Nomad podman Driver

Many thanks to @towe75 and Pascom for contributing this plugin to Nomad!

Features

• Use the jobs driver config to define the image for your container
• Start/stop containers with default or customer entrypoint and arguments
• Nomad runtime environment is populated
• Use Nomad alloc data in the container.
• Bind mount custom volumes into the container
• Publish ports
• Monitor the memory consumption
• Monitor CPU usage
• Task config cpu value is used to populate podman CpuShares
• Task config cores value is used to populate podman Cpuset
• Container log is forwarded to Nomad logger
• Utilize podmans --init feature
• Set username or UID used for the specified command within the container (podman --user option).
• Fine tune memory usage: standard Nomad memory resource plus additional driver specific swap, swappiness and reservation parameters, OOM handling
• Supports rootless containers with cgroup V2
• Set DNS servers, searchlist and options via Nomad dns parameters
• Support for nomad shared network namespaces and consul connect
• Quite flexible network configuration, allows to simply build pod-like structures within a nomad group

Redis Example job

Here is a simple redis "hello world" Example:

job "redis" {
  datacenters = ["dc1"]
  type        = "service"

  group "redis" {
    network {
      port "redis" { to = 6379 }
    }

    task "redis" {
      driver = "podman"

        config {
          image = "docker://redis"
          ports = ["redis"]
        }

      resources {
        cpu    = 500
        memory = 256
      }
    }
  }
}

nomad run redis.nomad

==> Monitoring evaluation "9fc25b88"
    Evaluation triggered by job "redis"
    Allocation "60fdc69b" created: node "f6bccd6d", group "redis"
    Evaluation status changed: "pending" -> "complete"
==> Evaluation "9fc25b88" finished with status "complete"

podman ps

CONTAINER ID  IMAGE                           COMMAND               CREATED         STATUS             PORTS  NAMES
6d2d700cbce6  docker.io/library/redis:latest  docker-entrypoint...  16 seconds ago  Up 16 seconds ago         redis-60fdc69b-65cb-8ece-8554-df49321b3462

Building The Driver from source

This project has a go.mod definition. So you can clone it to whatever directory you want. It is not necessary to setup a go path at all. Ensure that you use go 1.17 or newer.

git clone [email protected]:hashicorp/nomad-driver-podman
cd nomad-driver-podman
make dev

The compiled binary will be located at ./build/nomad-driver-podman.

Runtime dependencies

• Nomad 0.12.9+
• Linux host with podman installed
• For rootless containers you need a system supporting cgroup V2 and a few other things, follow this tutorial

You need a 3.0.x podman binary and a system socket activation unit, see https://www.redhat.com/sysadmin/podmans-new-rest-api

Nomad agent, nomad-driver-podman and podman will reside on the same host, so you do not have to worry about the ssh aspects of the podman api.

Ensure that Nomad can find the plugin, see plugin_dir

Driver Configuration

• volumes stanza:

  • enabled - Defaults to true. Allows tasks to bind host paths (volumes) inside their container.
  • selinuxlabel - Allows the operator to set a SELinux label to the allocation and task local bind-mounts to containers. If used with volumes.enabled set to false, the labels will still be applied to the standard binds in the container.

plugin "nomad-driver-podman" {
  config {
    volumes {
      enabled      = true
      selinuxlabel = "z"
    }
  }
}

• gc stanza:

  • container - Defaults to true. This option can be used to disable Nomad from removing a container when the task exits.

plugin "nomad-driver-podman" {
  config {
    gc {
      container = false
    }
  }
}

• recover_stopped (bool) Defaults to false. Allows the driver to start and reuse a previously stopped container after a Nomad client restart. Consider a simple single node system and a complete reboot. All previously managed containers will be reused instead of disposed and recreated.

  WARNING - use of recover_stopped may cause Nomad agent to not start on system restarts. This setting has been left in place for compatibility.

plugin "nomad-driver-podman" {
  config {
    recover_stopped = true
  }
}

• socket_path (string) Defaults to "unix:///run/podman/podman.sock" when running as root or a cgroup V1 system, and "unix:///run/user/<USER_ID>/podman/podman.sock" for rootless cgroup V2 systems

plugin "nomad-driver-podman" {
  config {
    socket_path = "unix:///run/podman/podman.sock"
  }
}

• disable_log_collection (string) Defaults to false. Setting this to true will disable Nomad logs collection of Podman tasks. If you don't rely on nomad log capabilities and exclusively use host based log aggregation, you may consider this option to disable nomad log collection overhead. Beware to you also loose automatic log rotation.

plugin "nomad-driver-podman" {
  config {
    disable_log_collection = false
  }
}

• extra_labels ([]string) Defaults to []. Setting this will automatically append Nomad-related labels to Podman tasks. Supports glob matching such as task*. Possible values are:

job_name
job_id
task_group_name
task_name
namespace
node_name
node_id

plugin "nomad-driver-podman" {
  config {
    extra_labels = ["job_name", "job_id", "task_group_name", "task_name", "namespace", "node_name", "node_id"]
  }
}

• logging stanza:

  • type - Defaults to "nomad". See the task configuration for details.
  • options - Defaults to {}. See the task configuration for details.

• client_http_timeout (string) Defaults to 60s default timeout used by http.Client requests

plugin "nomad-driver-podman" {
  config {
    client_http_timeout = "60s"
  }

Task Configuration

• image - The image to run. Accepted transports are docker (default if missing), oci-archive and docker-archive. Images reference as short-names will be treated according to user-configured preferences.

config {
  image = "docker://redis"
}

• auth - (Optional) Authenticate to the image registry using a static credential. tls_verify can be disabled for insecure registries.

config {
  image = "your.registry.tld/some/image"
  auth {
    username   = "someuser"
    password   = "sup3rs3creT"
    tls_verify = true
  }
}

• entrypoint - (Optional) A string list overriding the image's entrypoint. Defaults to the entrypoint set in the image.

config {
  entrypoint = [
    "/bin/bash",
    "-c"
  ]
}

• command - (Optional) The command to run when starting the container.

config {
  command = "some-command"
}

• args - (Optional) A list of arguments to the optional command. If no command is specified, the arguments are passed directly to the container.

config {
  args = [
    "arg1",
    "arg2",
  ]
}

• working_dir - (Optional) The working directory for the container. Defaults to the default set in the image.

config {
  working_dir = "/data"
}

• volumes - (Optional) A list of host_path:container_path:options strings to bind host paths to container paths. Named volumes are not supported.

config {
  volumes = [
    "/some/host/data:/container/data:ro,noexec"
  ]
}

• tmpfs - (Optional) A list of /container_path strings for tmpfs mount points. See podman run --tmpfs options for details.

config {
  tmpfs = [
    "/var"
  ]
}

• devices - (Optional) A list of host-device[:container-device][:permissions] definitions. Each entry adds a host device to the container. Optional permissions can be used to specify device permissions, it is combination of r for read, w for write, and m for mknod(2). See podman documentation for more details.

config {
  devices = [
    "/dev/net/tun"
  ]
}

• hostname - (Optional) The hostname to assign to the container. When launching more than one of a task (using count) with this option set, every container the task starts will have the same hostname.

• Forwarding and Exposing Ports - (Optional) See Docker Driver Configuration for details.

• init - Run an init inside the container that forwards signals and reaps processes.

config {
  init = true
}

• init_path - Path to the container-init binary.

config {
  init = true
  init_path = /usr/libexec/podman/catatonit
}

• user - Run the command as a specific user/uid within the container. See Task configuration

user = nobody

config {
}

• logging - Configure logging. See also plugin option disable_log_collection

driver = "nomad" (default) Podman redirects its combined stdout/stderr logstream directly to a Nomad fifo. Benefits of this mode are: zero overhead, don't have to worry about log rotation at system or Podman level. Downside: you cannot easily ship the logstream to a log aggregator plus stdout/stderr is multiplexed into a single stream..

config {
  logging = {
    driver = "nomad"
  }
}

driver = "journald" The container log is forwarded from Podman to the journald on your host. Next, it's pulled by the Podman API back from the journal into the Nomad fifo (controllable by disable_log_collection) Benefits: all containers can log into the host journal, you can ship a structured stream incl. metadata to your log aggregator. No log rotation at Podman level. You can add additional tags to the journal. Drawbacks: a bit more overhead, depends on Journal (will not work on WSL2). You should configure some rotation policy for your Journal. Ensure you're running Podman 3.1.0 or higher because of bugs in older versions.

config {
  logging = {
    driver = "journald"
    options = {
      "tag" = "redis"
    }
  }
}

• memory_reservation - Memory soft limit (nit = b (bytes), k (kilobytes), m (megabytes), or g (gigabytes))

After setting memory reservation, when the system detects memory contention or low memory, containers are forced to restrict their consumption to their reservation. So you should always set the value below --memory, otherwise the hard limit will take precedence. By default, memory reservation will be the same as memory limit.

config {
  memory_reservation = "100m"
}

• memory_swap - A limit value equal to memory plus swap. The swap LIMIT should always be larger than the memory value.

Unit can be b (bytes), k (kilobytes), m (megabytes), or g (gigabytes). If you don't specify a unit, b is used. Set LIMIT to -1 to enable unlimited swap.

config {
  memory_swap = "180m"
}

• memory_swappiness - Tune a container's memory swappiness behavior. Accepts an integer between 0 and 100.

config {
  memory_swappiness = 60
}

• network_mode - Set the network mode for the container.

By default the task uses the network stack defined in the task group, see network Stanza. If the groups network behavior is also undefined, it will fallback to bridge in rootful mode or slirp4netns for rootless containers.

• bridge: create a network stack on the default podman bridge.
• none: no networking
• host: use the Podman host network stack. Note: the host mode gives the container full access to local system services such as D-bus and is therefore considered insecure
• slirp4netns: use slirp4netns to create a user network stack. This is the default for rootless containers. Podman currently does not support it for root containers issue.
• container:id: reuse another podman containers network stack
• task:name-of-other-task: join the network of another task in the same allocation.

config {
  network_mode = "bridge"
}

• cap_add - (Optional) A list of Linux capabilities as strings to pass to --cap-add.

config {
  cap_add = [
    "SYS_TIME"
  ]
}

• cap_drop - (Optional) A list of Linux capabilities as strings to pass to --cap-drop.

config {
  cap_add = [
    "MKNOD"
  ]
}

• selinux_opts - (Optional) A list of process labels the container will use.

config {
  selinux_opts = [
    "type:my_container.process"
  ]
}

• sysctl - (Optional) A key-value map of sysctl configurations to set to the containers on start.

config {
  sysctl = {
    "net.core.somaxconn" = "16384"
  }
}

• privileged - (Optional) true or false (default). A privileged container turns off the security features that isolate the container from the host. Dropped Capabilities, limited devices, read-only mount points, Apparmor/SELinux separation, and Seccomp filters are all disabled.

• tty - (Optional) true or false (default). Allocate a pseudo-TTY for the container.

• labels - (Optional) Set labels on the container.

config {
  labels = {
    "nomad" = "job"
  }
}

• apparmor_profile - (Optional) Name of a apparmor profile to be used instead of the default profile. The special value unconfined disables apparmor for this container:

config {
  apparmor_profile = "your-profile"
}

• force_pull - (Optional) true or false (default). Always pull the latest image on container start.

config {
  force_pull = true
}

• readonly_rootfs - (Optional) true or false (default). Mount the rootfs as read-only.

config {
  readonly_rootfs = true
}

• ulimit - (Optional) A key-value map of ulimit configurations to set to the containers to start.

config {
  ulimit {
    nproc = "4242"
    nofile = "2048:4096"
  }

• userns - (Optional) Set the user namespace mode for the container.

config {
  userns = "keep-id:uid=200,gid=210"
}

• pids_limit - (Optional) An integer value that specifies the pid limit for the container.

config {
  pids_limit = 64
}

• image_pull_timeout - (Optional) time duration for your pull timeout (default to 5m).

config {
  image_pull_timeout = "5m"
}

Network Configuration

nomad lifecycle hooks combined with the drivers network_mode allows very flexible network namespace definitions. This feature does not build upon the native podman pod structure but simply reuses the networking namespace of one container for other tasks in the same group.

A typical example is a network server and a metric exporter or log shipping sidecar. The metric exporter needs access to i.E. a private monitoring Port which should not be exposed the the network and thus is usually bound to localhost.

The repository includes three different examples jobs for such a setup. All of them will start a nats server and a prometheus-nats-exporter using different approaches.

You can use curl to proof that the job is working correctly and that you can get prometheus metrics:

curl http://your-machine:7777/metrics

2 Task setup, server defines the network

See examples/jobs/nats_simple_pod.nomad

Here, the server task is started as main workload and the exporter runs as a poststart sidecar. Because of that, Nomad guarantees that the server is started first and thus the exporter can easily join the servers network namespace via network_mode = "task:server".

Note, that the server configuration file binds the http_port to localhost.

Be aware that ports must be defined in the parent network namespace, here server.

3 Task setup, a pause container defines the network

See examples/jobs/nats_pod.nomad

A slightly different setup is demonstrated in this job. It reassembles more closely the idea of a pod by starting a pause task, named pod via a prestart/sidecar hook.

Next, the main workload, server is started and joins the network namespace by using the network_mode = "task:pod" stanza. Finally, Nomad starts the poststart/sidecar exporter which also joins the network.

Note that all ports must be defined on the pod level.

2 Task setup, shared Nomad network namespace

See examples/jobs/nats_group.nomad

This example is very different. Both server and exporter join a network namespace which is created and managed by Nomad itself. See nomad network stanza to get started with this generic approach.

Rootless on ubuntu

edit /etc/default/grub to enable cgroups v2

GRUB_CMDLINE_LINUX_DEFAULT="quiet cgroup_enable=memory swapaccount=1 systemd.unified_cgroup_hierarchy=1"

sudo update-grub

ensure that podman socket is running

$ systemctl --user status podman.socket
* podman.socket - Podman API Socket
     Loaded: loaded (/usr/lib/systemd/user/podman.socket; disabled; vendor preset: disabled)
     Active: active (listening) since Sat 2020-10-31 19:21:29 CET; 22h ago
   Triggers: * podman.service
       Docs: man:podman-system-service(1)
     Listen: /run/user/1000/podman/podman.sock (Stream)
     CGroup: /user.slice/user-1000.slice/[email protected]/podman.socket

ensure that you have a recent version of crun

$ crun -V
crun version 0.13.227-d38b
commit: d38b8c28fc50a14978a27fa6afc69a55bfdd2c11
spec: 1.0.0
+SYSTEMD +SELINUX +APPARMOR +CAP +SECCOMP +EBPF +YAJL

nomad job run example.nomad

job "example" {
  datacenters = ["dc1"]
  type        = "service"

  group "cache" {
    count = 1
    restart {
      attempts = 2
      interval = "30m"
      delay    = "15s"
      mode     = "fail"
    }
    network {
      port "redis" { to = 6379 }
    }
    task "redis" {
      driver = "podman"

      config {
        image = "redis"
        ports = ["redis"]
      }

      resources {
        cpu    = 500 # 500 MHz
        memory = 256 # 256MB
      }
    }
  }
}

verify podman ps

$ podman ps
CONTAINER ID  IMAGE                           COMMAND       CREATED        STATUS            PORTS                                                 NAMES
2423ae3efa21  docker.io/library/redis:latest  redis-server  7 seconds ago  Up 6 seconds ago  127.0.0.1:21510->6379/tcp, 127.0.0.1:21510->6379/udp  redis-b640480f-4b93-65fd-7bba-c15722886395

Local Development

Requirements

• Vagrant >= 2.2
• VirtualBox >= v6.0

Vagrant Environment Setup

# create the vm
vagrant up

# ssh into the vm
vagrant ssh

Running a Nomad dev agent with the Podman plugin:

# Build the task driver plugin
make dev

# Copy the build nomad-driver-plugin executable to examples/plugins/
cp ./build/nomad-driver-podman examples/plugins/

# Start Nomad
nomad agent -config=examples/nomad/server.hcl 2>&1 > server.log &

# Run the client as sudo
sudo nomad agent -config=examples/nomad/client.hcl 2>&1 > client.log &

# Run a job
nomad job run examples/jobs/redis_ports.nomad

# Verify
nomad job status redis

sudo podman ps

Running the tests:

# Start the Podman server
systemctl --user start podman.socket

# Run the tests
CI=1 ./build/bin/gotestsum --junitfile ./build/test/result.xml -- -timeout=15m . ./api