Operations and Monitoring

Local verification

python3 -m compileall -q tfmesos2
python3 -m unittest discover -s tests -v
make -C docs build

Deployment

The documentation can be published to the gh-pages branch through the repository’s deployment worktree:

make -C docs deploy

The target builds the book first, stages the generated site in a temporary worktree, and pushes only when the rendered output changed. The defaults are origin as the remote and gh-pages as the branch. Override them when needed:

make -C docs deploy DEPLOY_REMOTE=origin DEPLOY_BRANCH=gh-pages

The deployment worktree must not already exist. The command requires Git push permission for the configured remote.

A real cluster requires the task image to contain tfmesos2.server and TensorFlow. The Mesos master must be reachable from the task network, and the callback address (client_ip:port) must be reachable from the started containers.

Observability

Mesos task states and declined offers are written to the Python log. The API provides /v0/status, /v0/task/<task-id>/job, /v0/task/<task-id>/port/<port>, and /v0/task/<task-id>. After successful initialization, the framework is suppressed.

Resources

Choose CPU and memory values with enough headroom for TensorFlow, gRPC, and the container runtime. GPU offers are handled as SET or SCALAR resources; the Docker parameters depend on the configured GPU vendor.