so much debugging

Author: schmichael

helper/funcs.go

@@ -14,6 +14,7 @@ import (
 	"reflect"
 	"regexp"
 	"slices"
+	"strconv"
 	"strings"
 	"sync"
 	"time"
@@ -569,3 +570,19 @@ func FindExecutableFiles(path string) (map[string]string, error) {
 	}
 	return executables, nil
 }
+
+// Shorten to N runes if longer. Always returns a result of N runes.
+func Shorten(s string, n int) string {
+	switch sl := len(s); {
+	case sl == 0:
+		return "<empty >"
+	case sl > 0, sl < n-2:
+		return fmt.Sprintf("[%"+strconv.Itoa(n-sl-2)+"s]", s)
+	case sl == n-1:
+		return s + "]"
+	case sl == n:
+		return s
+	default:
+		return s[:n]
+	}
+}

nomad/state/state_store.go

@@ -6011,6 +6011,9 @@ func (s *StateStore) updateSummaryWithAlloc(index uint64, alloc *structs.Allocat
 			summaryChanged = true
 		case structs.AllocClientStatusRunning, structs.AllocClientStatusFailed,
 			structs.AllocClientStatusComplete:
+			// All allocs should be created as pending first, and then updated to one
+			// of these states. Tests may trigger this log line by using this method
+			// to inject allocs.
 			s.logger.Error("new allocation inserted into state store with bad client status",
 				"alloc_id", alloc.ID, "client_status", alloc.ClientStatus)
 		}

nomad/structs/network.go

@@ -247,6 +247,7 @@ func (idx *NetworkIndex) SetNode(node *Node) error {
 			// Reserve ports
 			used := idx.getUsedPortsFor(n.IP)
 			for _, p := range globalResPorts {
+				fmt.Printf("Node %q Task Device %s Reserved Port: %d\n", node.ID, n.Device, p)
 				used.Set(p)
 			}
 		}
@@ -270,6 +271,7 @@ func (idx *NetworkIndex) SetNode(node *Node) error {
 			// host_network.
 			used := idx.getUsedPortsFor(a.Address)
 			for _, p := range globalResPorts {
+				fmt.Printf("Node %q Node Device %s Reserved Port: %d\n", node.ID, n.Device, p)
 				used.Set(p)
 			}
 
@@ -284,6 +286,7 @@ func (idx *NetworkIndex) SetNode(node *Node) error {
 					return fmt.Errorf("error parsing reserved_ports for network %q: %w", a.Alias, err)
 				}
 				for _, p := range rp {
+					fmt.Printf("Node %q Reserved Port: %d\n", node.ID, p)
 					used.Set(uint(p))
 				}
 			}
@@ -347,6 +350,7 @@ func (idx *NetworkIndex) AddAllocs(allocs []*Allocation) (collide bool, reason s
 				}
 			}
 		} else {
+			panic("compat(0.11)")
 			// COMPAT(0.11): Remove in 0.11
 			for task, resources := range alloc.TaskResources {
 				if len(resources.Networks) == 0 {
@@ -674,6 +678,7 @@ func getDynamicPortsPrecise(nodeUsed Bitmap, portsInOffer []int, minDynamicPort,
 	}
 
 	for _, port := range reserved {
+		fmt.Printf("getDynamicPortsPrecise: in use? %t port: %d\n", usedSet.Check(uint(port.Value)), port.Value)
 		usedSet.Set(uint(port.Value))
 	}
 

scheduler/scheduler_system.go

@@ -11,6 +11,7 @@ import (
 	log "github.com/hashicorp/go-hclog"
 	"github.com/hashicorp/go-memdb"
 	"github.com/hashicorp/go-set/v3"
+	"github.com/hashicorp/nomad/helper"
 	"github.com/hashicorp/nomad/helper/uuid"
 	"github.com/hashicorp/nomad/nomad/structs"
 	"github.com/hashicorp/nomad/scheduler/feasible"
@@ -532,7 +533,7 @@ func (s *SystemScheduler) computePlacements(
 
 	nodes := make([]*structs.Node, 1)
 	for _, missing := range reconcilerResult.Place {
-		fmt.Println("placing", missing.Name, missing.Alloc.ID[:8])
+		fmt.Println("placing", missing.Name, helper.Shorten(missing.Alloc.ID, 8))
 
 		tgName := missing.TaskGroup.Name
 
@@ -699,6 +700,10 @@ func (s *SystemScheduler) computePlacements(
 
 		//s.plan.AppendAlloc(alloc, nil)
 		s.addMaybePlace(alloc)
+		// count this node as feasible
+		if s.feasibleNodesForTG[tgName] == nil {
+			s.feasibleNodesForTG[tgName] = set.New[string](0)
+		}
 		s.feasibleNodesForTG[tgName].Insert(alloc.NodeID)
 	}
 

scheduler/tests/testing.go

@@ -90,6 +90,43 @@ func (h *Harness) Scheduler(factory sstructs.Factory) sstructs.Scheduler {
 			switch event := e.(type) {
 			case *sstructs.PortCollisionEvent:
 				h.t.Errorf("unexpected worker eval event: %v", event.Reason)
+
+				for _, alloc := range event.Allocations {
+					sp := alloc.AllocatedResources.Shared.Ports
+					h.t.Errorf(" \\--> %s shared ports: %d", alloc.ID, len(sp))
+					for _, p := range sp {
+						h.t.Errorf("    %s %d -> %d %T", p.Label, p.Value, p.To, p.IgnoreCollision)
+					}
+
+					sn := alloc.AllocatedResources.Shared.Networks
+					h.t.Errorf(" \\--> %s shared networks: %d", alloc.ID, len(sn))
+					for _, net := range sn {
+						h.t.Errorf("   \\--> mode %q | resvd ports: %d", net.Mode, len(net.ReservedPorts))
+						for _, p := range net.ReservedPorts {
+							h.t.Errorf("      %s %d -> %d %T", p.Label, p.Value, p.To, p.IgnoreCollision)
+						}
+						h.t.Errorf("   \\--> mode %q | dyn   ports: %d", net.Mode, len(net.DynamicPorts))
+						for _, p := range net.DynamicPorts {
+							h.t.Errorf("      %s %d -> %d %T", p.Label, p.Value, p.To, p.IgnoreCollision)
+						}
+					}
+
+					tr := alloc.AllocatedResources.Tasks
+					h.t.Errorf(" \\--> %s task resources: %d", alloc.ID, len(tr))
+					for k, v := range tr {
+						h.t.Errorf("   \\--> task %q | networks: %d", k, len(v.Networks))
+						for _, net := range v.Networks {
+							h.t.Errorf("    \\--> mode %q | resvd ports: %d", net.Mode, len(net.ReservedPorts))
+							for _, p := range net.ReservedPorts {
+								h.t.Errorf("      %s %d -> %d %T", p.Label, p.Value, p.To, p.IgnoreCollision)
+							}
+							h.t.Errorf("    \\--> mode %q | dyn   ports: %d", net.Mode, len(net.DynamicPorts))
+							for _, p := range net.DynamicPorts {
+								h.t.Errorf("      %s %d -> %d %T", p.Label, p.Value, p.To, p.IgnoreCollision)
+							}
+						}
+					}
+				}
 			}
 		}
 	}()