| 1 | package instancectl |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "errors" |
| 7 | "fmt" |
| 8 | "io" |
| 9 | "net" |
| 10 | "net/http" |
| 11 | "net/url" |
| 12 | "strconv" |
| 13 | "strings" |
| 14 | "time" |
| 15 | |
| 16 | "github.com/carlosframework/platform/internal/version" |
| 17 | ) |
| 18 | |
| 19 | // Deadlines per verb (design §5.3). A start or stop that exceeds its |
| 20 | // deadline is a failed wake or an unproven stop, never a hang: §6.7 puts |
| 21 | // one such call on a path that must not stall the activator. The seed's is |
| 22 | // long because a seed is allowed to be slow. |
| 23 | const ( |
| 24 | DeadlineControl = 30 * time.Second |
| 25 | DeadlineSeed = 150 * time.Second |
| 26 | DeadlineSweep = 60 * time.Second |
| 27 | DeadlineReady = 5 * time.Second |
| 28 | |
| 29 | // DeadlineCmdPoll bounds GET /cmd/{host}; the server answers within 30 s. |
| 30 | DeadlineCmdPoll = 35 * time.Second |
| 31 | ) |
| 32 | |
| 33 | // Client is the agent's side of the control channel. It is plain HTTP over |
| 34 | // the unix socket at Path; unixdial's checks are for UPSTREAMS the edge |
| 35 | // proxies to, and this socket is the one path they refuse, so the client |
| 36 | // dials it directly. |
| 37 | type Client struct { |
| 38 | Path string |
| 39 | // Dial is the dialer; nil means net.Dialer. Tests do not need to override |
| 40 | // it, but a caller composing several clients may. |
| 41 | Dial func(ctx context.Context, path string) (net.Conn, error) |
| 42 | // Build is the build this agent runs. Ready refuses a service whose |
| 43 | // BuildHeader does not name it (#256); empty skips the comparison, |
| 44 | // for tests that build a Client literal. NewClient sets version.Build. |
| 45 | Build string |
| 46 | } |
| 47 | |
| 48 | func NewClient(path string) *Client { return &Client{Path: path, Build: version.Build} } |
| 49 | |
| 50 | // StaleBuildError is Ready's answer to a control service running a build |
| 51 | // other than the agent's own: the service outlived a binary change and |
| 52 | // serves this release's units with the previous release's code (#256). |
| 53 | type StaleBuildError struct{ Running, Want string } |
| 54 | |
| 55 | func (e *StaleBuildError) Error() string { |
| 56 | if e.Running == "" { |
| 57 | return fmt.Sprintf("instance control: carlos-instance-control.service is running an older build than this agent (%s); restart carlos-instance-control.service", e.Want) |
| 58 | } |
| 59 | return fmt.Sprintf("instance control: carlos-instance-control.service is running build %s, this agent is %s; restart carlos-instance-control.service", e.Running, e.Want) |
| 60 | } |
| 61 | |
| 62 | // NotReadyError carries the probe verdict a 503 /ready returned. |
| 63 | type NotReadyError struct{ Report ProbeReport } |
| 64 | |
| 65 | func (e *NotReadyError) Error() string { |
| 66 | var fails []string |
| 67 | for _, l := range e.Report.Lines { |
| 68 | if l.Verdict != "ok" { |
| 69 | fails = append(fails, l.Check+" "+l.Verdict+" "+l.Detail) |
| 70 | } |
| 71 | } |
| 72 | if len(fails) == 0 { |
| 73 | return "instance control: box not ready (no verdict yet)" |
| 74 | } |
| 75 | return "instance control: box not ready: " + strings.Join(fails, "; ") |
| 76 | } |
| 77 | |
| 78 | // httpClient builds a fresh Transport per call rather than caching one on c. |
| 79 | // DisableKeepAlives is what makes that the right trade: a pooled idle |
| 80 | // connection nothing ever reuses (each call already builds its own client) |
| 81 | // would just sit open, leaking one file descriptor per call instead of |
| 82 | // closing when the response is read. |
| 83 | func (c *Client) httpClient() *http.Client { |
| 84 | dial := c.Dial |
| 85 | if dial == nil { |
| 86 | dial = func(ctx context.Context, path string) (net.Conn, error) { |
| 87 | return (&net.Dialer{}).DialContext(ctx, "unix", path) |
| 88 | } |
| 89 | } |
| 90 | return &http.Client{Transport: &http.Transport{ |
| 91 | DialContext: func(ctx context.Context, _, _ string) (net.Conn, error) { return dial(ctx, c.Path) }, |
| 92 | DisableKeepAlives: true, |
| 93 | }} |
| 94 | } |
| 95 | |
| 96 | // do issues one verb with the given deadline and decodes a JSON body when |
| 97 | // there is one. A non-2xx answer becomes an error carrying the server's |
| 98 | // message; a dial failure names the socket so a forgotten enable reads as |
| 99 | // what it is from carlos logs. |
| 100 | func (c *Client) do(ctx context.Context, deadline time.Duration, method, path string, q url.Values, out any) (int, http.Header, error) { |
| 101 | ctx, cancel := context.WithTimeout(ctx, deadline) |
| 102 | defer cancel() |
| 103 | u := "http://control" + path |
| 104 | if len(q) > 0 { |
| 105 | u += "?" + q.Encode() |
| 106 | } |
| 107 | req, err := http.NewRequestWithContext(ctx, method, u, nil) |
| 108 | if err != nil { |
| 109 | return 0, nil, err |
| 110 | } |
| 111 | resp, err := c.httpClient().Do(req) |
| 112 | if err != nil { |
| 113 | return 0, nil, fmt.Errorf("instance control socket %s: %w", c.Path, err) |
| 114 | } |
| 115 | defer resp.Body.Close() |
| 116 | body, _ := io.ReadAll(io.LimitReader(resp.Body, 64<<10)) |
| 117 | if resp.StatusCode == http.StatusServiceUnavailable && path == "/ready" { |
| 118 | var r ProbeReport |
| 119 | _ = json.Unmarshal(body, &r) |
| 120 | return resp.StatusCode, resp.Header, &NotReadyError{Report: r} |
| 121 | } |
| 122 | // A 502 from /sweep still carries the units it did and didn't manage to |
| 123 | // stop, so it is decoded into out here even though it is about to be |
| 124 | // reported as an error too; the caller needs the partial list alongside |
| 125 | // the failure, not instead of it. |
| 126 | if out != nil && len(body) > 0 && (resp.StatusCode/100 == 2 || (resp.StatusCode == http.StatusBadGateway && path == "/sweep")) { |
| 127 | if err := json.Unmarshal(body, out); err != nil { |
| 128 | return resp.StatusCode, resp.Header, fmt.Errorf("instance control %s: %v", path, err) |
| 129 | } |
| 130 | } |
| 131 | if resp.StatusCode/100 != 2 { |
| 132 | return resp.StatusCode, resp.Header, fmt.Errorf("instance control %s: %d: %s", path, resp.StatusCode, strings.TrimSpace(string(body))) |
| 133 | } |
| 134 | return resp.StatusCode, resp.Header, nil |
| 135 | } |
| 136 | |
| 137 | func hostRole(host string, role Role) url.Values { |
| 138 | q := url.Values{"host": {host}} |
| 139 | if role != "" { |
| 140 | q.Set("role", string(role)) |
| 141 | } |
| 142 | return q |
| 143 | } |
| 144 | |
| 145 | // Start starts the run or sidecar unit for host with the memory tier and |
| 146 | // returns the invocation id the server read after the start. |
| 147 | func (c *Client) Start(ctx context.Context, host string, role Role, memoryMax int64) (string, error) { |
| 148 | q := hostRole(host, role) |
| 149 | q.Set("memory_max", strconv.FormatInt(memoryMax, 10)) |
| 150 | var out StartResponse |
| 151 | if _, _, err := c.do(ctx, DeadlineControl, http.MethodPost, "/start", q, &out); err != nil { |
| 152 | return "", err |
| 153 | } |
| 154 | return out.InvocationID, nil |
| 155 | } |
| 156 | |
| 157 | // Seed runs the seed unit to completion and returns its retained result. |
| 158 | func (c *Client) Seed(ctx context.Context, host string, memoryMax int64) (SeedResponse, error) { |
| 159 | return c.seedQuery(ctx, host, memoryMax, nil) |
| 160 | } |
| 161 | |
| 162 | // SeedExe is Seed with the seed's executable named: the staged build the |
| 163 | // wake recorded (settings spec §4.2), which root validates and binds in |
| 164 | // place of the live link. A 403 is root's refusal, reported as |
| 165 | // ErrStagedExe: nothing was laid out or run. |
| 166 | func (c *Client) SeedExe(ctx context.Context, host string, memoryMax int64, x StagedExe) (SeedResponse, error) { |
| 167 | return c.seedQuery(ctx, host, memoryMax, &x) |
| 168 | } |
| 169 | |
| 170 | func (c *Client) seedQuery(ctx context.Context, host string, memoryMax int64, x *StagedExe) (SeedResponse, error) { |
| 171 | q := hostRole(host, "") |
| 172 | q.Set("memory_max", strconv.FormatInt(memoryMax, 10)) |
| 173 | if x != nil { |
| 174 | setStagedExe(q, *x) |
| 175 | } |
| 176 | var out SeedResponse |
| 177 | status, _, err := c.do(ctx, DeadlineSeed, http.MethodPost, "/seed", q, &out) |
| 178 | if err != nil && status == http.StatusForbidden && x != nil { |
| 179 | return out, fmt.Errorf("%w: %v", ErrStagedExe, err) |
| 180 | } |
| 181 | return out, err |
| 182 | } |
| 183 | |
| 184 | // setStagedExe puts x on a start or seed query. |
| 185 | func setStagedExe(q url.Values, x StagedExe) { |
| 186 | q.Set("exe", x.Path) |
| 187 | q.Set("version", x.Version) |
| 188 | q.Set("digest", x.Digest) |
| 189 | } |
| 190 | |
| 191 | func (c *Client) Stop(ctx context.Context, host string, role Role) error { |
| 192 | _, _, err := c.do(ctx, DeadlineControl, http.MethodPost, "/stop", hostRole(host, role), nil) |
| 193 | return err |
| 194 | } |
| 195 | |
| 196 | // Cmd starts host's cmd unit and returns its invocation (run spec §5.4). |
| 197 | func (c *Client) Cmd(ctx context.Context, host string, memoryMax int64) (string, error) { |
| 198 | return c.cmdQuery(ctx, host, memoryMax, nil) |
| 199 | } |
| 200 | |
| 201 | // CmdExe is Cmd with the executable named: the serving generation's staged |
| 202 | // build, which root validates and mounts in place of the live link, as for |
| 203 | // the legacy sidecar (zero-downtime spec §7.5). A 403 is root's refusal, |
| 204 | // reported as ErrStagedExe. |
| 205 | func (c *Client) CmdExe(ctx context.Context, host string, memoryMax int64, x StagedExe) (string, error) { |
| 206 | return c.cmdQuery(ctx, host, memoryMax, &x) |
| 207 | } |
| 208 | |
| 209 | func (c *Client) cmdQuery(ctx context.Context, host string, memoryMax int64, x *StagedExe) (string, error) { |
| 210 | q := hostRole(host, "") |
| 211 | q.Set("memory_max", strconv.FormatInt(memoryMax, 10)) |
| 212 | if x != nil { |
| 213 | q.Set("exe", x.Path) |
| 214 | q.Set("version", x.Version) |
| 215 | q.Set("digest", x.Digest) |
| 216 | } |
| 217 | var out StartResponse |
| 218 | status, _, err := c.do(ctx, DeadlineControl, http.MethodPost, "/cmd", q, &out) |
| 219 | if err != nil { |
| 220 | if status == http.StatusForbidden && x != nil { |
| 221 | return "", fmt.Errorf("%w: %v", ErrStagedExe, err) |
| 222 | } |
| 223 | return "", err |
| 224 | } |
| 225 | return out.InvocationID, nil |
| 226 | } |
| 227 | |
| 228 | // CmdStatus long-polls the named invocation of host's cmd unit. |
| 229 | func (c *Client) CmdStatus(ctx context.Context, host, invocation string) (CmdStatus, error) { |
| 230 | var out CmdStatus |
| 231 | _, _, err := c.do(ctx, DeadlineCmdPoll, http.MethodGet, "/cmd/"+url.PathEscape(host), url.Values{"invocation": {invocation}}, &out) |
| 232 | return out, err |
| 233 | } |
| 234 | |
| 235 | func (c *Client) Reset(ctx context.Context, host string, role Role) error { |
| 236 | _, _, err := c.do(ctx, DeadlineControl, http.MethodPost, "/reset", hostRole(host, role), nil) |
| 237 | return err |
| 238 | } |
| 239 | |
| 240 | // genValues is the query of a generation verb: the run role and gen=<n>. |
| 241 | func genValues(host string, n uint64) url.Values { |
| 242 | q := hostRole(host, RoleRun) |
| 243 | q.Set("gen", strconv.FormatUint(n, 10)) |
| 244 | return q |
| 245 | } |
| 246 | |
| 247 | // StartGen starts generation n's unit (zero-downtime spec §7.6) with its |
| 248 | // tier and returns the invocation id the server read after the start. A |
| 249 | // failure whose cleanup could not prove the unit gone wraps ErrGenUnproven. |
| 250 | func (c *Client) StartGen(ctx context.Context, host string, n uint64, memoryMax int64) (string, error) { |
| 251 | return c.startUnitQuery(ctx, genValues(host, n), memoryMax, nil) |
| 252 | } |
| 253 | |
| 254 | // StartGenExe is StartGen with the generation's executable named: the |
| 255 | // staged build root validates and mounts in place of the live link. A 403 |
| 256 | // is root's refusal, reported as ErrStagedExe. |
| 257 | func (c *Client) StartGenExe(ctx context.Context, host string, n uint64, memoryMax int64, x StagedExe) (string, error) { |
| 258 | return c.startUnitQuery(ctx, genValues(host, n), memoryMax, &x) |
| 259 | } |
| 260 | |
| 261 | // StartExe is Start with the unit's executable named: a staged build root |
| 262 | // validates and mounts in place of the live link, for the original |
| 263 | // template's run unit (the build a cold wake recorded, settings spec §4.2) |
| 264 | // or the legacy sidecar. A 403 is root's refusal, reported as |
| 265 | // ErrStagedExe. |
| 266 | func (c *Client) StartExe(ctx context.Context, host string, role Role, memoryMax int64, x StagedExe) (string, error) { |
| 267 | return c.startUnitQuery(ctx, hostRole(host, role), memoryMax, &x) |
| 268 | } |
| 269 | |
| 270 | // StartSidecarExe is StartExe for the legacy sidecar role: the serving |
| 271 | // generation's staged build (zero-downtime spec §7.5). |
| 272 | func (c *Client) StartSidecarExe(ctx context.Context, host string, memoryMax int64, x StagedExe) (string, error) { |
| 273 | return c.StartExe(ctx, host, RoleSidecar, memoryMax, x) |
| 274 | } |
| 275 | |
| 276 | // startUnitQuery is /start for a generation or the sidecar, optionally with |
| 277 | // a staged executable. One place maps root's two distinct refusals: 403 is |
| 278 | // ErrStagedExe (nothing was mounted), and the generation verbs' unproven |
| 279 | // status is ErrGenUnproven. |
| 280 | func (c *Client) startUnitQuery(ctx context.Context, q url.Values, memoryMax int64, x *StagedExe) (string, error) { |
| 281 | q.Set("memory_max", strconv.FormatInt(memoryMax, 10)) |
| 282 | if x != nil { |
| 283 | setStagedExe(q, *x) |
| 284 | } |
| 285 | var out StartResponse |
| 286 | status, _, err := c.do(ctx, DeadlineControl, http.MethodPost, "/start", q, &out) |
| 287 | if err != nil { |
| 288 | switch { |
| 289 | case status == http.StatusForbidden && x != nil: |
| 290 | return "", fmt.Errorf("%w: %v", ErrStagedExe, err) |
| 291 | case status == statusGenUnproven && q.Has("gen"): |
| 292 | return "", fmt.Errorf("%w: %w", ErrGenUnproven, err) |
| 293 | } |
| 294 | return "", err |
| 295 | } |
| 296 | return out.InvocationID, nil |
| 297 | } |
| 298 | |
| 299 | // StopGen stops generation n's unit only. kill kills it outright and |
| 300 | // returns once its cgroup is empty. |
| 301 | func (c *Client) StopGen(ctx context.Context, host string, n uint64, kill bool) error { |
| 302 | q := genValues(host, n) |
| 303 | if kill { |
| 304 | q.Set("kill", "1") |
| 305 | } |
| 306 | _, _, err := c.do(ctx, DeadlineControl, http.MethodPost, "/stop", q, nil) |
| 307 | return err |
| 308 | } |
| 309 | |
| 310 | func (c *Client) ResetGen(ctx context.Context, host string, n uint64) error { |
| 311 | _, _, err := c.do(ctx, DeadlineControl, http.MethodPost, "/reset", genValues(host, n), nil) |
| 312 | return err |
| 313 | } |
| 314 | |
| 315 | // Parent sets host's slice memory, the generation parent's backstop (spec |
| 316 | // §7.3): one tier, or two summed during an overlap. |
| 317 | func (c *Client) Parent(ctx context.Context, host string, memoryMax int64) error { |
| 318 | q := url.Values{"host": {host}, "memory_max": {strconv.FormatInt(memoryMax, 10)}} |
| 319 | _, _, err := c.do(ctx, DeadlineControl, http.MethodPost, "/parent", q, nil) |
| 320 | return err |
| 321 | } |
| 322 | |
| 323 | // Sweep stops every instance unit on the box (design §6.7). A 502 carries |
| 324 | // the units that were stopped and the ones that would not; both are |
| 325 | // returned with the error. |
| 326 | func (c *Client) Sweep(ctx context.Context) (SweepResponse, error) { |
| 327 | var out SweepResponse |
| 328 | status, _, err := c.do(ctx, DeadlineSweep, http.MethodPost, "/sweep", nil, &out) |
| 329 | if err != nil && status == http.StatusBadGateway { |
| 330 | return out, errors.Join(err, errors.New("some instance units would not stop")) |
| 331 | } |
| 332 | return out, err |
| 333 | } |
| 334 | |
| 335 | // Ready is nil on 204 from a service of this agent's own build, a |
| 336 | // *NotReadyError on 503, and a *StaleBuildError on a 204 whose BuildHeader |
| 337 | // is missing or names another build (#256). The verdict is asked first: |
| 338 | // a service that is both stale and not ready says why it is not ready. |
| 339 | func (c *Client) Ready(ctx context.Context) error { |
| 340 | _, hdr, err := c.do(ctx, DeadlineReady, http.MethodGet, "/ready", nil, nil) |
| 341 | if err != nil { |
| 342 | return err |
| 343 | } |
| 344 | if got := hdr.Get(BuildHeader); c.Build != "" && got != c.Build { |
| 345 | return &StaleBuildError{Running: got, Want: c.Build} |
| 346 | } |
| 347 | return nil |
| 348 | } |
| 349 | |
amadan