific version of Blender. At this time, only one Blender version is supported in this project.

Blender Failures

Flamenco Manager will attempt to schedule rendering on suitable Flamenco Workers and if jobs fail to complete on any Flamenco Worker they may retry a few times, but then are removed from the available pool of Flamenco Workers for future attempts.

This can be for any number of reasons related to the rendering of a project. Large projects, large image sizes can require more memory or compute than is available on any Flamenco Worker. In addition, depending on the the settings, Flamenco Worker may not be able to perform the render (eg: cycles vs evee or CPU vs GPU).

Flamenco Manager provides “Task” logs which will contain the console log output for Blender that may give you clues to the failure.