| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "errors" |
| 7 | "net" |
| 8 | "net/http" |
| 9 | "sync" |
| 10 | "time" |
| 11 | |
| 12 | "github.com/carlosframework/platform/internal/hostbox" |
| 13 | "github.com/carlosframework/platform/internal/hostclaim" |
| 14 | "github.com/carlosframework/platform/internal/instancectl" |
| 15 | "github.com/carlosframework/platform/internal/rootstate" |
| 16 | ) |
| 17 | |
| 18 | var openClaimServingParkAuthority = func() (rootstate.Store, error) { |
| 19 | return rootstate.Open("/var/lib/carlos-claims/authority/serving-parks", true, false) |
| 20 | } |
| 21 | |
| 22 | type servingParkAuthority interface { |
| 23 | parkServingDatabase(context.Context, net.Conn, servingArtifactIdentity, servingArtifactIdentity) (servingParkAuthorization, error) |
| 24 | parkLegacyServingDatabase(context.Context, net.Conn, servingArtifactIdentity, servingLegacyPredecessor) (servingParkAuthorization, error) |
| 25 | completeServingDatabasePark(context.Context, net.Conn, string, servingArtifactIdentity) error |
| 26 | retainLegacyNoProcess(context.Context, hostclaim.Observed, StopResult) error |
| 27 | } |
| 28 | |
| 29 | type servingEmptyAuthority interface { |
| 30 | EmptyAuthority(context.Context, hostclaim.Observed, ClaimLaunchInputs) error |
| 31 | } |
| 32 | |
| 33 | // Closing cancels and joins every request and stop-evidence retention before |
| 34 | // releasing private state. A timed-out join leaves the store open for retry. |
| 35 | type servingParkRuntime struct { |
| 36 | authority servingParkAuthority |
| 37 | state rootstate.Store |
| 38 | ctx context.Context |
| 39 | cancel context.CancelFunc |
| 40 | mu sync.Mutex |
| 41 | active int |
| 42 | recoveryMu sync.Mutex |
| 43 | recoveryAfter string |
| 44 | lastRecoveryLog time.Time |
| 45 | closing, closed, started bool |
| 46 | idle chan struct{} |
| 47 | } |
| 48 | |
| 49 | func withServingParkRoot(h ClaimRootHTTP, empty servingEmptyAuthority) (ClaimRootHTTP, error) { |
| 50 | state, err := openClaimServingParkAuthority() |
| 51 | if err != nil { |
| 52 | return ClaimRootHTTP{}, err |
| 53 | } |
| 54 | runtime, err := newServingParkRuntime(h, state, empty) |
| 55 | if err != nil { |
| 56 | state.Close() |
| 57 | return ClaimRootHTTP{}, err |
| 58 | } |
| 59 | h.servingParks = runtime |
| 60 | h.Stop.Coordinator.RetainNoProcess = runtime.retainNoProcess |
| 61 | return h, nil |
| 62 | } |
| 63 | |
| 64 | func newServingParkRuntime(h ClaimRootHTTP, state rootstate.Store, empty servingEmptyAuthority) (*servingParkRuntime, error) { |
| 65 | authority, err := newServingParkAuthority(h, state, empty) |
| 66 | if err != nil { |
| 67 | return nil, err |
| 68 | } |
| 69 | ctx, cancel := context.WithCancel(context.Background()) |
| 70 | return &servingParkRuntime{authority: authority, state: state, ctx: ctx, cancel: cancel, idle: make(chan struct{})}, nil |
| 71 | } |
| 72 | |
| 73 | func (p *servingParkRuntime) begin(ctx context.Context) (context.Context, func(), error) { |
| 74 | if p == nil || ctx == nil { |
| 75 | return nil, nil, hostbox.ErrUnsupported |
| 76 | } |
| 77 | p.mu.Lock() |
| 78 | defer p.mu.Unlock() |
| 79 | if p.closing { |
| 80 | return nil, nil, hostbox.ErrUnsupported |
| 81 | } |
| 82 | p.active++ |
| 83 | call, cancel := context.WithCancel(ctx) |
| 84 | stop := context.AfterFunc(p.ctx, cancel) |
| 85 | return call, func() { |
| 86 | stop() |
| 87 | cancel() |
| 88 | p.mu.Lock() |
| 89 | defer p.mu.Unlock() |
| 90 | p.active-- |
| 91 | if p.closing && p.active == 0 { |
| 92 | close(p.idle) |
| 93 | } |
| 94 | }, nil |
| 95 | } |
| 96 | |
| 97 | func (p *servingParkRuntime) Close(ctx context.Context) error { |
| 98 | if p == nil { |
| 99 | return nil |
| 100 | } |
| 101 | p.mu.Lock() |
| 102 | if !p.closing { |
| 103 | p.closing = true |
| 104 | p.cancel() |
| 105 | if p.active == 0 { |
| 106 | close(p.idle) |
| 107 | } |
| 108 | } |
| 109 | p.mu.Unlock() |
| 110 | select { |
| 111 | case <-p.idle: |
| 112 | case <-ctx.Done(): |
| 113 | select { |
| 114 | case <-p.idle: |
| 115 | default: |
| 116 | return ctx.Err() |
| 117 | } |
| 118 | } |
| 119 | p.mu.Lock() |
| 120 | defer p.mu.Unlock() |
| 121 | if p.closed { |
| 122 | return nil |
| 123 | } |
| 124 | p.closed = true |
| 125 | return p.state.Close() |
| 126 | } |
| 127 | |
| 128 | func (p *servingParkRuntime) retainNoProcess(ctx context.Context, o hostclaim.Observed, result StopResult) error { |
| 129 | call, done, err := p.begin(ctx) |
| 130 | if err != nil { |
| 131 | return err |
| 132 | } |
| 133 | defer done() |
| 134 | return p.authority.retainLegacyNoProcess(call, o, result) |
| 135 | } |
| 136 | |
| 137 | func (h ClaimRootHTTP) serveServingPark(w http.ResponseWriter, r *http.Request, q hostbox.JournalRequest) { |
| 138 | out := hostbox.JournalResponse{V: 1, ID: q.ID} |
| 139 | ctx, done, err := h.servingParks.begin(r.Context()) |
| 140 | if err != nil { |
| 141 | w.WriteHeader(http.StatusServiceUnavailable) |
| 142 | return |
| 143 | } |
| 144 | defer done() |
| 145 | conn := instancectl.RequestConn(ctx) |
| 146 | if conn == nil || h.RequireNewWork == nil || h.InventoryActor == nil || h.Journals == nil || q.ServingPark == nil { |
| 147 | w.WriteHeader(http.StatusForbidden) |
| 148 | return |
| 149 | } |
| 150 | // This gate also covers receipt-only replay. The outer HTTP handler has |
| 151 | // authenticated the live edge main process; effects authenticate it again |
| 152 | // after acquiring root's namespace and host locks. |
| 153 | if err = h.RequireNewWork(ctx); err != nil { |
| 154 | w.WriteHeader(http.StatusForbidden) |
| 155 | return |
| 156 | } |
| 157 | actor, err := h.InventoryActor(ctx) |
| 158 | if err != nil || !actor.Eligible || actor.Quarantined || actor.Box == "" || actor.Enrollment == "" || actor.Revision == 0 || actor.Binding == "" || q.ServingPark.Successor.Box != actor.Box || q.ServingPark.Successor.Region != actor.Region { |
| 159 | w.WriteHeader(http.StatusForbidden) |
| 160 | return |
| 161 | } |
| 162 | out, err = h.Journals.ExecuteServingPark(ctx, actor.Box+"/"+actor.Enrollment, q, func(ctx context.Context) (hostbox.ServingParkResult, error) { |
| 163 | p, a := *q.ServingPark, h.servingParks.authority |
| 164 | switch q.Op { |
| 165 | case "serving-park": |
| 166 | return a.parkServingDatabase(ctx, conn, p.Successor, p.Predecessor) |
| 167 | case "serving-park-legacy": |
| 168 | return a.parkLegacyServingDatabase(ctx, conn, p.Successor, p.Legacy) |
| 169 | case "serving-park-complete": |
| 170 | err := a.completeServingDatabasePark(ctx, conn, p.Retirement, p.Successor) |
| 171 | return hostbox.ServingParkResult{ID: p.Retirement, Successor: p.Successor}, err |
| 172 | default: |
| 173 | return hostbox.ServingParkResult{}, hostbox.ErrUnsupported |
| 174 | } |
| 175 | }) |
| 176 | if err != nil { |
| 177 | out.ServingPark = nil |
| 178 | switch { |
| 179 | case errors.Is(err, hostbox.ErrJournalUncertain): |
| 180 | out.Error = "uncertain" |
| 181 | case errors.Is(err, hostbox.ErrChanged), errors.Is(err, hostbox.ErrTypedJournal): |
| 182 | out.Error = "refused" |
| 183 | case errors.Is(err, hostbox.ErrUnsupported): |
| 184 | out.Error = "unsupported" |
| 185 | default: |
| 186 | out.Error = "uncertain" |
| 187 | } |
| 188 | } |
| 189 | w.Header().Set("Content-Type", "application/json") |
| 190 | _ = json.NewEncoder(w).Encode(out) |
| 191 | } |
| 192 | |
amadan