Documentation
¶
Overview ¶
Dispatcher service for Crunch that submits containers to the slurm queue.
Index ¶
Constants ¶
This section is empty.
Variables ¶
Functions ¶
func NewSlurmCLI ¶
func NewSlurmCLI() *slurmCLI
func SlurmNodeTypeFeatureKludge ¶
SlurmNodeTypeFeatureKludge ensures SLURM accepts every instance type name as a valid feature name, even if no instances of that type have appeared yet.
It takes advantage of some SLURM peculiarities:
(1) A feature is valid after it has been offered by a node, even if it is no longer offered by any node. So, to make a feature name valid, we can add it to a dummy node ("compute0"), then remove it.
(2) To test whether a set of feature names are valid without actually submitting a job, we can call srun --test-only with the desired features.
SlurmNodeTypeFeatureKludge does a test-and-fix operation immediately, and then periodically, in case slurm restarts and forgets the list of valid features. It never returns (unless there are no node types configured, in which case it returns immediately), so it should generally be invoked with "go".
Types ¶
type Dispatcher ¶
type Dispatcher struct {
*dispatch.Dispatcher
Client arvados.Client
// contains filtered or unexported fields
}
func (*Dispatcher) CheckHealth ¶
func (disp *Dispatcher) CheckHealth() error
func (*Dispatcher) Done ¶
func (disp *Dispatcher) Done() <-chan struct{}
func (*Dispatcher) ServeHTTP ¶
func (disp *Dispatcher) ServeHTTP(w http.ResponseWriter, r *http.Request)
type Slurm ¶
type SqueueChecker ¶
type SqueueChecker struct {
Logger logger
Period time.Duration
PrioritySpread int64
Slurm Slurm
// contains filtered or unexported fields
}
SqueueChecker implements asynchronous polling monitor of the SLURM queue using the command 'squeue'.
func (*SqueueChecker) All ¶
func (sqc *SqueueChecker) All() []string
All waits for the next squeue invocation, and returns all job names reported by squeue.
func (*SqueueChecker) HasUUID ¶
func (sqc *SqueueChecker) HasUUID(uuid string) bool
HasUUID checks if a given container UUID is in the slurm queue. This does not run squeue directly, but instead blocks until woken up by next successful update of squeue.
func (*SqueueChecker) SetPriority ¶
func (sqc *SqueueChecker) SetPriority(uuid string, want int64)
SetPriority sets or updates the desired (Arvados) priority for a container.
func (*SqueueChecker) Stop ¶
func (sqc *SqueueChecker) Stop()
Stop stops the squeue monitoring goroutine. Do not call HasUUID after calling Stop.
Source Files