|
| 1 | +package api |
| 2 | + |
| 3 | +import ( |
| 4 | + "context" |
| 5 | + "errors" |
| 6 | + "time" |
| 7 | + |
| 8 | + "github.com/kernel/hypeman/lib/guestmemory" |
| 9 | + "github.com/kernel/hypeman/lib/logger" |
| 10 | + "github.com/kernel/hypeman/lib/oapi" |
| 11 | + "go.opentelemetry.io/otel" |
| 12 | + "go.opentelemetry.io/otel/attribute" |
| 13 | + "go.opentelemetry.io/otel/codes" |
| 14 | + "go.opentelemetry.io/otel/trace" |
| 15 | +) |
| 16 | + |
| 17 | +const ( |
| 18 | + defaultMemoryReclaimHold = 5 * time.Minute |
| 19 | + maxMemoryReclaimHold = 1 * time.Hour |
| 20 | +) |
| 21 | + |
| 22 | +// ReclaimMemory triggers proactive guest memory reclaim via runtime ballooning. |
| 23 | +func (s *ApiService) ReclaimMemory(ctx context.Context, request oapi.ReclaimMemoryRequestObject) (oapi.ReclaimMemoryResponseObject, error) { |
| 24 | + log := logger.FromContext(ctx) |
| 25 | + if request.Body == nil { |
| 26 | + return oapi.ReclaimMemory400JSONResponse{ |
| 27 | + Code: "bad_request", |
| 28 | + Message: "request body is required", |
| 29 | + }, nil |
| 30 | + } |
| 31 | + if s.GuestMemoryController == nil { |
| 32 | + return oapi.ReclaimMemory500JSONResponse{ |
| 33 | + Code: "internal_error", |
| 34 | + Message: "guest memory controller not initialized", |
| 35 | + }, nil |
| 36 | + } |
| 37 | + |
| 38 | + holdFor, err := parseMemoryReclaimHold(request.Body) |
| 39 | + if err != nil { |
| 40 | + return oapi.ReclaimMemory400JSONResponse{ |
| 41 | + Code: "bad_request", |
| 42 | + Message: err.Error(), |
| 43 | + }, nil |
| 44 | + } |
| 45 | + |
| 46 | + tracer := otel.Tracer("hypeman/guestmemory") |
| 47 | + ctx, span := tracer.Start(ctx, "guestmemory.manual_reclaim", |
| 48 | + traceAttrsForManualReclaim(request.Body.ReclaimBytes, holdFor, request.Body.DryRun != nil && *request.Body.DryRun, request.Body.Reason != nil)) |
| 49 | + defer span.End() |
| 50 | + |
| 51 | + log.InfoContext(ctx, |
| 52 | + "manual guest memory reclaim requested", |
| 53 | + "operation", "manual_reclaim", |
| 54 | + "requested_reclaim_bytes", request.Body.ReclaimBytes, |
| 55 | + "hold_for_seconds", holdFor.Seconds(), |
| 56 | + "dry_run", request.Body.DryRun != nil && *request.Body.DryRun, |
| 57 | + "reason_present", request.Body.Reason != nil, |
| 58 | + ) |
| 59 | + |
| 60 | + resp, err := s.GuestMemoryController.TriggerReclaim(ctx, guestmemory.ManualReclaimRequest{ |
| 61 | + ReclaimBytes: request.Body.ReclaimBytes, |
| 62 | + HoldFor: holdFor, |
| 63 | + DryRun: request.Body.DryRun != nil && *request.Body.DryRun, |
| 64 | + Reason: derefString(request.Body.Reason), |
| 65 | + }) |
| 66 | + if err != nil { |
| 67 | + switch { |
| 68 | + case errors.Is(err, guestmemory.ErrGuestMemoryDisabled), errors.Is(err, guestmemory.ErrActiveBallooningDisabled): |
| 69 | + span.RecordError(err) |
| 70 | + span.SetStatus(codes.Error, err.Error()) |
| 71 | + log.WarnContext(ctx, "manual guest memory reclaim rejected", "operation", "manual_reclaim", "error", err) |
| 72 | + return oapi.ReclaimMemory400JSONResponse{ |
| 73 | + Code: "feature_disabled", |
| 74 | + Message: err.Error(), |
| 75 | + }, nil |
| 76 | + default: |
| 77 | + span.RecordError(err) |
| 78 | + span.SetStatus(codes.Error, err.Error()) |
| 79 | + log.ErrorContext(ctx, "manual guest memory reclaim failed", "operation", "manual_reclaim", "error", err) |
| 80 | + return oapi.ReclaimMemory500JSONResponse{ |
| 81 | + Code: "internal_error", |
| 82 | + Message: err.Error(), |
| 83 | + }, nil |
| 84 | + } |
| 85 | + } |
| 86 | + |
| 87 | + span.SetAttributes( |
| 88 | + attribute.Int64("planned_reclaim_bytes", resp.PlannedReclaimBytes), |
| 89 | + attribute.Int64("applied_reclaim_bytes", resp.AppliedReclaimBytes), |
| 90 | + attribute.Int64("host_available_bytes", resp.HostAvailableBytes), |
| 91 | + attribute.String("host_pressure_state", string(resp.HostPressureState)), |
| 92 | + attribute.Int("action_count", len(resp.Actions)), |
| 93 | + ) |
| 94 | + span.SetStatus(codes.Ok, "") |
| 95 | + log.InfoContext(ctx, |
| 96 | + "manual guest memory reclaim completed", |
| 97 | + "operation", "manual_reclaim", |
| 98 | + "planned_reclaim_bytes", resp.PlannedReclaimBytes, |
| 99 | + "applied_reclaim_bytes", resp.AppliedReclaimBytes, |
| 100 | + "host_available_bytes", resp.HostAvailableBytes, |
| 101 | + "host_pressure_state", resp.HostPressureState, |
| 102 | + "action_count", len(resp.Actions), |
| 103 | + ) |
| 104 | + |
| 105 | + return oapi.ReclaimMemory200JSONResponse(memoryReclaimResponseToOAPI(resp)), nil |
| 106 | +} |
| 107 | + |
| 108 | +func traceAttrsForManualReclaim(reclaimBytes int64, holdFor time.Duration, dryRun bool, reasonPresent bool) trace.SpanStartOption { |
| 109 | + return trace.WithAttributes( |
| 110 | + attribute.Int64("requested_reclaim_bytes", reclaimBytes), |
| 111 | + attribute.Float64("hold_for_seconds", holdFor.Seconds()), |
| 112 | + attribute.Bool("dry_run", dryRun), |
| 113 | + attribute.Bool("reason_present", reasonPresent), |
| 114 | + ) |
| 115 | +} |
| 116 | + |
| 117 | +func parseMemoryReclaimHold(req *oapi.MemoryReclaimRequest) (time.Duration, error) { |
| 118 | + if req == nil { |
| 119 | + return 0, nil |
| 120 | + } |
| 121 | + |
| 122 | + if req.HoldFor == nil { |
| 123 | + if req.ReclaimBytes > 0 { |
| 124 | + return defaultMemoryReclaimHold, nil |
| 125 | + } |
| 126 | + return 0, nil |
| 127 | + } |
| 128 | + |
| 129 | + holdFor, err := time.ParseDuration(*req.HoldFor) |
| 130 | + if err != nil { |
| 131 | + return 0, errors.New("hold_for must be a valid duration") |
| 132 | + } |
| 133 | + if holdFor < 0 { |
| 134 | + return 0, errors.New("hold_for must be non-negative") |
| 135 | + } |
| 136 | + if holdFor > maxMemoryReclaimHold { |
| 137 | + return 0, errors.New("hold_for must be less than or equal to 1h") |
| 138 | + } |
| 139 | + return holdFor, nil |
| 140 | +} |
| 141 | + |
| 142 | +func memoryReclaimResponseToOAPI(resp guestmemory.ManualReclaimResponse) oapi.MemoryReclaimResponse { |
| 143 | + out := oapi.MemoryReclaimResponse{ |
| 144 | + RequestedReclaimBytes: resp.RequestedReclaimBytes, |
| 145 | + PlannedReclaimBytes: resp.PlannedReclaimBytes, |
| 146 | + AppliedReclaimBytes: resp.AppliedReclaimBytes, |
| 147 | + HoldUntil: resp.HoldUntil, |
| 148 | + HostAvailableBytes: resp.HostAvailableBytes, |
| 149 | + HostPressureState: oapi.MemoryReclaimResponseHostPressureState(resp.HostPressureState), |
| 150 | + Actions: make([]oapi.MemoryReclaimAction, 0, len(resp.Actions)), |
| 151 | + } |
| 152 | + |
| 153 | + for _, action := range resp.Actions { |
| 154 | + item := oapi.MemoryReclaimAction{ |
| 155 | + InstanceId: action.InstanceID, |
| 156 | + InstanceName: action.InstanceName, |
| 157 | + Hypervisor: oapi.MemoryReclaimActionHypervisor(action.Hypervisor), |
| 158 | + AssignedMemoryBytes: action.AssignedMemoryBytes, |
| 159 | + ProtectedFloorBytes: action.ProtectedFloorBytes, |
| 160 | + PreviousTargetGuestMemoryBytes: action.PreviousTargetGuestMemoryBytes, |
| 161 | + PlannedTargetGuestMemoryBytes: action.PlannedTargetGuestMemoryBytes, |
| 162 | + TargetGuestMemoryBytes: action.TargetGuestMemoryBytes, |
| 163 | + AppliedReclaimBytes: action.AppliedReclaimBytes, |
| 164 | + Status: action.Status, |
| 165 | + } |
| 166 | + if action.Error != "" { |
| 167 | + item.Error = &action.Error |
| 168 | + } |
| 169 | + out.Actions = append(out.Actions, item) |
| 170 | + } |
| 171 | + |
| 172 | + return out |
| 173 | +} |
| 174 | + |
| 175 | +func derefString(v *string) string { |
| 176 | + if v == nil { |
| 177 | + return "" |
| 178 | + } |
| 179 | + return *v |
| 180 | +} |
0 commit comments