| 1 | // Package runs is what the console, the box and the CLI share about one |
| 2 | // `carlos run` (run spec): ids, the run's objects under its app's subtree, |
| 3 | // the wire shapes, the limits and the reasons a run ends without an exit. |
| 4 | // It holds no policy beyond validation; the record's run state is |
| 5 | // provision.RunState. |
| 6 | package runs |
| 7 | |
| 8 | import ( |
| 9 | "crypto/rand" |
| 10 | "encoding/base32" |
| 11 | "fmt" |
| 12 | "io" |
| 13 | "regexp" |
| 14 | "strconv" |
| 15 | "strings" |
| 16 | "time" |
| 17 | |
| 18 | "github.com/carlosframework/platform/internal/release" |
| 19 | ) |
| 20 | |
| 21 | const ( |
| 22 | MaxArgs = 256 |
| 23 | MaxArgvBytes = 64 << 10 |
| 24 | MaxFileBytes = 64 << 20 |
| 25 | MaxOutputBytes = 64 << 20 |
| 26 | // MetaAllowance bounds the meta part and the multipart framing on top |
| 27 | // of the files (run spec §5.1). |
| 28 | MetaAllowance = 1 << 20 |
| 29 | MinTimeout = time.Second |
| 30 | MaxTimeout = 6 * time.Hour |
| 31 | DefaultTimeout = 30 * time.Minute |
| 32 | PickupWindow = 60 * time.Second |
| 33 | LostAfter = time.Hour |
| 34 | KeepFor = 24 * time.Hour |
| 35 | ChunkBytes = 256 << 10 |
| 36 | ChunkEvery = time.Second |
| 37 | StreamOut = "o" |
| 38 | StreamErr = "e" |
| 39 | // A follow page holds at most PageChunks chunks and PageBytes of |
| 40 | // output, so its JSON stays under the CLI's 1 MiB response cap. |
| 41 | PageChunks = 64 |
| 42 | PageBytes = 512 << 10 |
| 43 | ) |
| 44 | |
| 45 | // Why the box refused a run (result.json's refused). |
| 46 | const ( |
| 47 | RefusedFlat = "flat" |
| 48 | RefusedClaim = "claim" |
| 49 | RefusedWake = "wake" |
| 50 | RefusedIncarnation = "incarnation" |
| 51 | RefusedImport = "import" |
| 52 | // RefusedExpired is written by the console, not a box: the run expired |
| 53 | // before any box picked it up (run spec §5.2), so no box ever wrote a |
| 54 | // result. The CLI reports it as "not picked up". |
| 55 | RefusedExpired = "expired" |
| 56 | ) |
| 57 | |
| 58 | // Why a run was interrupted, beside activator's Stop* reasons (RUN-C44 to |
| 59 | // RUN-C46, RUN-C82). |
| 60 | const ( |
| 61 | InterruptedLost = "its box stopped reporting" |
| 62 | InterruptedAgentRestart = "the box's agent restarted" |
| 63 | InterruptedGone = "the command's unit went away" |
| 64 | InterruptedStartFailed = "the box could not start the command" |
| 65 | ) |
| 66 | |
| 67 | // DirName is the directory under an instance's own data directory that |
| 68 | // holds each run's files: activator.CmdRunDirName, restated here because |
| 69 | // runs imports no activator (Plan 3 pins the two equal). |
| 70 | const DirName = ".carlos-run" |
| 71 | |
| 72 | var idEnc = base32.NewEncoding("abcdefghijklmnopqrstuvwxyz234567").WithPadding(base32.NoPadding) |
| 73 | |
| 74 | var idShape = regexp.MustCompile(`^run_[a-z2-7]{26}$`) |
| 75 | |
| 76 | // NewID mints a run id from r: "run_" and 26 lowercase base32 characters, |
| 77 | // the encoding of 16 random bytes (128 bits). A run id is the only thing that |
| 78 | // stops the record from admitting a run twice and the only thing that names |
| 79 | // a run's objects, so production code mints with MintID; NewID takes a reader |
| 80 | // only so a test can supply a fixed one. |
| 81 | func NewID(r io.Reader) (string, error) { |
| 82 | var b [16]byte |
| 83 | if _, err := io.ReadFull(r, b[:]); err != nil { |
| 84 | return "", err |
| 85 | } |
| 86 | return "run_" + idEnc.EncodeToString(b[:]), nil |
| 87 | } |
| 88 | |
| 89 | // MintID mints a run id from the operating system's cryptographic random |
| 90 | // source. |
| 91 | func MintID() (string, error) { return NewID(rand.Reader) } |
| 92 | |
| 93 | // ValidID reports whether id is a run id. Every id that passes is exactly 30 |
| 94 | // bytes of [a-z2-7_] and so is one clean path element: never empty, with no |
| 95 | // slash, backslash or dot, and never "." or "..". The console validates an |
| 96 | // id from a request with it, the box validates one from a record with it, and |
| 97 | // the store helpers below refuse an id that fails it. |
| 98 | func ValidID(id string) bool { return idShape.MatchString(id) } |
| 99 | |
| 100 | // RunsPrefix is where an app's runs live: under its own subtree, so the |
| 101 | // owner gate maps onto them and app trash and purge sweep them. |
| 102 | func RunsPrefix(acct, app string) string { return release.AppPrefix(acct, app) + "runs/" } |
| 103 | |
| 104 | func Prefix(acct, app, id string) string { return RunsPrefix(acct, app) + id + "/" } |
| 105 | func RequestKey(acct, app, id string) string { return Prefix(acct, app, id) + "request.json" } |
| 106 | func ResultKey(acct, app, id string) string { return Prefix(acct, app, id) + "result.json" } |
| 107 | func AuditedKey(acct, app, id string) string { return Prefix(acct, app, id) + "audited" } |
| 108 | func OutPrefix(acct, app, id string) string { return Prefix(acct, app, id) + "out/" } |
| 109 | |
| 110 | // InputKey is where an uploaded file lives. name must have passed PlainName |
| 111 | // (CheckFiles does that); InputKey does not check it, and a name with a |
| 112 | // slash or ".." would leave in/. |
| 113 | func InputKey(acct, app, id, name string) string { return Prefix(acct, app, id) + "in/" + name } |
| 114 | |
| 115 | // ChunkKey is out/<seq>-<stream>, the seq zero-padded so keys sort in the |
| 116 | // order the agent read the output. stream must be StreamOut or StreamErr |
| 117 | // (ChunkKey does not check it), and seq at most 99,999,999, the 8-digit |
| 118 | // format; the 64 MB output cap in 256 KB chunks never gets near it. |
| 119 | func ChunkKey(acct, app, id string, seq int64, stream string) string { |
| 120 | return OutPrefix(acct, app, id) + fmt.Sprintf("%08d-%s", seq, stream) |
| 121 | } |
| 122 | |
| 123 | // ParseChunkKey reads a key ChunkKey made under outPrefix. |
| 124 | func ParseChunkKey(key, outPrefix string) (int64, string, bool) { |
| 125 | name, ok := strings.CutPrefix(key, outPrefix) |
| 126 | if !ok || len(name) != 10 || name[8] != '-' { |
| 127 | return 0, "", false |
| 128 | } |
| 129 | stream := name[9:] |
| 130 | if stream != StreamOut && stream != StreamErr { |
| 131 | return 0, "", false |
| 132 | } |
| 133 | seq, err := strconv.ParseInt(name[:8], 10, 64) |
| 134 | if err != nil || seq < 1 { |
| 135 | return 0, "", false |
| 136 | } |
| 137 | for i := 0; i < 8; i++ { // ParseInt would also take a sign |
| 138 | if name[i] < '0' || name[i] > '9' { |
| 139 | return 0, "", false |
| 140 | } |
| 141 | } |
| 142 | return seq, stream, true |
| 143 | } |
| 144 | |
| 145 | // Meta is the create request's first part. |
| 146 | type Meta struct { |
| 147 | Argv []string `json:"argv"` |
| 148 | TimeoutS int64 `json:"timeout_s"` |
| 149 | Files []MetaFile `json:"files"` |
| 150 | // OnBehalfOf is the person a run service says it is running for. Only |
| 151 | // a run service may send it; it is recorded on the service's word and |
| 152 | // grants nothing (service runs spec §2). |
| 153 | OnBehalfOf string `json:"on_behalf_of,omitempty"` |
| 154 | } |
| 155 | |
| 156 | type MetaFile struct { |
| 157 | Name string `json:"name"` |
| 158 | Size int64 `json:"size"` |
| 159 | // SHA256 is the file's hex sha256 as the caller declares it. A keyed |
| 160 | // create must declare it, and the console checks the uploaded bytes |
| 161 | // against it whenever it is declared (service runs spec § 3). |
| 162 | SHA256 string `json:"sha256,omitempty"` |
| 163 | } |
| 164 | |
| 165 | // File is one uploaded file as request.json records it. |
| 166 | type File struct { |
| 167 | Name string `json:"name"` |
| 168 | Size int64 `json:"size"` |
| 169 | SHA256 string `json:"sha256"` |
| 170 | } |
| 171 | |
| 172 | // Request is runs/<id>/request.json (run spec §5.1). Incarnation is the |
| 173 | // record's (hostclaim.Incarnation of account, app, host and CreatedAt). |
| 174 | type Request struct { |
| 175 | Schema int `json:"schema"` |
| 176 | ID string `json:"id"` |
| 177 | Account string `json:"account"` |
| 178 | App string `json:"app"` |
| 179 | Host string `json:"host"` |
| 180 | Incarnation string `json:"incarnation"` |
| 181 | Argv []string `json:"argv"` |
| 182 | TimeoutS int64 `json:"timeout_s"` |
| 183 | Files []File `json:"files"` |
| 184 | By string `json:"by"` |
| 185 | RequestedAt time.Time `json:"requested_at"` |
| 186 | // A service's run (service runs spec §2): the credential's name, for |
| 187 | // display; its stable id, which is what decides whose run it is (a |
| 188 | // name is reused after a revoke, an id never); and the person it said |
| 189 | // it ran for, unverified. All three are empty for a person's run. |
| 190 | ByService string `json:"by_service,omitempty"` |
| 191 | ByCredential string `json:"by_credential,omitempty"` |
| 192 | OnBehalfOf string `json:"on_behalf_of,omitempty"` |
| 193 | // IdempotencyKey is a keyed create's key as the record's run_keys holds |
| 194 | // it, scoped to its caller and hashed (ScopedKey), never the caller's |
| 195 | // own words. It links the run to its key after the record has pruned |
| 196 | // the entry (service runs spec §§ 2, 3). Empty for an unkeyed create. |
| 197 | IdempotencyKey string `json:"idempotency_key,omitempty"` |
| 198 | } |
| 199 | |
| 200 | // Timeout is the run's timeout as a duration. |
| 201 | func (r Request) Timeout() time.Duration { return time.Duration(r.TimeoutS) * time.Second } |
| 202 | |
| 203 | // ServiceLine is how every surface names the requester of a service's run, |
| 204 | // "" for a person's. The person a service names is always shown as that |
| 205 | // service's statement, never as the one who ran it. |
| 206 | func (r Request) ServiceLine() string { |
| 207 | if by := r.ServiceBy(); by != "" { |
| 208 | return "run " + by |
| 209 | } |
| 210 | return "" |
| 211 | } |
| 212 | |
| 213 | // ServiceBy is ServiceLine without its leading "run ", for a sentence that |
| 214 | // already says "run" ("run <id> started by service ci (scheduled)"). "" for |
| 215 | // a person's run. |
| 216 | func (r Request) ServiceBy() string { |
| 217 | switch { |
| 218 | case r.ByService == "": |
| 219 | return "" |
| 220 | case r.OnBehalfOf == "": |
| 221 | return "by service " + r.ByService + " (scheduled)" |
| 222 | } |
| 223 | return "by service " + r.ByService + " for " + r.OnBehalfOf + " (as stated by the service)" |
| 224 | } |
| 225 | |
| 226 | // Result is runs/<id>/result.json (run spec §5.5), written only after the |
| 227 | // last chunk. Signal is the decimal signal number when a signal ended the |
| 228 | // command (index item 5); Detail is the member-safe wake error for a |
| 229 | // refused wake (index item 4). |
| 230 | type Result struct { |
| 231 | Exit int `json:"exit"` |
| 232 | Signal string `json:"signal"` |
| 233 | TimedOut bool `json:"timed_out"` |
| 234 | Cancelled bool `json:"cancelled"` |
| 235 | Interrupted string `json:"interrupted"` |
| 236 | Refused string `json:"refused"` |
| 237 | Detail string `json:"detail,omitempty"` |
| 238 | Started time.Time `json:"started"` |
| 239 | Ended time.Time `json:"ended"` |
| 240 | Box string `json:"box"` |
| 241 | LastSeq int64 `json:"last_seq"` |
| 242 | Truncated bool `json:"truncated"` |
| 243 | } |
| 244 | |
| 245 | // Chunk is one output chunk in a follow page; Data is base64 in JSON. |
| 246 | type Chunk struct { |
| 247 | Seq int64 `json:"seq"` |
| 248 | Stream string `json:"stream"` |
| 249 | Data []byte `json:"data"` |
| 250 | } |
| 251 | |
| 252 | // Page is GET …/runs/{id}?after=<seq>. State is the record's state for |
| 253 | // this run, or "" when the record no longer names it. More says the page |
| 254 | // was cut short: ask again at once. |
| 255 | type Page struct { |
| 256 | ID string `json:"id"` |
| 257 | State string `json:"state"` |
| 258 | Chunks []Chunk `json:"chunks"` |
| 259 | Result *Result `json:"result,omitempty"` |
| 260 | More bool `json:"more"` |
| 261 | } |
| 262 | |
| 263 | // Ack is POST …/runs/{id}/ack. |
| 264 | type Ack struct { |
| 265 | LastSeq int64 `json:"last_seq"` |
| 266 | } |
| 267 | |
| 268 | // Created is the 202 to a create. State is the record's state for the run, |
| 269 | // "" when the record no longer names it. Replayed says the run was admitted |
| 270 | // earlier under the same Idempotency-Key, and this create started nothing. |
| 271 | type Created struct { |
| 272 | ID string `json:"id"` |
| 273 | Host string `json:"host"` |
| 274 | State string `json:"state"` |
| 275 | Replayed bool `json:"replayed,omitempty"` |
| 276 | } |
| 277 | |
| 278 | // KeyAnswer is GET …/runs?key=<key>: the run admitted under the caller's |
| 279 | // key. |
| 280 | type KeyAnswer struct { |
| 281 | RunID string `json:"run_id"` |
| 282 | } |
| 283 | |
| 284 | // StateAnswer is the answer to cancel, expire and ack. |
| 285 | type StateAnswer struct { |
| 286 | State string `json:"state"` |
| 287 | } |
| 288 | |
amadan