Example #1
0
// gcEval returns whether the eval should be garbage collected given a raft
// threshold index. The eval disqualifies for garbage collection if it or its
// allocs are not older than the threshold. If the eval should be garbage
// collected, the associated alloc ids that should also be removed are also
// returned
func (c *CoreScheduler) gcEval(eval *structs.Evaluation, thresholdIndex uint64) (
	bool, []string, error) {
	// Ignore non-terminal and new evaluations
	if !eval.TerminalStatus() || eval.ModifyIndex > thresholdIndex {
		return false, nil, nil
	}

	// Get the allocations by eval
	allocs, err := c.snap.AllocsByEval(eval.ID)
	if err != nil {
		c.srv.logger.Printf("[ERR] sched.core: failed to get allocs for eval %s: %v",
			eval.ID, err)
		return false, nil, err
	}

	// Scan the allocations to ensure they are terminal and old
	for _, alloc := range allocs {
		if !alloc.TerminalStatus() || alloc.ModifyIndex > thresholdIndex {
			return false, nil, nil
		}
	}

	allocIds := make([]string, len(allocs))
	for i, alloc := range allocs {
		allocIds[i] = alloc.ID
	}

	// Evaluation is eligible for garbage collection
	return true, allocIds, nil
}
Example #2
0
// gcEval returns whether the eval should be garbage collected given a raft
// threshold index. The eval disqualifies for garbage collection if it or its
// allocs are not older than the threshold. If the eval should be garbage
// collected, the associated alloc ids that should also be removed are also
// returned
func (c *CoreScheduler) gcEval(eval *structs.Evaluation, thresholdIndex uint64, allowBatch bool) (
	bool, []string, error) {
	// Ignore non-terminal and new evaluations
	if !eval.TerminalStatus() || eval.ModifyIndex > thresholdIndex {
		return false, nil, nil
	}

	// If the eval is from a running "batch" job we don't want to garbage
	// collect its allocations. If there is a long running batch job and its
	// terminal allocations get GC'd the scheduler would re-run the
	// allocations.
	if eval.Type == structs.JobTypeBatch {
		if !allowBatch {
			return false, nil, nil
		}

		// Check if the job is running
		job, err := c.snap.JobByID(eval.JobID)
		if err != nil {
			return false, nil, err
		}

		// We don't want to gc anything related to a job which is not dead
		if job != nil && job.Status != structs.JobStatusDead {
			return false, nil, nil
		}
	}

	// Get the allocations by eval
	allocs, err := c.snap.AllocsByEval(eval.ID)
	if err != nil {
		c.srv.logger.Printf("[ERR] sched.core: failed to get allocs for eval %s: %v",
			eval.ID, err)
		return false, nil, err
	}

	// Scan the allocations to ensure they are terminal and old
	gcEval := true
	var gcAllocIDs []string
	for _, alloc := range allocs {
		if !alloc.TerminalStatus() || alloc.ModifyIndex > thresholdIndex {
			// Can't GC the evaluation since not all of the allocations are
			// terminal
			gcEval = false
		} else {
			// The allocation is eligible to be GC'd
			gcAllocIDs = append(gcAllocIDs, alloc.ID)
		}
	}

	return gcEval, gcAllocIDs, nil
}
Example #3
0
// evalGC is used to garbage collect old evaluations
func (c *CoreScheduler) evalGC(eval *structs.Evaluation) error {
	// Iterate over the evaluations
	iter, err := c.snap.Evals()
	if err != nil {
		return err
	}

	// Compute the old threshold limit for GC using the FSM
	// time table.  This is a rough mapping of a time to the
	// Raft index it belongs to.
	tt := c.srv.fsm.TimeTable()
	cutoff := time.Now().UTC().Add(-1 * c.srv.config.EvalGCThreshold)
	oldThreshold := tt.NearestIndex(cutoff)
	c.srv.logger.Printf("[DEBUG] sched.core: eval GC: scanning before index %d (%v)",
		oldThreshold, c.srv.config.EvalGCThreshold)

	// Collect the allocations and evaluations to GC
	var gcAlloc, gcEval []string

OUTER:
	for {
		raw := iter.Next()
		if raw == nil {
			break
		}
		eval := raw.(*structs.Evaluation)

		// Ignore non-terminal and new evaluations
		if !eval.TerminalStatus() || eval.ModifyIndex > oldThreshold {
			continue
		}

		// Get the allocations by eval
		allocs, err := c.snap.AllocsByEval(eval.ID)
		if err != nil {
			c.srv.logger.Printf("[ERR] sched.core: failed to get allocs for eval %s: %v",
				eval.ID, err)
			continue
		}

		// Scan the allocations to ensure they are terminal and old
		for _, alloc := range allocs {
			if !alloc.TerminalStatus() || alloc.ModifyIndex > oldThreshold {
				continue OUTER
			}
		}

		// Evaluation is eligible for garbage collection
		gcEval = append(gcEval, eval.ID)
		for _, alloc := range allocs {
			gcAlloc = append(gcAlloc, alloc.ID)
		}
	}

	// Fast-path the nothing case
	if len(gcEval) == 0 && len(gcAlloc) == 0 {
		return nil
	}
	c.srv.logger.Printf("[DEBUG] sched.core: eval GC: %d evaluations, %d allocs eligible",
		len(gcEval), len(gcAlloc))

	// Call to the leader to issue the reap
	req := structs.EvalDeleteRequest{
		Evals:  gcEval,
		Allocs: gcAlloc,
		WriteRequest: structs.WriteRequest{
			Region: c.srv.config.Region,
		},
	}
	var resp structs.GenericResponse
	if err := c.srv.RPC("Eval.Reap", &req, &resp); err != nil {
		c.srv.logger.Printf("[ERR] sched.core: eval reap failed: %v", err)
		return err
	}
	return nil
}