Documentation
¶
Index ¶
- Constants
- Variables
- func AddUserToContext(ctx context.Context, user string) context.Context
- func GetUserFromContext(ctx context.Context) (string, bool)
- func InitCgroups() error
- type ArgMaker
- type DiskIOLimits
- type Job
- type JobDescription
- type JobSpec
- type JobState
- type JobStatus
- type Log
- type ResourceLimits
- type Tracker
- func (t *Tracker) Get(ctx context.Context, id string) (JobDescription, error)
- func (t *Tracker) GetLogChannel(id string, follow bool, ctx context.Context) (<-chan Log, error)
- func (t *Tracker) List(ctx context.Context, completed, all bool) []JobDescription
- func (t *Tracker) Shutdown(ctx context.Context) (int, error)
- func (t *Tracker) Start(ctx context.Context, spec JobSpec) (string, error)
- func (t *Tracker) Stop(ctx context.Context, id string, cleanup bool) error
Constants ¶
const ( JobStatePreStart = iota JobStateRunning JobStateCompleted )
const JobberCG = "/sys/fs/cgroup/jobber"
Variables ¶
var ( ErrMissingID = errors.New("missing job ID") ErrNoCommand = errors.New("missing job command") ErrNotStarted = errors.New("could not start job") ErrShutdown = errors.New("service is shut down") ErrUnknown = errors.New("unknown job") )
var (
ErrAlreadyStarted = errors.New("job already started")
)
Functions ¶
func InitCgroups ¶
func InitCgroups() error
Types ¶
type ArgMaker ¶
type ArgMaker func(JobDescription) (string, []string)
type DiskIOLimits ¶
type DiskIOLimits struct {
Device string
Major uint32
Minor uint32
ReadBPS uint64
WriteBPS uint64
ReadIOPS uint32
WriteIOPS uint32
}
func (*DiskIOLimits) ResolveDevice ¶
func (d *DiskIOLimits) ResolveDevice() error
func (*DiskIOLimits) String ¶
func (d *DiskIOLimits) String() string
func (*DiskIOLimits) UnmarshalText ¶
func (d *DiskIOLimits) UnmarshalText(b []byte) (err error)
UnmarshalText unmarshals a string ([]byte) into a DiskIOLimits. It is used by kong to unmarshal the command line argument into a structured value.
The format of the input string is 5 or 6 colon separated values. If 5 values, the first should be a block device filesystem path that can be stat'ed to get its major and minor number. If 6 values, they are directly the major and minor number.
The remaining 4 values are the disk IO limits for that block device. A field may be empty which is parsed as zero, which means no setting for that throttle.
type Job ¶
type Job struct {
ID string
Spec JobSpec
Status JobStatus
// contains filtered or unexported fields
}
func (*Job) AttachOutfeed ¶
func (*Job) Description ¶
func (j *Job) Description() JobDescription
func (*Job) ExecPart1 ¶
func (j *Job) ExecPart1() (io.ReadCloser, error)
ExecPart1 starts the execution of a job's command, ensuring it runs in new namespaces where appropriate, attaching pipes to capture the output of the command and any errors that come from not being able to run the command. It uses an ArgMaker to construct the command line as we do not know anything about the program we are embedded in and what command line args it takes. The ArgMaker abstracts that for us and allows the user of this package to define how to propagate Job parameters into a Job for ExecPart2 in a child process.
If successful, it returns an io.ReadCloser that can be read for the command's combined stdout/stderr stream. Once that has closed, Job.cmd.Wait() should be called on the job to capture the exit code of the process and reap it.
func (*Job) ExecPart2 ¶
func (j *Job) ExecPart2()
ExecPart2 runs the job in a cgroup configured from the job's parameters and configures the namespaces it is already in. It is expected that the process is already running in "empty" namespaces based on the job's configuration.
It is expected that the standard io streams are set up as follows:
- stdin: /dev/null
- stdout: where the process's stdout and stderr are sent
- stderr: where error messages due to the inability to run the program are sent - e.g. errors setting up the cgroup, being unable to exec the program (not found), etc.
When the command is executed, it will have the stderr stream it received closed and will instead have the stdout stream on stderr too.
It does not return an error, instead writing errors to stderr to be captured by the parent process in ExecPart1().
type JobDescription ¶
type JobSpec ¶
type JobSpec struct {
Command string `arg:"" help:"Command for jobber server to run"`
Args []string `arg:"" optional:"" help:"Arguments to command"`
Root string `help:"run in isolated root directory"`
IsolateNetwork bool `help:"run in isolated network namespace"`
Resources ResourceLimits `embed:""`
}
type ResourceLimits ¶
type ResourceLimits struct {
MaxProcesses uint32 `help:"maximum number of processes"`
Memory uint64 `help:"maximum memory (bytes)"`
CPU uint32 `help:"maximum CPU (milliCPU)"`
IO []DiskIOLimits `name:"io" help:"disk io limits (dev:rbps:wbps:riops:wiops)"`
}
type Tracker ¶
type Tracker struct {
// contains filtered or unexported fields
}
Tracker maintains a set of Jobs that are either running or have completed. Jobs can be added (started), stopped (including removed via cleanup if desired), listed and attached to for log output.
func NewTracker ¶
func (*Tracker) Get ¶
Get returns a copy of the job identified by id if it exists in the tracker, otherwise an error. The copy returned is not an active job that can be manipulated - it is just for the data.
func (*Tracker) GetLogChannel ¶
GetLogChannel returns a channel that streams the logs of the job identified by id. If follow is set, the stream will continue until the job terminates. Regardless of the follow flag, if the context is closed, then the returned log channel is detached from the log feeder and is closed.
func (*Tracker) List ¶
func (t *Tracker) List(ctx context.Context, completed, all bool) []JobDescription
List returns a copy of all the jobs for a owner, or all jobs if the given owner is empty. Only running jobs are returned, unless completed is true.
Source Files