| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "sync" |
| 7 | |
| 8 | "github.com/carlosframework/platform/internal/activator" |
| 9 | "github.com/carlosframework/platform/internal/hostbox" |
| 10 | "github.com/carlosframework/platform/internal/hostclaim" |
| 11 | "github.com/carlosframework/platform/internal/hostguard" |
| 12 | "github.com/carlosframework/platform/internal/hostwrite" |
| 13 | "github.com/carlosframework/platform/internal/lbhealth" |
| 14 | "github.com/carlosframework/platform/internal/provision" |
| 15 | "github.com/carlosframework/platform/internal/schedule" |
| 16 | ) |
| 17 | |
| 18 | // ClaimProviders contains deployment providers after identity, floor and |
| 19 | // quarantine verification. Build must fail if any of those gates is unknown. |
| 20 | type RegionalLoop interface { |
| 21 | Loop(context.Context, <-chan struct{}, func(error)) error |
| 22 | } |
| 23 | |
| 24 | type ClaimProviders struct { |
| 25 | guard hostguard.Guard |
| 26 | registerLanding func(context.Context, hostclaim.Observed) error |
| 27 | moveTargetFence func(context.Context, hostclaim.Observed) (hostclaim.ReturnFence, error) |
| 28 | moveParks *claimMoveParks |
| 29 | Delivery schedule.DeliveryAdmission |
| 30 | Selector hostwrite.Selector |
| 31 | Wake *activator.ClaimRuntime |
| 32 | Regional RegionalLoop |
| 33 | Stop *ClaimStopCoordinator |
| 34 | Provision func(*provision.ReconcileConfig) |
| 35 | Close func(context.Context) error |
| 36 | } |
| 37 | type ClaimFactory func(context.Context) (ClaimProviders, error) |
| 38 | |
| 39 | type ClaimWiring struct { |
| 40 | Writes hostwrite.Config |
| 41 | Providers ClaimProviders |
| 42 | cancel context.CancelFunc |
| 43 | joined chan struct{} |
| 44 | once sync.Once |
| 45 | } |
| 46 | |
| 47 | func newClaimWiring(ctx context.Context, getenv func(string) string, build ClaimFactory) (*ClaimWiring, error) { |
| 48 | w := &ClaimWiring{} |
| 49 | cfg, err := hostbox.Startup(ctx, getenv, func(ctx context.Context) (hostwrite.Selector, error) { |
| 50 | if build == nil { |
| 51 | return nil, hostwrite.ErrUnavailable |
| 52 | } |
| 53 | providers, err := build(ctx) |
| 54 | if err != nil { |
| 55 | return nil, err |
| 56 | } |
| 57 | if providers.Selector == nil || providers.Wake == nil || providers.Regional == nil || providers.Stop == nil { |
| 58 | return nil, hostwrite.ErrUnavailable |
| 59 | } |
| 60 | w.Providers = providers |
| 61 | return providers.Selector, nil |
| 62 | }) |
| 63 | w.Writes = cfg |
| 64 | return w, err |
| 65 | } |
| 66 | |
| 67 | func (w *ClaimWiring) configure(cfg *activator.Config, runner *schedule.Runner, converger *schedule.Converger) error { |
| 68 | if w == nil || cfg == nil || runner == nil || converger == nil { |
| 69 | return hostwrite.ErrUnavailable |
| 70 | } |
| 71 | if !w.Writes.Enabled { |
| 72 | return nil |
| 73 | } |
| 74 | if w.Writes.Selector == nil || w.Providers.Wake == nil { |
| 75 | return hostwrite.ErrUnavailable |
| 76 | } |
| 77 | cfg.HostWrites = w.Writes |
| 78 | cfg.Claims = w.Providers.Wake |
| 79 | runner.HostWrites = w.Writes |
| 80 | runner.StartDelivery = w.Providers.Delivery |
| 81 | converger.HostWrites = w.Writes |
| 82 | return nil |
| 83 | } |
| 84 | func (w *ClaimWiring) startRegional(ctx context.Context, serial <-chan struct{}, report func(error)) error { |
| 85 | if w == nil { |
| 86 | return hostwrite.ErrUnavailable |
| 87 | } |
| 88 | if !w.Writes.Enabled { |
| 89 | return nil |
| 90 | } |
| 91 | if w.Providers.Regional == nil || w.joined != nil { |
| 92 | return hostwrite.ErrUnavailable |
| 93 | } |
| 94 | workerCtx, cancel := context.WithCancel(ctx) |
| 95 | w.cancel = cancel |
| 96 | w.joined = make(chan struct{}) |
| 97 | // The load balancer's hosting condition (region HA spec 2 §5, §6) is |
| 98 | // this process's mode, as RunAgent's gate chose it (agentHosting): |
| 99 | // the claim agent of an enrolled box with an armed guard. It is set |
| 100 | // before the worker starts and cleared when the worker ends, so a box |
| 101 | // whose claim worker has stopped never answers healthy. |
| 102 | hosting := agentHosting.Load() |
| 103 | lbhealth.Default.SetHosting(hosting) |
| 104 | go func() { |
| 105 | defer close(w.joined) |
| 106 | defer lbhealth.Default.SetHosting(false) |
| 107 | w.Providers.Regional.Loop(workerCtx, serial, report) |
| 108 | }() |
| 109 | return nil |
| 110 | } |
| 111 | func (w *ClaimWiring) close(ctx context.Context) error { |
| 112 | if w == nil { |
| 113 | return nil |
| 114 | } |
| 115 | w.once.Do(func() { |
| 116 | if w.cancel != nil { |
| 117 | w.cancel() |
| 118 | } |
| 119 | }) |
| 120 | if w.joined == nil { |
| 121 | if w.Providers.Close != nil { |
| 122 | return w.Providers.Close(ctx) |
| 123 | } |
| 124 | return nil |
| 125 | } |
| 126 | select { |
| 127 | case <-w.joined: |
| 128 | if w.Providers.Close != nil { |
| 129 | return w.Providers.Close(ctx) |
| 130 | } |
| 131 | return nil |
| 132 | case <-ctx.Done(): |
| 133 | return errors.Join(ctx.Err(), hostwrite.ErrUnavailable) |
| 134 | } |
| 135 | } |
| 136 | |
amadan