Fuzzball Documentation
Toggle Dark/Light/Auto mode Toggle Dark/Light/Auto mode Toggle Dark/Light/Auto mode Back to homepage

Job Hostnames

Fuzzball automatically manages hostname resolution within the containers that support jobs. Fuzzball creates a new subdomain for each workflow and gives each container a hostname matching the name of the job it supports, so jobs in a workflow can reach one another by job name without knowing any IP addresses. Fuzzball also builds a longer fully qualified domain name (FQDN) for each container from the job name, the workflow UUID, the group UUID, and the domain suffix fuzzball.

Consider a job in a workflow and group with these identifiers:

  • Job name: myjob
  • Workflow UUID: 29344cd9-72a1-426b-a4c9-e724177b7929
  • Group UUID: 3f363570-30d8-4e55-84ec-6046dcca6126

Fuzzball creates a subdomain for the workflow and assigns the container this FQDN:

myjob.29344cd9-72a1-426b-a4c9-e724177b7929.3f363570-30d8-4e55-84ec-6046dcca6126.fuzzball
|---| |----------------------------------| |----------------------------------| |------|
 job              workflow UUID                       group UUID                 domain

Other jobs in the workflow can use the short hostname myjob to resolve the IP address of the container.

Job and service name uniqueness

Jobs and services share one DNS namespace within a workflow, so a job and a service in the same workflow cannot have the same name. Submitting a workflow with a duplicate name fails validation:

workflow validation failed: name "myjob" is used by both a job and a service; job and service names must be unique within a workflow

Without this rule a name used by both would be ambiguous: a container asking for database could not reliably reach the right IP address.

Images, volumes, and ingress and egress file IDs share this namespace too, so every name in a workflow must be distinct across all of those sections, not just across jobs and services.

Task array and multinode job hostnames

Fuzzball appends an integer to task array and multinode job hostnames so their containers can be told apart. The bare job name does not resolve for these jobs – only the numbered forms do.

For a task array, the integer is the task ID. A task array named myjob produces the hostname myjob-1 for the task with FB_TASK_ID=1.

For a multinode job, the integer is the node index, starting at 0. A multinode job named myjob running on three nodes produces the hostnames myjob-0, myjob-1, and myjob-2.

A multinode hostname identifies a node, not an MPI rank. When more than one process runs per node, several ranks share one hostname – an MPI job on two nodes with four slots each has ranks 0 through 7, but only the hostnames myjob-0 and myjob-1.

Services are reachable by their own FQDNs rather than the job forms above. See Workflow Endpoints and Connecting to Workflow Services for how to find and reach them.