TaskConfig: Add option for file_limit to set RLIMIT_NOFILE

Depending on the workload, the default resource limit (ulimit) for the
max number of file descriptors per process may need to be raised, such
as for database servers. Add a `file_limit` option to the task config
which will allow the limit to be set per task.

Signed-off-by: Ryan Sundberg <[email protected]>

Author: sundbry

README.md

@@ -112,6 +112,7 @@ To interact with `images` and `containers` directly, you can use [`nerdctl`](htt
 | **privileged** | bool | no | Run container in privileged mode. Your container will have all linux capabilities when running in privileged mode. |
 | **pids_limit** | int64 | no | An integer value that specifies the pid limit for the container. Defaults to unlimited. |
 | **pid_mode** | string | no | `host` or not set (default). Set to `host` to share the PID namespace with the host. |
+| **file_limit** | int64 | no | An integer value that specifies the file descriptor ulimit for the container. Defaults to 1024 by containerd. |
 | **hostname** | string | no | 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. |
 | **host_dns** | bool | no | Default (`true`). By default, a container launched using `containerd-driver` will use host `/etc/resolv.conf`. This is similar to [`docker behavior`](https://docs.docker.com/config/containers/container-networking/#dns-services). However, if you don't want to use host DNS, you can turn off this flag by setting `host_dns=false`. |
 | **seccomp** | bool | no | Enable default seccomp profile. List of [`allowed syscalls`](https://github.com/containerd/containerd/blob/master/contrib/seccomp/seccomp_default.go#L51-L395). |

containerd/containerd.go

@@ -25,6 +25,7 @@ import (
 
 	etchosts "github.com/Roblox/nomad-driver-containerd/etchosts"
 	"github.com/containerd/containerd"
+	"github.com/containerd/containerd/containers"
 	"github.com/containerd/containerd/cio"
 	"github.com/containerd/containerd/contrib/seccomp"
 	"github.com/containerd/containerd/oci"
@@ -93,6 +94,27 @@ func withResolver(creds CredentialsOpt) containerd.RemoteOpt {
 	return containerd.WithResolver(resolver)
 }
 
+func withFileLimit(maxOpenFiles uint64) oci.SpecOpts {
+	return func(_ context.Context, _ oci.Client, _ *containers.Container, spec *oci.Spec) error {
+		newRlimits := []specs.POSIXRlimit{{
+			Type: "RLIMIT_NOFILE",
+			Hard: maxOpenFiles,
+			Soft: maxOpenFiles,
+		}}
+
+		// Copy existing rlimits excluding previous RLIMIT_NOFILE
+		for _, rlimit := range spec.Process.Rlimits {
+			if rlimit.Type != "RLIMIT_NOFILE" {
+				newRlimits = append(newRlimits, rlimit)
+			}
+		}
+
+		spec.Process.Rlimits = newRlimits
+
+		return nil
+	}
+}
+
 func (d *Driver) pullImage(imageName, imagePullTimeout string, auth *RegistryAuth) (containerd.Image, error) {
 	pullTimeout, err := time.ParseDuration(imagePullTimeout)
 	if err != nil {
@@ -167,6 +189,11 @@ func (d *Driver) createContainer(containerConfig *ContainerConfig, config *TaskC
 		}
 	}
 
+	// Set the resource limit for open file descriptors
+	if config.FileLimit > 0 {
+		opts = append(opts, withFileLimit(uint64(config.FileLimit)))
+	}
+
 	// Size of /dev/shm
 	if len(config.ShmSize) > 0 {
 		shmBytes, err := units.RAMInBytes(config.ShmSize)

containerd/driver.go

@@ -106,6 +106,7 @@ var (
 		"privileged": hclspec.NewAttr("privileged", "bool", false),
 		"pids_limit": hclspec.NewAttr("pids_limit", "number", false),
 		"pid_mode":   hclspec.NewAttr("pid_mode", "string", false),
+		"file_limit": hclspec.NewAttr("file_limit", "number", false),
 		"hostname":   hclspec.NewAttr("hostname", "string", false),
 		"host_dns": hclspec.NewDefault(
 			hclspec.NewAttr("host_dns", "bool", false),
@@ -190,6 +191,7 @@ type TaskConfig struct {
 	Privileged       bool               `codec:"privileged"`
 	PidsLimit        int64              `codec:"pids_limit"`
 	PidMode          string             `codec:"pid_mode"`
+	FileLimit        int64              `codec:"file_limit"`
 	Hostname         string             `codec:"hostname"`
 	HostDNS          bool               `codec:"host_dns"`
 	ImagePullTimeout string             `codec:"image_pull_timeout"`