zlayer_agent/runtime.rs
1//! Abstract container runtime interface
2//!
3//! Defines the Runtime trait that can be implemented for different container runtimes
4//! (containerd, CRI-O, etc.)
5
6use crate::cgroups_stats::ContainerStats;
7use crate::error::{AgentError, Result};
8use futures_util::Stream;
9use std::collections::VecDeque;
10use std::future::Future;
11use std::net::IpAddr;
12use std::pin::Pin;
13use std::sync::Arc;
14use std::time::Duration;
15use tokio::io::{AsyncRead, AsyncWrite};
16use tokio::sync::Mutex;
17use tokio::task::JoinHandle;
18use zlayer_observability::logs::{LogEntry, LogSource, LogStream};
19use zlayer_spec::{PullPolicy, RegistryAuth, ServiceSpec};
20
21/// Container state
22#[derive(Debug, Clone, PartialEq, Eq)]
23pub enum ContainerState {
24 /// Container is being pulled/created
25 Pending,
26 /// Init actions are running
27 Initializing,
28 /// Container is running
29 Running,
30 /// Container is stopping
31 Stopping,
32 /// Container has exited
33 Exited { code: i32 },
34 /// Container failed
35 Failed { reason: String },
36}
37
38impl ContainerState {
39 /// Stable lowercase string representation of the state.
40 ///
41 /// Used when surfacing container state through the API / `ps` output.
42 /// `Running` stringifies to `"running"` (matched case-insensitively by the
43 /// raft e2e harness when counting healthy replicas).
44 #[must_use]
45 pub fn as_str(&self) -> &'static str {
46 match self {
47 Self::Pending => "pending",
48 Self::Initializing => "initializing",
49 Self::Running => "running",
50 Self::Stopping => "stopping",
51 Self::Exited { .. } => "exited",
52 Self::Failed { .. } => "failed",
53 }
54 }
55}
56
57impl std::fmt::Display for ContainerState {
58 fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
59 f.write_str(self.as_str())
60 }
61}
62
63/// Container identifier.
64///
65/// Identifies a container by `(service, replica)` for the legacy single-group
66/// case, and extends with `role` + `node_id` for cluster-aware
67/// multi-group services and cross-node identification.
68///
69/// Defaults: `role = "default"`, `node_id = 0`. Existing constructors
70/// (`ContainerId::new(service, replica)`) produce these defaults. Use
71/// `ContainerId::with_role_and_node(...)` when the new fields matter.
72///
73/// Display:
74/// - With defaults: `{service}-rep-{replica}` (backward compat).
75/// - Otherwise: `{service}-{role}-{replica}-on-{node_id}`.
76#[derive(Debug, Clone, PartialEq, Eq, Hash, serde::Serialize, serde::Deserialize)]
77pub struct ContainerId {
78 pub service: String,
79 pub replica: u32,
80 /// Role within `replica_groups`. `"default"` for services without groups.
81 #[serde(default = "default_container_role")]
82 pub role: String,
83 /// Cluster node that owns this container. `0` in single-node deployments
84 /// or before the cluster is initialized.
85 #[serde(default)]
86 pub node_id: u64,
87}
88
89fn default_container_role() -> String {
90 "default".to_string()
91}
92
93impl ContainerId {
94 /// Build a legacy `{service, replica}` `ContainerId` with default `role`
95 /// and `node_id`. Used by all existing callsites — behavior is unchanged.
96 #[must_use]
97 pub fn new(service: impl Into<String>, replica: u32) -> Self {
98 Self {
99 service: service.into(),
100 replica,
101 role: default_container_role(),
102 node_id: 0,
103 }
104 }
105
106 /// Build a cluster-aware `ContainerId` with explicit `role` and `node_id`.
107 /// Used by `ServiceManager` when a service has `replica_groups` or when
108 /// the daemon participates in a multi-node cluster.
109 #[must_use]
110 pub fn with_role_and_node(
111 service: impl Into<String>,
112 replica: u32,
113 role: impl Into<String>,
114 node_id: u64,
115 ) -> Self {
116 Self {
117 service: service.into(),
118 replica,
119 role: role.into(),
120 node_id,
121 }
122 }
123
124 /// True when both `role` and `node_id` are at their defaults — i.e.
125 /// this is a legacy-shape `ContainerId`.
126 #[must_use]
127 pub fn is_legacy_shape(&self) -> bool {
128 self.role == "default" && self.node_id == 0
129 }
130
131 /// Parse a `ContainerId` back from its [`Display`](std::fmt::Display) form.
132 ///
133 /// This is the exact inverse of `Display`:
134 /// - `"{service}-rep-{replica}"` (legacy shape) → `ContainerId::new`.
135 /// - `"{service}-{role}-{replica}-on-{node_id}"` (cluster shape) →
136 /// `ContainerId::with_role_and_node`.
137 ///
138 /// The service name may itself contain `-`, so parsing anchors on the
139 /// rightmost structural markers (`-on-` then the trailing `-rep-`/`-{role}-`
140 /// segment) rather than splitting left-to-right. Returns `None` for any
141 /// string that does not match either shape (e.g. a hex id or a bare name).
142 #[must_use]
143 pub fn parse_display(s: &str) -> Option<Self> {
144 // Cluster shape: `{service}-{role}-{replica}-on-{node_id}`.
145 if let Some((head, node_str)) = s.rsplit_once("-on-") {
146 let node_id: u64 = node_str.parse().ok()?;
147 // `head` = `{service}-{role}-{replica}`. The replica is the last
148 // `-`-segment; the role is the segment before it; everything left
149 // of that is the (possibly hyphenated) service.
150 let (service_role, replica_str) = head.rsplit_once('-')?;
151 let replica: u32 = replica_str.parse().ok()?;
152 let (service, role) = service_role.rsplit_once('-')?;
153 if service.is_empty() || role.is_empty() {
154 return None;
155 }
156 return Some(Self::with_role_and_node(service, replica, role, node_id));
157 }
158 // Legacy shape: `{service}-rep-{replica}`.
159 if let Some((service, replica_str)) = s.rsplit_once("-rep-") {
160 let replica: u32 = replica_str.parse().ok()?;
161 if service.is_empty() {
162 return None;
163 }
164 return Some(Self::new(service, replica));
165 }
166 None
167 }
168}
169
170impl std::fmt::Display for ContainerId {
171 fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
172 if self.is_legacy_shape() {
173 write!(f, "{}-rep-{}", self.service, self.replica)
174 } else {
175 write!(
176 f,
177 "{}-{}-{}-on-{}",
178 self.service, self.role, self.replica, self.node_id
179 )
180 }
181 }
182}
183
184/// Container handle
185pub struct Container {
186 pub id: ContainerId,
187 /// Image reference this container was created from (canonical form, e.g.
188 /// `docker.io/library/nginx:1.29-alpine`). Surfaced through the API/`ps`.
189 pub image: String,
190 pub state: ContainerState,
191 pub pid: Option<u32>,
192 pub task: Option<JoinHandle<std::io::Result<()>>>,
193 /// Overlay network IP address assigned to this container
194 pub overlay_ip: Option<IpAddr>,
195 /// Health monitor task handle for this container
196 pub health_monitor: Option<JoinHandle<()>>,
197 /// Runtime-assigned port override (used by macOS sandbox where all
198 /// containers share the host network and need unique ports).
199 /// When `Some(port)`, the proxy should use this port instead of the
200 /// spec-declared endpoint port for this specific container's backend address.
201 pub port_override: Option<u16>,
202}
203
204/// Summary information about a cached image on the host runtime.
205#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
206pub struct ImageInfo {
207 /// Canonical image reference (e.g. `zachhandley/zlayer-manager:latest`).
208 pub reference: String,
209 /// Content-addressed digest if known (`sha256:...`). `None` when the
210 /// backend only tracks images by tag.
211 pub digest: Option<String>,
212 /// Total on-disk / in-cache size in bytes, when available.
213 pub size_bytes: Option<u64>,
214}
215
216/// Result of a prune operation.
217#[derive(Debug, Clone, Default, serde::Serialize, serde::Deserialize)]
218pub struct PruneResult {
219 /// Image references that were removed.
220 pub deleted: Vec<String>,
221 /// Bytes reclaimed from the cache. `0` when the backend cannot report.
222 pub space_reclaimed: u64,
223}
224
225/// Reason a container stopped running, as reported by [`Runtime::wait_outcome`].
226///
227/// Serialized as `snake_case` strings on the wire (`exited`, `signal`,
228/// `oom_killed`, `runtime_error`) so the API DTO can emit the reason as-is
229/// without a second translation layer.
230#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
231#[serde(rename_all = "snake_case")]
232pub enum WaitReason {
233 /// Container exited normally.
234 Exited,
235 /// Container was killed by a signal (e.g. `SIGKILL`, `SIGTERM`).
236 Signal,
237 /// Container was killed by the OOM killer.
238 OomKilled,
239 /// Runtime-side failure (pre-start error, runtime crash, etc.).
240 RuntimeError,
241}
242
243/// Wait condition mirroring Docker's `POST /containers/{id}/wait?condition=`.
244///
245/// Maps 1:1 to the wire form (`not-running`, `next-exit`, `removed`) via
246/// kebab-case serde so the daemon and the Docker compat shim can deserialize
247/// the query parameter in a single step. The default condition is
248/// [`WaitCondition::NotRunning`], matching Docker's behaviour when the
249/// `condition` query param is omitted.
250#[derive(Debug, Clone, Copy, PartialEq, Eq, Default, serde::Serialize, serde::Deserialize)]
251#[serde(rename_all = "kebab-case")]
252pub enum WaitCondition {
253 /// Wait until the container is not running. This is the default when
254 /// the caller doesn't specify a condition. Returns immediately if the
255 /// container has already exited; otherwise blocks until it does.
256 #[default]
257 NotRunning,
258 /// Wait for the next container exit, even if the container is already
259 /// stopped at the time of the call. Restarts the wait loop on each
260 /// observed exit.
261 NextExit,
262 /// Wait until the container is removed. Useful in conjunction with
263 /// `--rm` / `AutoRemove` lifecycle wiring.
264 Removed,
265}
266
267impl WaitCondition {
268 /// Return the wire string this variant serializes to (`"not-running"`,
269 /// `"next-exit"`, `"removed"`), matching Docker's `condition=` query
270 /// parameter spelling.
271 #[must_use]
272 pub const fn as_wire_str(self) -> &'static str {
273 match self {
274 Self::NotRunning => "not-running",
275 Self::NextExit => "next-exit",
276 Self::Removed => "removed",
277 }
278 }
279
280 /// Parse a Docker-style condition string (`"not-running"`, `"next-exit"`,
281 /// `"removed"`). Returns `None` for unknown values so callers can
282 /// distinguish "default" (omitted) from "rejected".
283 #[must_use]
284 pub fn from_wire_str(s: &str) -> Option<Self> {
285 match s {
286 "not-running" | "" => Some(Self::NotRunning),
287 "next-exit" => Some(Self::NextExit),
288 "removed" => Some(Self::Removed),
289 _ => None,
290 }
291 }
292}
293
294/// Richer wait result returned by [`Runtime::wait_outcome`].
295///
296/// Backwards-compatible with [`Runtime::wait_container`] (which returns just
297/// `exit_code`). The API handler uses this to populate the extended
298/// `ContainerWaitResponse` fields (`reason`, `signal`, `finished_at`) while
299/// existing callers that only need the exit code can keep using
300/// `wait_container`.
301#[derive(Debug, Clone)]
302pub struct WaitOutcome {
303 /// Process exit code (0 = success). When the container was killed by
304 /// signal `N`, this is typically `128 + N`.
305 pub exit_code: i32,
306 /// Classification of the exit.
307 pub reason: WaitReason,
308 /// Signal name when `reason == WaitReason::Signal`, e.g. `"SIGKILL"`.
309 /// Derived from `exit_code - 128` on a best-effort basis.
310 pub signal: Option<String>,
311 /// Time the container exited, if the runtime reports it.
312 pub finished_at: Option<chrono::DateTime<chrono::Utc>>,
313}
314
315impl WaitOutcome {
316 /// Build a plain `Exited` outcome with no signal/timestamp metadata — the
317 /// default that matches the pre-§3.12 behaviour.
318 #[must_use]
319 pub fn exited(exit_code: i32) -> Self {
320 Self {
321 exit_code,
322 reason: WaitReason::Exited,
323 signal: None,
324 finished_at: None,
325 }
326 }
327}
328
329/// Map a signal-style exit code (`128 + N`) to a canonical signal name.
330///
331/// Recognises common POSIX signals and falls back to `signal_<n>` for
332/// unknown numbers so the caller always gets *something* readable.
333#[must_use]
334pub fn signal_name_from_exit_code(exit_code: i32) -> Option<String> {
335 if exit_code <= 128 {
336 return None;
337 }
338 let n = exit_code - 128;
339 let name = match n {
340 1 => "SIGHUP",
341 2 => "SIGINT",
342 3 => "SIGQUIT",
343 4 => "SIGILL",
344 6 => "SIGABRT",
345 7 => "SIGBUS",
346 8 => "SIGFPE",
347 9 => "SIGKILL",
348 10 => "SIGUSR1",
349 11 => "SIGSEGV",
350 12 => "SIGUSR2",
351 13 => "SIGPIPE",
352 14 => "SIGALRM",
353 15 => "SIGTERM",
354 17 => "SIGSTOP",
355 18 => "SIGCONT",
356 _ => return Some(format!("signal_{n}")),
357 };
358 Some(name.to_string())
359}
360
361/// One streaming event emitted by [`Runtime::exec_stream`].
362///
363/// Runtimes push these events as the exec'd command produces output. The final
364/// event for any successful stream is always an [`ExecEvent::Exit`] carrying
365/// the process's exit code.
366#[derive(Debug, Clone, PartialEq, Eq)]
367pub enum ExecEvent {
368 /// A chunk of stdout from the running command. Emitted line-by-line by
369 /// Docker; other runtimes may emit the full buffered output in one event.
370 Stdout(String),
371 /// A chunk of stderr from the running command.
372 Stderr(String),
373 /// The command has exited with this exit code. Always the final event.
374 Exit(i32),
375}
376
377/// Boxed async stream of [`ExecEvent`]s returned by [`Runtime::exec_stream`].
378pub type ExecEventStream = Pin<Box<dyn Stream<Item = ExecEvent> + Send>>;
379
380/// Options accepted by [`Runtime::exec_pty`].
381///
382/// Mirrors the union of fields exposed by Docker's `POST /containers/{id}/exec`
383/// (`ExecConfig`) so the daemon can pass them through with minimal translation.
384/// Unlike [`Runtime::exec`] / [`Runtime::exec_stream`] (which capture stdout
385/// and stderr separately and return only after the process exits), `exec_pty`
386/// is the interactive entry point: it allocates a PTY when `tty` is set,
387/// streams I/O bidirectionally over a single duplex byte stream, and returns
388/// an [`ExecHandle`] that the caller drives concurrently with the running
389/// process.
390#[derive(Debug, Clone, Default, PartialEq, Eq)]
391#[allow(clippy::struct_excessive_bools)] // mirrors Docker's `ExecConfig` 1:1
392pub struct ExecOptions {
393 /// The argv vector for the exec'd process (`command[0]` is the binary).
394 pub command: Vec<String>,
395 /// Extra environment variables, in `KEY=VALUE` form. Merged into the
396 /// container's existing env on the runtime side.
397 pub env: Vec<String>,
398 /// Optional working directory inside the container. `None` keeps the
399 /// container's default `WORKDIR`.
400 pub working_dir: Option<String>,
401 /// Optional `user[:group]` override. `None` keeps the container's
402 /// configured user.
403 pub user: Option<String>,
404 /// Run the exec with privileged capabilities (Docker `Privileged`).
405 pub privileged: bool,
406 /// Allocate a TTY for the exec'd process (Docker `Tty`). When `true`, the
407 /// runtime should set up a pseudo-terminal pair and the duplex stream on
408 /// the returned [`ExecHandle`] carries multiplexed PTY traffic; when
409 /// `false`, the stream carries raw stdout/stderr without PTY framing.
410 pub tty: bool,
411 /// Attach stdin so the caller can write to the process (Docker
412 /// `AttachStdin`). When `false`, the writable half of the duplex stream
413 /// is effectively a no-op.
414 pub attach_stdin: bool,
415 /// Attach stdout so the caller receives the process's stdout on the
416 /// readable half of the duplex stream (Docker `AttachStdout`).
417 pub attach_stdout: bool,
418 /// Attach stderr so the caller receives the process's stderr on the
419 /// readable half of the duplex stream (Docker `AttachStderr`).
420 pub attach_stderr: bool,
421}
422
423/// Marker supertrait combining [`AsyncRead`] + [`AsyncWrite`] so they can be
424/// used together as a single trait object. Rust forbids stacking two
425/// non-auto traits directly in `dyn`, so [`ExecPtyStream`] is built on top
426/// of this helper instead.
427///
428/// A blanket impl below covers every type that already satisfies the four
429/// component bounds, so callers never need to implement `ExecDuplex`
430/// manually — they just hand the runtime any concrete duplex stream that's
431/// `AsyncRead + AsyncWrite + Send + Unpin`.
432pub trait ExecDuplex: AsyncRead + AsyncWrite + Send + Unpin {}
433
434impl<T> ExecDuplex for T where T: AsyncRead + AsyncWrite + Send + Unpin + ?Sized {}
435
436/// Duplex byte stream used by [`ExecHandle`] to shuttle stdin/stdout (and,
437/// when [`ExecOptions::tty`] is set, multiplexed PTY traffic) between the
438/// caller and the exec'd process.
439///
440/// `Unpin` is required (via [`ExecDuplex`]) so callers can poll the trait
441/// object directly via the usual `tokio::io::AsyncReadExt` /
442/// `AsyncWriteExt` extension methods without having to pin the box
443/// themselves.
444pub type ExecPtyStream = Box<dyn ExecDuplex + 'static>;
445
446/// Future returned by [`ExecHandle`] that resolves with the exec'd process's
447/// exit code once the runtime observes it has terminated.
448pub type ExecExitFuture = Pin<Box<dyn Future<Output = Result<i32>> + Send>>;
449
450/// Runtime-side handle returned by [`Runtime::exec_pty`].
451///
452/// Bundles everything a long-lived interactive exec session needs:
453///
454/// 1. A duplex [`ExecPtyStream`] for shuttling stdin/stdout (or full PTY
455/// traffic when `tty` is set) between the caller and the running process.
456/// 2. A `tokio::sync::mpsc::Sender<(rows, cols)>` so the caller can resize
457/// the allocated PTY in response to terminal-size changes (mirrors
458/// Docker's `POST /exec/{id}/resize` endpoint). Runtimes that don't allocate
459/// a PTY should still accept the channel and treat resize messages as
460/// no-ops; the channel is dropped on the runtime side once the process
461/// exits.
462/// 3. A boxed [`ExecExitFuture`] that resolves with the exit code once the
463/// runtime detects the process has terminated.
464///
465/// `ExecHandle` deliberately does not implement `Debug` / `Clone` because
466/// every field holds a trait object (or, in the exit future's case, an opaque
467/// boxed future). Consumers move the fields out by destructuring and then
468/// drive the I/O stream, the resize channel, and the exit future
469/// independently.
470pub struct ExecHandle {
471 /// Bidirectional byte channel between the caller and the exec'd process.
472 pub stream: ExecPtyStream,
473 /// Channel for sending `(rows, cols)` resize requests for the PTY
474 /// allocated to the exec session. Bounded so a stuck runtime can't make
475 /// the caller buffer unbounded resize events; senders should drop the
476 /// most recent excess size on backpressure.
477 pub resize: tokio::sync::mpsc::Sender<(u16, u16)>,
478 /// Future that resolves with the process's exit code once the runtime
479 /// observes the exec has terminated. Consumers typically `await` this on
480 /// a dedicated task while pumping the duplex stream on another.
481 pub exit: ExecExitFuture,
482}
483
484/// Which standard stream a [`LogChunk`] originated from.
485///
486/// Mirrors the three POSIX file descriptors. `Stdin` is included for
487/// completeness — Docker's multiplexed log header carries a `stdin` channel
488/// for attached interactive containers — but most container runtimes only
489/// emit `Stdout` / `Stderr` chunks.
490#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
491#[serde(rename_all = "snake_case")]
492pub enum LogChannel {
493 /// Standard input (rarely emitted; included for completeness with
494 /// Docker's stdcopy framing).
495 Stdin,
496 /// Standard output.
497 Stdout,
498 /// Standard error.
499 Stderr,
500}
501
502/// One chunk of container log output emitted by [`Runtime::logs_stream`].
503///
504/// Streams emit one `LogChunk` per line (or per Docker stdcopy frame) as
505/// data is produced by the container. Backends that expose timestamps
506/// populate `timestamp`; ones that don't leave it `None`.
507#[derive(Debug, Clone)]
508pub struct LogChunk {
509 /// Which standard stream produced this chunk.
510 pub stream: LogChannel,
511 /// Raw bytes of the chunk. Not necessarily UTF-8 — container output is
512 /// arbitrary binary data and consumers must handle invalid UTF-8.
513 pub bytes: bytes::Bytes,
514 /// When the runtime reported this chunk, when known.
515 pub timestamp: Option<chrono::DateTime<chrono::Utc>>,
516}
517
518/// Options accepted by [`Runtime::logs_stream`].
519///
520/// Mirrors the `GET /containers/{id}/logs` query parameters in the Docker
521/// Engine API so backends can pass them through with minimal translation.
522#[derive(Debug, Clone, Default)]
523#[allow(clippy::struct_excessive_bools)] // mirrors Docker's logs query params 1:1
524pub struct LogsStreamOptions {
525 /// Continue streaming after the current end-of-log marker. When `false`,
526 /// the stream terminates once the runtime has emitted all currently
527 /// buffered logs.
528 pub follow: bool,
529 /// Tail the last N lines before starting to stream. `None` means "all
530 /// available logs from the start".
531 pub tail: Option<u64>,
532 /// Earliest timestamp (Unix seconds) to include. `None` means "no
533 /// lower bound".
534 pub since: Option<i64>,
535 /// Latest timestamp (Unix seconds) to include. `None` means "no upper
536 /// bound".
537 pub until: Option<i64>,
538 /// When `true`, the runtime should populate [`LogChunk::timestamp`] for
539 /// every chunk. Backends that always carry timestamps may ignore this
540 /// flag.
541 pub timestamps: bool,
542 /// Include stdout in the stream.
543 pub stdout: bool,
544 /// Include stderr in the stream.
545 pub stderr: bool,
546}
547
548/// One periodic resource-usage sample emitted by [`Runtime::stats_stream`].
549///
550/// Mirrors the union of fields exposed by Docker's `/containers/{id}/stats`
551/// endpoint and Linux cgroup stat files so downstream consumers (autoscaler,
552/// `docker stats`-compat HTTP shim) can read a single shape regardless of
553/// backend. Counters that the runtime cannot report are left at zero —
554/// missing data is signalled separately via the surrounding stream
555/// metadata.
556#[derive(Debug, Clone)]
557pub struct StatsSample {
558 /// Cumulative container CPU time consumed, in nanoseconds.
559 pub cpu_total_ns: u64,
560 /// Cumulative system CPU time observed at the same moment, in
561 /// nanoseconds. Used to compute relative CPU percentage between
562 /// successive samples (Docker's classic `cpu_delta / system_delta`
563 /// formula).
564 pub cpu_system_ns: u64,
565 /// Number of CPUs currently online for this container. Used as the
566 /// final scaling factor in the CPU-percentage calculation.
567 pub online_cpus: u32,
568 /// Resident memory currently in use, in bytes (cgroup
569 /// `memory.usage_in_bytes` minus inactive page cache for v1, or
570 /// `memory.current` for v2).
571 pub mem_used_bytes: u64,
572 /// Memory limit configured on the container, in bytes. `0` when no
573 /// limit is set (cgroup reports its sentinel value).
574 pub mem_limit_bytes: u64,
575 /// Cumulative bytes received across all attached network interfaces.
576 pub net_rx_bytes: u64,
577 /// Cumulative bytes transmitted across all attached network interfaces.
578 pub net_tx_bytes: u64,
579 /// Cumulative bytes read from block devices.
580 pub blkio_read_bytes: u64,
581 /// Cumulative bytes written to block devices.
582 pub blkio_write_bytes: u64,
583 /// Number of process IDs currently running inside the container's pid
584 /// namespace.
585 pub pids_current: u64,
586 /// Configured pids limit, if any. `None` means unlimited.
587 pub pids_limit: Option<u64>,
588 /// Wallclock time the sample was taken.
589 pub timestamp: chrono::DateTime<chrono::Utc>,
590}
591
592/// One progress event emitted by [`Runtime::pull_image_stream`].
593///
594/// Backends emit a series of `Status` events as layers are downloaded /
595/// extracted, followed by exactly one `Done` event when the pull completes
596/// successfully. Errors mid-pull are propagated as the stream's `Err`
597/// variant and terminate the stream.
598#[derive(Debug, Clone)]
599pub enum PullProgress {
600 /// Progress update for an in-flight layer or stage.
601 Status {
602 /// Layer ID or other backend-specific identifier, when available.
603 id: Option<String>,
604 /// Human-readable status text, e.g. `"Pulling fs layer"`,
605 /// `"Downloading"`, `"Extracting"`. Always present; may be empty
606 /// when the backend has nothing to report this tick.
607 status: String,
608 /// Pre-formatted progress bar (Docker emits a string like
609 /// `"[========> ] 12.3MB/45.6MB"`). `None` when the
610 /// backend reports raw `current`/`total` only.
611 progress: Option<String>,
612 /// Bytes transferred so far for this layer, when reported.
613 current: Option<u64>,
614 /// Expected total bytes for this layer, when reported.
615 total: Option<u64>,
616 },
617 /// Pull completed successfully.
618 Done {
619 /// Resolved canonical image reference (typically the same as the
620 /// requested reference, but may include a digest the backend
621 /// resolved).
622 reference: String,
623 /// Content-addressed digest, when the backend reports one.
624 digest: Option<String>,
625 },
626}
627
628/// Boxed async stream of `Result<LogChunk, AgentError>` items returned by
629/// [`Runtime::logs_stream`].
630///
631/// `'static` lifetime so handlers can hold the stream past the trait method's
632/// borrow of `self`.
633pub type LogsStream = Pin<Box<dyn Stream<Item = Result<LogChunk>> + Send + 'static>>;
634
635/// Boxed async stream of `Result<StatsSample, AgentError>` items returned by
636/// [`Runtime::stats_stream`].
637pub type StatsStream = Pin<Box<dyn Stream<Item = Result<StatsSample>> + Send + 'static>>;
638
639/// Boxed async stream of `Result<PullProgress, AgentError>` items returned by
640/// [`Runtime::pull_image_stream`].
641pub type PullProgressStream = Pin<Box<dyn Stream<Item = Result<PullProgress>> + Send + 'static>>;
642
643/// Boxed async stream of TAR-archive byte chunks returned by
644/// [`Runtime::archive_get`].
645///
646/// Each yielded `Bytes` is a contiguous slice of the TAR archive that the
647/// runtime is producing for the requested container path. The stream ends
648/// once the archive is fully written. Mid-stream errors map to
649/// [`AgentError`] variants.
650pub type ArchiveStream = Pin<Box<dyn Stream<Item = Result<bytes::Bytes>> + Send + 'static>>;
651
652/// Stat metadata for a single path inside a container, returned by
653/// [`Runtime::archive_head`].
654///
655/// Mirrors Docker's `X-Docker-Container-Path-Stat` header payload (a
656/// base64-encoded JSON object with `name`, `size`, `mode`, `mtime`, and
657/// `linkTarget` fields). The API layer serializes this back into the same
658/// header for the `HEAD /containers/{id}/archive` response.
659#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
660pub struct PathStat {
661 /// Base name of the path (e.g. `"foo.txt"` for `/etc/foo.txt`).
662 pub name: String,
663 /// File size in bytes. For directories this is the size reported by the
664 /// runtime (typically the directory entry size, not the recursive sum).
665 pub size: i64,
666 /// Unix file mode bits (`S_IFMT | S_IRWXU | ...`). Encoded as a `u32` so
667 /// it round-trips through the Docker JSON header losslessly.
668 pub mode: u32,
669 /// Last-modification time as an RFC 3339 string. `None` when the runtime
670 /// cannot report it.
671 pub mtime: Option<String>,
672 /// Target of a symbolic link, when the path is a symlink. Empty string
673 /// for non-symlink paths (matching Docker's wire shape).
674 pub link_target: String,
675}
676
677/// Options accepted by [`Runtime::archive_put`].
678///
679/// Mirrors the query parameters Docker accepts on
680/// `PUT /containers/{id}/archive`:
681///
682/// * `noOverwriteDirNonDir=1` — refuse to replace a non-directory with a
683/// directory or vice versa. Default `false`.
684/// * `copyUIDGID=1` — preserve UID/GID of files in the archive instead of
685/// chown'ing them to the container's user. Default `false`.
686#[derive(Debug, Clone, Copy, Default, PartialEq, Eq)]
687pub struct ArchivePutOptions {
688 /// When `true`, the runtime must reject puts that would replace a
689 /// non-directory with a directory (or vice versa) at the destination.
690 pub no_overwrite_dir_non_dir: bool,
691 /// When `true`, preserve UID/GID of files in the archive verbatim.
692 pub copy_uid_gid: bool,
693}
694
695/// Per-network attachment reported by [`Runtime::inspect_detailed`].
696///
697/// Mirrors the subset of bollard's `EndpointSettings` that the API needs to
698/// populate `ContainerInfo.networks` for standalone containers. Kept in
699/// `zlayer-agent` (rather than `zlayer-spec`) because it's a runtime-level
700/// inspect result, not a deployment specification.
701#[derive(Debug, Clone, Default, PartialEq, Eq)]
702pub struct NetworkAttachmentDetail {
703 /// Network name as reported by the runtime (Docker's key in
704 /// `NetworkSettings.Networks`, e.g. `"bridge"` or a user-defined network
705 /// name).
706 pub network: String,
707 /// DNS aliases the container answers to on this network. Empty when the
708 /// runtime doesn't surface aliases.
709 pub aliases: Vec<String>,
710 /// Assigned IPv4 address on this network, if any. Empty strings are
711 /// normalised to `None`.
712 pub ipv4: Option<String>,
713}
714
715/// Per-container health detail reported by [`Runtime::inspect_detailed`].
716///
717/// Sourced directly from bollard's `ContainerState.health` (Docker's native
718/// healthcheck tracking). Our internal `HealthMonitor` in
719/// `crates/zlayer-agent/src/health.rs` drives service-level health events; for
720/// standalone containers the API reports the runtime-native status instead so
721/// that images with a baked-in `HEALTHCHECK` show up correctly.
722#[derive(Debug, Clone, Default, PartialEq, Eq)]
723pub struct HealthDetail {
724 /// One of `"none"`, `"starting"`, `"healthy"`, `"unhealthy"` (Docker's
725 /// `HealthStatusEnum`). Empty string is normalised to `"none"` upstream.
726 pub status: String,
727 /// Consecutive failing probe count, if the runtime reports it.
728 pub failing_streak: Option<u32>,
729 /// Output from the most recent failing probe, when available.
730 pub last_output: Option<String>,
731}
732
733/// Rich inspect details for a single container, returned by
734/// [`Runtime::inspect_detailed`].
735///
736/// Carries the fields `ContainerInfo` needs on top of the bare
737/// [`ContainerState`] reported by [`Runtime::container_state`]:
738/// published ports, attached networks, first IPv4, health, and `exit_code`.
739///
740/// Default is an all-empty record — that's what the default trait method
741/// returns for runtimes that don't (yet) implement rich inspect, and the API
742/// layer treats all fields as purely additive, so a default record still
743/// produces a backwards-compatible `ContainerInfo`.
744#[derive(Debug, Clone, Default, PartialEq, Eq)]
745pub struct ContainerInspectDetails {
746 /// Published port mappings (container → host), translated back from the
747 /// runtime's internal port-binding map.
748 pub ports: Vec<zlayer_spec::PortMapping>,
749 /// Networks the container is attached to, plus the aliases + IPv4 for each.
750 pub networks: Vec<NetworkAttachmentDetail>,
751 /// First non-empty IPv4 address found across the container's networks,
752 /// useful as a "primary" IP for simple clients that don't want to iterate
753 /// `networks`. `None` when the container isn't on any network with an IP.
754 pub ipv4: Option<String>,
755 /// Health status when the container has a Docker-native `HEALTHCHECK` or
756 /// the runtime otherwise reports a health state.
757 pub health: Option<HealthDetail>,
758 /// Most recent exit code, when the runtime reports one. `None` for
759 /// containers that are still running and have never exited.
760 pub exit_code: Option<i32>,
761}
762
763/// Lightweight summary of a single container reported by
764/// [`Runtime::list_containers`].
765///
766/// Reconciliation only needs to match runtime containers against
767/// `ZLayer`'s own metadata, which lives in the `com.zlayer.container_id`
768/// label (see `zlayer-api::handlers::container_id_map::ZLAYER_CONTAINER_ID_LABEL`).
769/// Carrying that label value plus the runtime-native id is sufficient for
770/// the standalone-container reconcile pass; richer fields can be added
771/// when concrete callers need them.
772#[derive(Debug, Clone, Default, PartialEq, Eq)]
773pub struct RuntimeContainerSummary {
774 /// Backend-native container handle (Docker's 64-char hex, the youki
775 /// state-dir name, etc.). Opaque to the reconciler — only used for
776 /// logging and to disambiguate listings.
777 pub runtime_id: String,
778 /// Value of the `com.zlayer.container_id` label, if the runtime
779 /// reports one for this container. Containers without this label are
780 /// foreign (not ZLayer-managed) and should be ignored by reconcile.
781 pub zlayer_container_id_label: Option<String>,
782}
783
784/// One row of process information returned by [`Runtime::top_container`].
785///
786/// Mirrors Docker's `GET /containers/{id}/top` response shape: a `Titles`
787/// vector that names each column, plus a `Processes` matrix where each row is
788/// the per-process column values. The runtime decides which `ps` fields to
789/// emit; the API/Docker shim forwards them verbatim.
790#[derive(Debug, Clone, Default, PartialEq, Eq)]
791pub struct ContainerTopOutput {
792 /// Column titles (e.g. `["UID", "PID", "PPID", "C", "STIME", "TTY", "TIME", "CMD"]`).
793 pub titles: Vec<String>,
794 /// One row per process; each row has the same length as `titles`.
795 pub processes: Vec<Vec<String>>,
796}
797
798/// Filesystem change kind reported by [`Runtime::changes_container`].
799///
800/// Matches Docker's numeric encoding: `0 = Modified`, `1 = Added`, `2 = Deleted`.
801#[derive(Debug, Clone, Copy, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
802#[serde(rename_all = "snake_case")]
803pub enum FilesystemChangeKind {
804 /// File or directory was modified in the container's writable layer.
805 Modified,
806 /// File or directory was added.
807 Added,
808 /// File or directory was deleted.
809 Deleted,
810}
811
812impl FilesystemChangeKind {
813 /// Numeric wire value used by Docker's `/containers/{id}/changes`.
814 #[must_use]
815 pub const fn as_docker_kind(self) -> u8 {
816 match self {
817 Self::Modified => 0,
818 Self::Added => 1,
819 Self::Deleted => 2,
820 }
821 }
822}
823
824/// One filesystem change reported by [`Runtime::changes_container`].
825#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
826pub struct FilesystemChangeEntry {
827 /// Path inside the container that changed (absolute, e.g. `/etc/hosts`).
828 pub path: String,
829 /// Kind of change.
830 pub kind: FilesystemChangeKind,
831}
832
833/// One published port mapping entry returned by [`Runtime::port_mappings_container`].
834///
835/// Mirrors one entry of Docker's `/containers/{id}/port` map. A single
836/// container port may bind multiple host endpoints (e.g. IPv4 + IPv6); each
837/// such binding yields one [`PortMappingEntry`].
838#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
839pub struct PortMappingEntry {
840 /// Container port number that's published.
841 pub container_port: u16,
842 /// Protocol (`"tcp"`, `"udp"`, `"sctp"`).
843 pub protocol: String,
844 /// Host IP address that the container's port is mapped to.
845 pub host_ip: Option<String>,
846 /// Host port number that the container's port is mapped to.
847 pub host_port: Option<u16>,
848}
849
850/// Result of a [`Runtime::prune_containers`] call.
851#[derive(Debug, Clone, Default, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
852pub struct ContainerPruneResult {
853 /// Container IDs that were removed (runtime-native or hex form).
854 pub deleted: Vec<String>,
855 /// Bytes reclaimed from the runtime's container storage. `0` when the
856 /// backend cannot report.
857 pub space_reclaimed: u64,
858}
859
860/// Runtime-level container restart policy attached to a
861/// [`ContainerResourceUpdate`].
862///
863/// Mirrors Docker's `HostConfig.RestartPolicy`. `name` is one of `""`,
864/// `"no"`, `"always"`, `"unless-stopped"`, or `"on-failure"`. Backends that
865/// can persist a restart policy (Docker via bollard, Youki via the
866/// supervisor's stored spec) honour it; backends that cannot
867/// (WASM, Mock) leave it unmodified.
868#[derive(Debug, Clone, Default, PartialEq, Eq)]
869pub struct ContainerRestartPolicyUpdate {
870 /// Restart policy name.
871 pub name: Option<String>,
872 /// Maximum retry count (only honoured when `name == "on-failure"`).
873 pub maximum_retry_count: Option<i64>,
874}
875
876/// Resource-update payload passed to [`Runtime::update_container_resources`].
877///
878/// Mirrors the resource subset of Docker's `POST /containers/{id}/update`
879/// body. Every field is `Option<...>`; backends apply only the fields that
880/// are `Some`. A request with all fields `None` is a no-op (consistent with
881/// Docker's behaviour).
882///
883/// Field semantics match Docker's wire shape: see [`ContainerUpdateRequest`]
884/// in `zlayer-types::api::containers` for the JSON encoding.
885#[derive(Debug, Clone, Default, PartialEq, Eq)]
886pub struct ContainerResourceUpdate {
887 /// CPU shares (cgroup `cpu.weight` / `cpu.shares`).
888 pub cpu_shares: Option<i64>,
889 /// Memory limit in bytes.
890 pub memory: Option<i64>,
891 /// CPU CFS period in microseconds.
892 pub cpu_period: Option<i64>,
893 /// CPU CFS quota in microseconds.
894 pub cpu_quota: Option<i64>,
895 /// CPU real-time period in microseconds.
896 pub cpu_realtime_period: Option<i64>,
897 /// CPU real-time runtime in microseconds.
898 pub cpu_realtime_runtime: Option<i64>,
899 /// CPUs allowed for execution (e.g. `"0-3"`).
900 pub cpuset_cpus: Option<String>,
901 /// Memory nodes (NUMA) allowed for execution.
902 pub cpuset_mems: Option<String>,
903 /// Soft memory limit in bytes.
904 pub memory_reservation: Option<i64>,
905 /// Total memory limit (memory + swap) in bytes. `-1` removes swap.
906 pub memory_swap: Option<i64>,
907 /// Kernel memory limit in bytes (deprecated upstream).
908 pub kernel_memory: Option<i64>,
909 /// Block IO weight (10-1000).
910 pub blkio_weight: Option<u16>,
911 /// PIDs limit. `0` or `-1` for unlimited.
912 pub pids_limit: Option<i64>,
913 /// Replacement restart policy. `None` leaves the policy unchanged.
914 pub restart_policy: Option<ContainerRestartPolicyUpdate>,
915}
916
917impl ContainerResourceUpdate {
918 /// Returns `true` when this update would not change anything — every
919 /// field is `None`. Backends short-circuit no-op updates rather than
920 /// touching the cgroup hierarchy.
921 #[must_use]
922 pub fn is_empty(&self) -> bool {
923 self.cpu_shares.is_none()
924 && self.memory.is_none()
925 && self.cpu_period.is_none()
926 && self.cpu_quota.is_none()
927 && self.cpu_realtime_period.is_none()
928 && self.cpu_realtime_runtime.is_none()
929 && self.cpuset_cpus.is_none()
930 && self.cpuset_mems.is_none()
931 && self.memory_reservation.is_none()
932 && self.memory_swap.is_none()
933 && self.kernel_memory.is_none()
934 && self.blkio_weight.is_none()
935 && self.pids_limit.is_none()
936 && self.restart_policy.is_none()
937 }
938}
939
940/// Result of a [`Runtime::update_container_resources`] call. Mirrors
941/// Docker's `{"Warnings": [...]}` response shape: backends append a string
942/// per "we accepted this but did not apply it" or "field deprecated"
943/// warning.
944#[derive(Debug, Clone, Default, PartialEq, Eq)]
945pub struct ContainerUpdateOutcome {
946 /// Human-readable warnings emitted while applying the update.
947 pub warnings: Vec<String>,
948}
949
950/// Detailed image inspect record returned by [`Runtime::inspect_image_native`].
951///
952/// Mirrors the union of fields exposed by Docker's `GET /images/{name}/json`
953/// (bollard's `ImageInspect`) so the API / Docker compat shim can surface a
954/// Docker-shaped JSON response without re-translating later. Optional fields
955/// remain `None` when the backend cannot provide them.
956#[derive(Debug, Clone, Default, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
957pub struct ImageInspectInfo {
958 /// Content-addressed image id (`sha256:...`), when known.
959 pub id: Option<String>,
960 /// All tags (`repo:tag`) currently pointing at this image.
961 pub repo_tags: Vec<String>,
962 /// Manifest digests this image is known under (`repo@sha256:...`).
963 pub repo_digests: Vec<String>,
964 /// Parent image id (`sha256:...`), when the image was built locally.
965 pub parent: Option<String>,
966 /// Human-readable comment recorded at commit/import time.
967 pub comment: Option<String>,
968 /// Creation timestamp in RFC 3339 form.
969 pub created: Option<String>,
970 /// Container id this image was committed from, when applicable.
971 pub container: Option<String>,
972 /// Daemon version that built / imported this image.
973 pub docker_version: Option<String>,
974 /// Author recorded on the image (e.g. `MAINTAINER` instruction).
975 pub author: Option<String>,
976 /// Hardware architecture the image targets (`amd64`, `arm64`, ...).
977 pub architecture: Option<String>,
978 /// Operating system the image targets (`linux`, `windows`, ...).
979 pub os: Option<String>,
980 /// Total on-disk size in bytes, when known.
981 pub size: Option<u64>,
982 /// Layer order: list of `sha256:...` digests, root-most first.
983 pub layers: Vec<String>,
984 /// Container env (`KEY=VALUE`).
985 pub env: Vec<String>,
986 /// Default command vector.
987 pub cmd: Vec<String>,
988 /// Default entrypoint vector.
989 pub entrypoint: Vec<String>,
990 /// Working directory inside the image.
991 pub working_dir: Option<String>,
992 /// User the image runs as by default.
993 pub user: Option<String>,
994 /// Image labels.
995 pub labels: std::collections::BTreeMap<String, String>,
996}
997
998/// One row of an image's history, returned by [`Runtime::image_history`].
999///
1000/// Mirrors Docker's `GET /images/{name}/history` response.
1001#[derive(Debug, Clone, Default, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
1002pub struct ImageHistoryEntry {
1003 /// Layer / image id (`sha256:...`). May be `<missing>` for layers that
1004 /// were dropped during a squash.
1005 pub id: String,
1006 /// Unix-seconds timestamp when this layer was created.
1007 pub created: i64,
1008 /// Dockerfile-style instruction that produced this layer.
1009 pub created_by: String,
1010 /// Tags that point at this specific layer.
1011 pub tags: Vec<String>,
1012 /// Layer size in bytes.
1013 pub size: u64,
1014 /// Optional comment recorded with the layer.
1015 pub comment: String,
1016}
1017
1018/// One result returned by [`Runtime::search_images`].
1019///
1020/// Mirrors Docker's `GET /images/search` response items.
1021#[derive(Debug, Clone, Default, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
1022pub struct ImageSearchResult {
1023 /// Image name (e.g. `library/nginx`).
1024 pub name: String,
1025 /// Free-text description of the image.
1026 pub description: String,
1027 /// Number of stars on the source registry, when reported.
1028 pub star_count: u64,
1029 /// Whether the image is officially curated.
1030 pub official: bool,
1031 /// Whether the image was produced by an automated build (deprecated by
1032 /// Docker but still surfaced for compatibility).
1033 pub automated: bool,
1034}
1035
1036/// Result of a [`Runtime::commit_container`] call.
1037#[derive(Debug, Clone, Default, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
1038pub struct CommitOutcome {
1039 /// Content-addressed image id of the freshly created image.
1040 pub id: String,
1041}
1042
1043/// Options accepted by [`Runtime::commit_container`].
1044///
1045/// Mirrors Docker's `POST /commit` query parameters. All fields are optional
1046/// — when `repo`/`tag` are both empty the runtime creates an untagged image.
1047#[derive(Debug, Clone, Default, PartialEq, Eq)]
1048pub struct CommitOptions {
1049 /// Repository name to apply to the committed image (e.g. `myapp`).
1050 pub repo: Option<String>,
1051 /// Tag to apply (defaults to `latest` when `repo` is set and tag is empty).
1052 pub tag: Option<String>,
1053 /// Free-form comment to record on the committed image.
1054 pub comment: Option<String>,
1055 /// Author to record on the committed image.
1056 pub author: Option<String>,
1057 /// Whether to pause the container before committing (defaults to `true`).
1058 pub pause: bool,
1059 /// Dockerfile-style instructions to apply during commit.
1060 pub changes: Option<String>,
1061}
1062
1063/// Boxed async stream of TAR-archive byte chunks returned by image save /
1064/// container export endpoints. Each yielded `Bytes` is a contiguous slice
1065/// of an uncompressed TAR archive.
1066pub type ImageExportStream = Pin<Box<dyn Stream<Item = Result<bytes::Bytes>> + Send + 'static>>;
1067
1068/// The runtime's already-open image-store handles: its
1069/// [`LocalRegistry`](zlayer_registry::LocalRegistry) plus the shared,
1070/// single-process-exclusive blob cache. Returned by
1071/// [`Runtime::image_store_handles`] so callers (the `POST /images/import`
1072/// handler) can reuse the daemon's open store rather than opening a second
1073/// handle (which would fail with "database is locked by another process").
1074pub type ImageStoreHandles = (
1075 Arc<zlayer_registry::LocalRegistry>,
1076 Arc<Box<dyn zlayer_registry::BlobCacheBackend>>,
1077);
1078
1079/// One progress event emitted by [`Runtime::load_images`].
1080///
1081/// `Status` events carry per-line progress reported by the daemon while
1082/// the tar is being unpacked; `Done` is emitted exactly once when load
1083/// completes successfully.
1084#[derive(Debug, Clone, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
1085#[serde(tag = "kind", rename_all = "snake_case")]
1086pub enum LoadProgress {
1087 /// Mid-load progress entry.
1088 Status {
1089 /// Layer or image id when reported.
1090 #[serde(default, skip_serializing_if = "Option::is_none")]
1091 id: Option<String>,
1092 /// Human-readable status text.
1093 status: String,
1094 },
1095 /// Load completed. `references` lists the image references that were
1096 /// loaded into the cache.
1097 Done {
1098 /// Loaded image references (`repo:tag` or `sha256:...`).
1099 references: Vec<String>,
1100 },
1101}
1102
1103/// Boxed async stream of [`LoadProgress`] events returned by
1104/// [`Runtime::load_images`].
1105pub type LoadProgressStream = Pin<Box<dyn Stream<Item = Result<LoadProgress>> + Send + 'static>>;
1106
1107/// How a runtime joins a container to the cross-node `WireGuard` overlay.
1108///
1109/// The overlay can be attached three ways depending on whether the container is
1110/// a host process, a Windows compute system, or a full VM:
1111#[derive(Debug, Clone, Copy, PartialEq, Eq)]
1112pub enum OverlayAttachKind {
1113 /// Linux host process: overlayd enters `/proc/<pid>/ns/net` and plumbs a
1114 /// veth into the container's network namespace (the default).
1115 HostNetns,
1116 /// Windows: the HCN endpoint + namespace were created at container-create
1117 /// time; the agent only registers the assigned IP.
1118 HostIp,
1119 /// Any guest with no host-visible netns/PID (macOS VZ-Linux over vsock
1120 /// today; also macOS native guests and the Windows WSL2 distro): overlayd
1121 /// allocates the overlay identity and the agent pushes it into the guest,
1122 /// where a kernel `WireGuard` device is brought up. Pushed via
1123 /// [`Runtime::push_overlay_config`].
1124 GuestManaged,
1125 /// Host-shared native runtime with no per-container netns (macOS Seatbelt;
1126 /// macOS native-VZ): the NODE joins the overlay at host level (a `utun`
1127 /// bound to the node overlay IP) and every container reaches the daemon and
1128 /// its siblings over the shared host stack (Seatbelt) or a host
1129 /// `node_ip:port -> guest` forward (native-VZ). There is no per-container
1130 /// veth, no host PID to attach, and no guest push; per-replica addressing is
1131 /// the node overlay IP plus [`Runtime::get_container_port_override`].
1132 HostProxy,
1133}
1134
1135/// Abstract container runtime trait
1136///
1137/// This trait abstracts over different container runtimes (containerd, CRI-O, etc.)
1138#[async_trait::async_trait]
1139pub trait Runtime: Send + Sync {
1140 /// Pull an image to local storage
1141 async fn pull_image(&self, image: &str) -> Result<()>;
1142
1143 /// Pull an image to local storage with a specific policy.
1144 ///
1145 /// When `auth` is `Some`, the runtime uses those inline credentials for
1146 /// the pull (§3.10 of `ZLAYER_SDK_FIXES.md`). When `auth` is `None`, the
1147 /// runtime falls back to its existing credential-store lookup keyed by
1148 /// registry hostname (or anonymous access when no match exists).
1149 ///
1150 /// Non-Docker runtimes may accept but ignore the `auth` argument — their
1151 /// OCI puller (`zlayer-registry`) already resolves credentials from the
1152 /// store by hostname, and inline auth is primarily a Docker-backend
1153 /// concern. Ignoring it is safe: callers that need inline auth should use
1154 /// the Docker runtime.
1155 ///
1156 /// `source` is the per-image [`zlayer_spec::SourcePolicy`] selecting which
1157 /// tiers (local store, S3, remote registry) the pull may consult and in
1158 /// what order. Runtimes that build the `zlayer-registry` `ImagePuller`
1159 /// chain (youki, macOS VZ-Linux) honor it; runtimes that delegate to an
1160 /// external daemon (Docker/WSL/HCS) accept but ignore it.
1161 async fn pull_image_with_policy(
1162 &self,
1163 image: &str,
1164 policy: PullPolicy,
1165 auth: Option<&RegistryAuth>,
1166 source: zlayer_spec::SourcePolicy,
1167 ) -> Result<()>;
1168
1169 /// Create a container
1170 async fn create_container(&self, id: &ContainerId, spec: &ServiceSpec) -> Result<()>;
1171
1172 /// Inject the daemon's secrets provider so container env vars using the
1173 /// `$S:` reference syntax can be resolved at container-create time.
1174 ///
1175 /// Called once, post-construction, after the daemon has selected its
1176 /// secrets store. Default is a no-op; runtimes that build OCI bundles with
1177 /// secret resolution (`YoukiRuntime`; `CompositeRuntime` forwards to its inner
1178 /// runtime) override this.
1179 fn set_secrets_provider(&self, _provider: std::sync::Arc<dyn zlayer_secrets::SecretsProvider>) {
1180 }
1181
1182 /// Start a container
1183 async fn start_container(&self, id: &ContainerId) -> Result<()>;
1184
1185 /// Stop a container
1186 async fn stop_container(&self, id: &ContainerId, timeout: Duration) -> Result<()>;
1187
1188 /// Remove a container
1189 async fn remove_container(&self, id: &ContainerId) -> Result<()>;
1190
1191 /// Get container state
1192 async fn container_state(&self, id: &ContainerId) -> Result<ContainerState>;
1193
1194 /// Get container logs as structured entries
1195 async fn container_logs(&self, id: &ContainerId, tail: usize) -> Result<Vec<LogEntry>>;
1196
1197 /// Execute command in container
1198 async fn exec(&self, id: &ContainerId, cmd: &[String]) -> Result<(i32, String, String)>;
1199
1200 /// Execute a command in a container with Docker `exec` options (`--user`,
1201 /// `-w`/`--workdir`, `-e`/`--env`) applied.
1202 ///
1203 /// The default implementation ignores the user/cwd/env overrides and
1204 /// delegates to [`Runtime::exec`], so runtimes that don't (yet) plumb them
1205 /// keep working unchanged. Runtimes that can honour them — notably the macOS
1206 /// VZ-Linux runtime, which drives the in-guest agent over vsock — override
1207 /// this to drop to the requested uid/gid, chdir, and inject env before exec.
1208 ///
1209 /// `opts.command` is the argv (`command[0]` is the binary); `opts.user`,
1210 /// `opts.working_dir`, and `opts.env` carry the Docker overrides. The
1211 /// `tty`/`attach_*`/`privileged` fields of [`ExecOptions`] are not consulted
1212 /// by this buffered entry point.
1213 async fn exec_with_opts(
1214 &self,
1215 id: &ContainerId,
1216 opts: &ExecOptions,
1217 ) -> Result<(i32, String, String)> {
1218 self.exec(id, &opts.command).await
1219 }
1220
1221 /// Execute a command in a container and stream stdout / stderr / exit
1222 /// events as they are produced.
1223 ///
1224 /// The default implementation calls the buffered [`Runtime::exec`] and
1225 /// emits everything as a single `Stdout` event, a single `Stderr` event,
1226 /// and a final `Exit` event. Runtimes that support true streaming (e.g.
1227 /// Docker via bollard) override this to produce line-by-line events as
1228 /// the command runs.
1229 ///
1230 /// The returned stream always terminates with exactly one
1231 /// [`ExecEvent::Exit`] as the final item on success. Errors that occur
1232 /// before the stream is returned (e.g. container not found, failure to
1233 /// create the exec) are surfaced via the outer `Result`; errors that
1234 /// occur mid-stream are logged by the runtime and the stream closes.
1235 async fn exec_stream(&self, id: &ContainerId, cmd: &[String]) -> Result<ExecEventStream> {
1236 let (exit, stdout, stderr) = self.exec(id, cmd).await?;
1237 let mut events: Vec<ExecEvent> = Vec::with_capacity(3);
1238 if !stdout.is_empty() {
1239 events.push(ExecEvent::Stdout(stdout));
1240 }
1241 if !stderr.is_empty() {
1242 events.push(ExecEvent::Stderr(stderr));
1243 }
1244 events.push(ExecEvent::Exit(exit));
1245 Ok(Box::pin(futures_util::stream::iter(events)))
1246 }
1247
1248 /// Start an interactive exec session against a container, returning an
1249 /// [`ExecHandle`] the caller drives concurrently with the running process.
1250 ///
1251 /// Unlike [`Runtime::exec`] (which buffers stdout/stderr and returns only
1252 /// after the process exits) and [`Runtime::exec_stream`] (which streams
1253 /// line-by-line events one-way), `exec_pty` is the long-lived bidirectional
1254 /// entry point: when [`ExecOptions::tty`] is set the runtime allocates a
1255 /// pseudo-terminal pair and the returned [`ExecHandle::stream`] shuttles
1256 /// raw PTY bytes; when `tty` is false the stream still carries
1257 /// stdin/stdout/stderr but without PTY framing. The handle's
1258 /// [`ExecHandle::resize`] channel mirrors Docker's
1259 /// `POST /exec/{id}/resize` and the [`ExecHandle::exit`] future resolves
1260 /// with the process exit code once the runtime detects termination.
1261 ///
1262 /// The default implementation returns [`AgentError::Unsupported`].
1263 /// Backends that can host interactive execs (Docker via bollard's
1264 /// `start_exec` with hijacked stream, Youki via libcontainer's exec API,
1265 /// HCS via the Windows console) override this. Runtimes that have no
1266 /// notion of an interactive exec (WASM in-process, mocks that don't need
1267 /// PTY traffic) should leave the default in place — callers then surface
1268 /// a clear error rather than degrade silently to a buffered exec.
1269 async fn exec_pty(&self, _id: &ContainerId, _opts: ExecOptions) -> Result<ExecHandle> {
1270 Err(AgentError::Unsupported(
1271 "exec_pty is not supported by this runtime".into(),
1272 ))
1273 }
1274
1275 /// Get container resource statistics from cgroups
1276 ///
1277 /// Returns CPU and memory statistics for the specified container.
1278 /// Used for metrics collection and autoscaling decisions.
1279 async fn get_container_stats(&self, id: &ContainerId) -> Result<ContainerStats>;
1280
1281 /// Wait for a container to exit and return its exit code
1282 ///
1283 /// This method blocks until the container exits or an error occurs.
1284 /// Used primarily for job execution to implement run-to-completion semantics.
1285 async fn wait_container(&self, id: &ContainerId) -> Result<i32>;
1286
1287 /// Wait for a container to exit and return a [`WaitOutcome`] with richer
1288 /// classification (exit code + reason + signal + `finished_at` timestamp).
1289 ///
1290 /// The default implementation delegates to [`Runtime::wait_container`] and
1291 /// synthesizes a [`WaitReason::Exited`] result with no signal/timestamp.
1292 /// Runtimes that can distinguish OOM kills, signal deaths, or report a
1293 /// finished-at time (e.g. the Docker runtime, which has
1294 /// `ContainerInspectResponse.state.oom_killed` / `.finished_at`) should
1295 /// override this.
1296 ///
1297 /// This is the legacy entry point that always uses
1298 /// [`WaitCondition::NotRunning`]. Callers that need to honour Docker's
1299 /// `condition=` query parameter should use
1300 /// [`Runtime::wait_outcome_with_condition`] instead.
1301 async fn wait_outcome(&self, id: &ContainerId) -> Result<WaitOutcome> {
1302 let exit_code = self.wait_container(id).await?;
1303 Ok(WaitOutcome::exited(exit_code))
1304 }
1305
1306 /// Wait for a container to reach a [`WaitCondition`] and return a
1307 /// [`WaitOutcome`].
1308 ///
1309 /// Mirrors Docker's `POST /containers/{id}/wait?condition=<...>` semantics:
1310 ///
1311 /// * [`WaitCondition::NotRunning`] (the default) — block until the
1312 /// container is no longer running. Returns the exit-code outcome.
1313 /// * [`WaitCondition::NextExit`] — wait for the next observed exit, even
1314 /// if the container is already stopped at call time. The default
1315 /// implementation cannot distinguish "already stopped" from "next exit",
1316 /// so it falls back to the same wait as `NotRunning`. Backends that can
1317 /// subscribe to runtime events (Docker via bollard's wait stream)
1318 /// override this to honour the semantic.
1319 /// * [`WaitCondition::Removed`] — block until the container has been
1320 /// removed. The default implementation again falls back to a normal
1321 /// wait; the Docker runtime overrides it via bollard's `condition`
1322 /// parameter.
1323 ///
1324 /// Default implementation delegates to [`Runtime::wait_outcome`] for all
1325 /// conditions, ignoring the condition argument. This keeps existing
1326 /// runtimes (Youki, WASM, mocks) working without code changes.
1327 async fn wait_outcome_with_condition(
1328 &self,
1329 id: &ContainerId,
1330 _condition: WaitCondition,
1331 ) -> Result<WaitOutcome> {
1332 self.wait_outcome(id).await
1333 }
1334
1335 /// Rename a container. Mirrors Docker's
1336 /// `POST /containers/{id}/rename?name=<new>` endpoint.
1337 ///
1338 /// `new_name` is the requested human-readable name (without any leading
1339 /// `/`). Backends are expected to validate the name against their own
1340 /// constraints (e.g. uniqueness, allowed characters) and return an
1341 /// appropriate [`AgentError`] on rejection.
1342 ///
1343 /// The default implementation returns [`AgentError::Unsupported`].
1344 /// Runtimes that can rename a live container override this:
1345 ///
1346 /// * Docker — calls bollard's `rename_container` with
1347 /// `RenameContainerOptions { name }`.
1348 /// * Youki — currently returns `Unsupported` because the libcontainer
1349 /// state-dir is keyed off the immutable `ContainerId` and renaming the
1350 /// on-disk layout safely while a container is alive would require
1351 /// coordination with the supervisor that owns the bundle path.
1352 /// * Other backends (WASM, HCS, mocks) inherit the `Unsupported` default.
1353 async fn rename_container(&self, _id: &ContainerId, _new_name: &str) -> Result<()> {
1354 Err(AgentError::Unsupported(
1355 "rename_container is not supported by this runtime".into(),
1356 ))
1357 }
1358
1359 /// Get container logs (stdout/stderr combined)
1360 ///
1361 /// Returns logs as structured entries.
1362 /// Used to capture job output after completion.
1363 async fn get_logs(&self, id: &ContainerId) -> Result<Vec<LogEntry>>;
1364
1365 /// Get the PID of a container's main process
1366 ///
1367 /// Returns:
1368 /// - `Ok(Some(pid))` for runtimes with real processes (Youki, Docker)
1369 /// - `Ok(None)` for runtimes without separate PIDs (WASM in-process)
1370 /// - `Err` if the container doesn't exist or there's an error
1371 ///
1372 /// Used for overlay network attachment and process management.
1373 async fn get_container_pid(&self, id: &ContainerId) -> Result<Option<u32>>;
1374
1375 /// How this runtime joins a container to the cross-node overlay network.
1376 ///
1377 /// Defaults to [`OverlayAttachKind::HostNetns`] (the Linux veth-by-PID path).
1378 /// VM runtimes with no host netns (macOS VZ-Linux) override this to
1379 /// [`OverlayAttachKind::GuestManaged`], which makes the service layer push a
1380 /// host-allocated overlay config into the guest via [`push_overlay_config`]
1381 /// instead of attaching a veth by PID.
1382 ///
1383 /// [`push_overlay_config`]: Runtime::push_overlay_config
1384 fn overlay_attach_kind(&self) -> OverlayAttachKind {
1385 OverlayAttachKind::HostNetns
1386 }
1387
1388 /// Per-container variant of [`Runtime::overlay_attach_kind`]. A composite
1389 /// runtime that multiplexes several backends (e.g. Windows HCS + WSL2, or
1390 /// macOS Seatbelt + VZ) overrides this to return the attach kind of the
1391 /// specific backend that owns `id` (resolved via an async backend lookup),
1392 /// so a Seatbelt container reports [`OverlayAttachKind::HostProxy`] while a
1393 /// co-located VZ-Linux container reports [`OverlayAttachKind::GuestManaged`].
1394 /// Defaults to the runtime-wide [`Runtime::overlay_attach_kind`].
1395 async fn overlay_attach_kind_for(&self, _id: &ContainerId) -> OverlayAttachKind {
1396 self.overlay_attach_kind()
1397 }
1398
1399 /// Push a host-allocated overlay (`WireGuard`) config into a guest that manages
1400 /// its own overlay interface ([`OverlayAttachKind::GuestManaged`]).
1401 ///
1402 /// The service layer obtains `config` from overlayd (which allocated the
1403 /// address + keypair and registered the public key in the mesh) and calls
1404 /// this so the runtime can deliver it to the guest (over vsock) and bring up
1405 /// the in-guest interface. Runtimes that attach by netns/PID never call this
1406 /// and use the default, which errors.
1407 async fn push_overlay_config(
1408 &self,
1409 _id: &ContainerId,
1410 _config: &zlayer_types::overlayd::GuestOverlayConfig,
1411 ) -> Result<()> {
1412 Err(AgentError::Unsupported(
1413 "push_overlay_config is not supported by this runtime".to_string(),
1414 ))
1415 }
1416
1417 /// Record the overlay `/32` overlayd assigned to a host-shared container and
1418 /// stand up userspace forwarders binding `<overlay_ip>:<published_port>` to
1419 /// the container's local delivery address. Called by the service/job layer
1420 /// AFTER a successful [`OverlayAttachKind::HostProxy`] attach, once the
1421 /// node-side `utun` alias for `overlay_ip` is live. Default no-op:
1422 /// host-netns (Linux) and guest-managed (VZ-Linux) runtimes surface their
1423 /// overlay IP by other means.
1424 async fn attach_overlay_ip(&self, _id: &ContainerId, _overlay_ip: IpAddr) -> Result<()> {
1425 Ok(())
1426 }
1427
1428 /// Tear down the forwarders and clear the stored overlay IP for `id`.
1429 /// Called on container teardown / scale-down. Default no-op.
1430 async fn detach_overlay_ip(&self, _id: &ContainerId) -> Result<()> {
1431 Ok(())
1432 }
1433
1434 /// Get the IP address of a container
1435 ///
1436 /// Returns:
1437 /// - `Ok(Some(ip))` if the container has a known IP address
1438 /// - `Ok(None)` if the container exists but has no IP assigned yet
1439 /// - `Err` if the container doesn't exist or there's an error
1440 ///
1441 /// Used for proxy backend registration when overlay networking is unavailable.
1442 async fn get_container_ip(&self, id: &ContainerId) -> Result<Option<IpAddr>>;
1443
1444 /// Get a runtime-assigned port override for a container.
1445 ///
1446 /// Returns:
1447 /// - `Ok(Some(port))` if the runtime assigned a dynamic port to this container
1448 /// - `Ok(None)` if the container should use the spec-declared endpoint port
1449 ///
1450 /// This exists for runtimes where all containers share the host network stack
1451 /// (e.g., macOS sandbox). Without network namespaces, multiple replicas of
1452 /// the same service would conflict on the same port. The runtime assigns
1453 /// each replica a unique port and passes it via the `PORT` environment variable.
1454 /// The proxy then routes to `container_ip:override_port` instead of
1455 /// `container_ip:spec_port`.
1456 ///
1457 /// Runtimes with per-container networking (overlay, VMs, Docker) return `None`.
1458 async fn get_container_port_override(&self, _id: &ContainerId) -> Result<Option<u16>> {
1459 Ok(None)
1460 }
1461
1462 /// Get the HCN namespace GUID of a Windows container.
1463 ///
1464 /// Windows-only. Linux/macOS runtimes have no HCN namespace concept and
1465 /// return `Ok(None)`. The `HcsRuntime` overrides this to return the
1466 /// namespace GUID attached during `create_container`; `OverlayManager`
1467 /// then uses the GUID to register the container's assigned overlay IP
1468 /// against the right HCN compartment (analogous to how Linux uses PID
1469 /// to enter the netns via `/proc/{pid}/ns/net`).
1470 #[cfg(target_os = "windows")]
1471 async fn get_container_namespace_id(
1472 &self,
1473 _id: &ContainerId,
1474 ) -> Result<Option<windows::core::GUID>> {
1475 Ok(None)
1476 }
1477
1478 /// Sync all named volumes associated with this container to S3.
1479 ///
1480 /// Called after a container is stopped but before it is removed, giving
1481 /// the runtime a chance to flush persistent volume data to remote storage.
1482 ///
1483 /// The default implementation is a no-op. Runtimes that support S3-backed
1484 /// volume sync (e.g., Youki with the `s3` feature) override this.
1485 async fn sync_container_volumes(&self, _id: &ContainerId) -> Result<()> {
1486 Ok(())
1487 }
1488
1489 /// Stream container logs as raw byte chunks tagged with their channel.
1490 ///
1491 /// Mirrors the `GET /containers/{id}/logs` endpoint of the Docker Engine
1492 /// API: callers can request `follow`, `tail`, `since`/`until` time
1493 /// windows, per-channel filtering (`stdout` / `stderr`), and inline
1494 /// timestamps via [`LogsStreamOptions`]. Backends that demultiplex
1495 /// Docker's stdcopy framing emit one [`LogChunk`] per frame; line-based
1496 /// runtimes emit one chunk per line.
1497 ///
1498 /// The default implementation returns [`AgentError::Unsupported`].
1499 /// Concrete runtimes override this with backend-specific streaming
1500 /// (bollard's `logs` for Docker, log-file tailing for Youki/HCS, etc.).
1501 /// The stream is `'static` so HTTP handlers can drive it independently
1502 /// of the trait-method borrow.
1503 async fn logs_stream(&self, _id: &ContainerId, _opts: LogsStreamOptions) -> Result<LogsStream> {
1504 Err(AgentError::Unsupported(
1505 "logs_stream is not supported by this runtime".into(),
1506 ))
1507 }
1508
1509 /// Stream periodic resource-usage samples for a container.
1510 ///
1511 /// Mirrors the streaming form of `GET /containers/{id}/stats` in the
1512 /// Docker Engine API: each yielded [`StatsSample`] is one full snapshot
1513 /// of CPU / memory / network / block-IO / pids counters at the moment
1514 /// it was taken. Sampling cadence is backend-defined (Docker emits one
1515 /// sample per second by default).
1516 ///
1517 /// The default implementation returns [`AgentError::Unsupported`].
1518 /// Backends that can produce this data implement it directly: Docker
1519 /// via bollard's `stats`, Youki by polling cgroup stat files,
1520 /// `MockRuntime` by emitting a deterministic single sample.
1521 async fn stats_stream(&self, _id: &ContainerId) -> Result<StatsStream> {
1522 Err(AgentError::Unsupported(
1523 "stats_stream is not supported by this runtime".into(),
1524 ))
1525 }
1526
1527 /// Pull an image, streaming progress events as layers are downloaded.
1528 ///
1529 /// Mirrors the streaming form of `POST /images/create` in the Docker
1530 /// Engine API. Backends emit a series of [`PullProgress::Status`]
1531 /// events for in-flight layers, followed by exactly one
1532 /// [`PullProgress::Done`] event on success. Errors that occur mid-pull
1533 /// surface as `Err` items on the stream and terminate it.
1534 ///
1535 /// `auth` carries inline credentials for this pull. When `None`, the
1536 /// runtime falls back to its credential-store lookup keyed by registry
1537 /// hostname (matching the semantics of [`Runtime::pull_image_with_policy`]).
1538 ///
1539 /// Backends override this with their native streaming pull
1540 /// (bollard's `create_image` for Docker, `zlayer-registry` for Youki).
1541 ///
1542 /// The default implementation performs a BLOCKING pull via
1543 /// [`Runtime::pull_image_with_policy`] and then synthesizes a minimal
1544 /// Docker-style progress stream (a `Status` line + a terminal `Done`). This
1545 /// keeps the streaming `POST /images/create` (e.g. `docker pull` through
1546 /// the `zlayer-docker` compat socket) working on runtimes that lack a native
1547 /// streaming pull — notably the macOS sandbox. The streaming form means
1548 /// "make this image available now", so it pulls if not already present.
1549 async fn pull_image_stream(
1550 &self,
1551 image: &str,
1552 auth: Option<&RegistryAuth>,
1553 ) -> Result<PullProgressStream> {
1554 self.pull_image_with_policy(
1555 image,
1556 PullPolicy::IfNotPresent,
1557 auth,
1558 zlayer_spec::SourcePolicy::default(),
1559 )
1560 .await?;
1561 let reference = image.to_string();
1562 let events: Vec<Result<PullProgress>> = vec![
1563 Ok(PullProgress::Status {
1564 id: None,
1565 status: format!("Pulling from {reference}"),
1566 progress: None,
1567 current: None,
1568 total: None,
1569 }),
1570 Ok(PullProgress::Done {
1571 reference,
1572 digest: None,
1573 }),
1574 ];
1575 Ok(Box::pin(futures_util::stream::iter(events)))
1576 }
1577
1578 /// List all images managed by this runtime's image storage.
1579 ///
1580 /// The default implementation returns `AgentError::Unsupported` — individual
1581 /// runtimes override this with backend-specific logic (bollard for Docker,
1582 /// zlayer-registry cache walk for Youki, etc.).
1583 async fn list_images(&self) -> Result<Vec<ImageInfo>> {
1584 Err(AgentError::Unsupported(
1585 "list_images is not supported by this runtime".into(),
1586 ))
1587 }
1588
1589 /// Remove an image by reference from local storage.
1590 ///
1591 /// When `force` is true, also removes the image even when other containers
1592 /// reference it. The default implementation returns `AgentError::Unsupported`.
1593 async fn remove_image(&self, _image: &str, _force: bool) -> Result<()> {
1594 Err(AgentError::Unsupported(
1595 "remove_image is not supported by this runtime".into(),
1596 ))
1597 }
1598
1599 /// Prune dangling / unused images from local storage.
1600 ///
1601 /// Returns a [`PruneResult`] describing what was removed. The default
1602 /// implementation returns `AgentError::Unsupported`.
1603 async fn prune_images(&self) -> Result<PruneResult> {
1604 Err(AgentError::Unsupported(
1605 "prune_images is not supported by this runtime".into(),
1606 ))
1607 }
1608
1609 /// Send a signal to a running container.
1610 ///
1611 /// When `signal` is `None`, the runtime sends `SIGKILL` (matching Docker's
1612 /// `docker kill` default). Backends validate the signal name and reject
1613 /// anything outside the standard POSIX set (`SIGKILL`, `SIGTERM`, `SIGINT`,
1614 /// `SIGHUP`, `SIGUSR1`, `SIGUSR2`).
1615 ///
1616 /// Used by `POST /api/v1/containers/{id}/kill` and Docker-compat
1617 /// `docker kill`. The default implementation returns
1618 /// [`AgentError::Unsupported`].
1619 async fn kill_container(&self, _id: &ContainerId, _signal: Option<&str>) -> Result<()> {
1620 Err(AgentError::Unsupported(
1621 "kill_container is not supported by this runtime".into(),
1622 ))
1623 }
1624
1625 /// Write a chunk of stdin to a running container's main process.
1626 ///
1627 /// Powers the host→guest direction of interactive (`-it`) sessions: the
1628 /// daemon's `POST /api/v1/containers/{id}/stdin` endpoint forwards raw
1629 /// terminal bytes here, which the runtime relays to the workload (for the
1630 /// macOS VZ-Linux backend, as `Msg::Stdin` frames to the in-guest agent).
1631 ///
1632 /// The default implementation returns [`AgentError::Unsupported`] so
1633 /// non-interactive backends keep compiling.
1634 async fn write_stdin(&self, _id: &ContainerId, _data: &[u8]) -> Result<()> {
1635 Err(AgentError::Unsupported(
1636 "write_stdin is not supported by this runtime".into(),
1637 ))
1638 }
1639
1640 /// Signal end-of-input (close stdin) for a running container.
1641 ///
1642 /// Powers Ctrl-D / detach for interactive sessions: the daemon's
1643 /// `DELETE /api/v1/containers/{id}/stdin` endpoint calls this, which causes
1644 /// the runtime to stop forwarding stdin and emit a final close marker (for
1645 /// the macOS VZ-Linux backend, `Msg::StdinEof` to the in-guest agent).
1646 ///
1647 /// The default implementation returns [`AgentError::Unsupported`].
1648 async fn close_stdin(&self, _id: &ContainerId) -> Result<()> {
1649 Err(AgentError::Unsupported(
1650 "close_stdin is not supported by this runtime".into(),
1651 ))
1652 }
1653
1654 /// Create a new tag pointing at an existing image.
1655 ///
1656 /// `source` is the reference to an already-cached image. `target` is the
1657 /// new reference to create — it must be a full reference (repository + tag).
1658 ///
1659 /// Used by `POST /api/v1/images/tag` and Docker-compat `docker tag`. The
1660 /// default implementation returns [`AgentError::Unsupported`].
1661 async fn tag_image(&self, _source: &str, _target: &str) -> Result<()> {
1662 Err(AgentError::Unsupported(
1663 "tag_image is not supported by this runtime".into(),
1664 ))
1665 }
1666
1667 /// Inspect an image and return a Docker-shaped detail record.
1668 ///
1669 /// Mirrors Docker's `GET /images/{name}/json`. Backends translate their
1670 /// native inspect output into [`ImageInspectInfo`]; the API/Docker
1671 /// compat shim emits the JSON body. The default implementation returns
1672 /// [`AgentError::Unsupported`] so non-Docker backends keep compiling.
1673 async fn inspect_image_native(&self, _image: &str) -> Result<ImageInspectInfo> {
1674 Err(AgentError::Unsupported(
1675 "inspect_image_native is not supported by this runtime".into(),
1676 ))
1677 }
1678
1679 /// Return the parent-layer history for an image.
1680 ///
1681 /// Mirrors Docker's `GET /images/{name}/history`. The default
1682 /// implementation returns [`AgentError::Unsupported`].
1683 async fn image_history(&self, _image: &str) -> Result<Vec<ImageHistoryEntry>> {
1684 Err(AgentError::Unsupported(
1685 "image_history is not supported by this runtime".into(),
1686 ))
1687 }
1688
1689 /// Search a registry for images matching `term`.
1690 ///
1691 /// Mirrors Docker's `GET /images/search`. `limit` caps the number of
1692 /// returned items; `0` means "let the registry decide". The default
1693 /// implementation returns [`AgentError::Unsupported`].
1694 async fn search_images(&self, _term: &str, _limit: u32) -> Result<Vec<ImageSearchResult>> {
1695 Err(AgentError::Unsupported(
1696 "search_images is not supported by this runtime".into(),
1697 ))
1698 }
1699
1700 /// Stream a tar archive containing one or more images.
1701 ///
1702 /// Mirrors Docker's `GET /images/get?names=...`. Multi-image archives
1703 /// dedupe shared layers. The default implementation returns
1704 /// [`AgentError::Unsupported`].
1705 async fn save_images(&self, _names: &[String]) -> Result<ImageExportStream> {
1706 Err(AgentError::Unsupported(
1707 "save_images is not supported by this runtime".into(),
1708 ))
1709 }
1710
1711 /// Load images from a tar archive.
1712 ///
1713 /// Mirrors Docker's `POST /images/load`. `tar_bytes` is the
1714 /// uncompressed (or gzip-compressed) tar produced by [`Self::save_images`].
1715 /// `quiet` suppresses progress events when set. The default
1716 /// implementation returns [`AgentError::Unsupported`].
1717 async fn load_images(
1718 &self,
1719 _tar_bytes: bytes::Bytes,
1720 _quiet: bool,
1721 ) -> Result<LoadProgressStream> {
1722 Err(AgentError::Unsupported(
1723 "load_images is not supported by this runtime".into(),
1724 ))
1725 }
1726
1727 /// Import a single image from a tar root filesystem.
1728 ///
1729 /// Mirrors the `fromSrc=`-mode of Docker's `POST /images/create`.
1730 /// `tar_bytes` is a tar of the root filesystem; `repo`/`tag` are the
1731 /// reference to apply to the resulting image. The default
1732 /// implementation returns [`AgentError::Unsupported`].
1733 async fn import_image(
1734 &self,
1735 _tar_bytes: bytes::Bytes,
1736 _repo: Option<&str>,
1737 _tag: Option<&str>,
1738 ) -> Result<String> {
1739 Err(AgentError::Unsupported(
1740 "import_image is not supported by this runtime".into(),
1741 ))
1742 }
1743
1744 /// Expose the runtime's already-open image store handles: its
1745 /// [`LocalRegistry`](zlayer_registry::LocalRegistry) and shared blob cache.
1746 ///
1747 /// `zlayer-registry`'s on-disk blob cache (ZQL / redb) is **single-process
1748 /// exclusive** — only one open handle may exist per process. The youki
1749 /// runtime opens both at construction and holds them for the daemon's
1750 /// lifetime. Callers that need to write into the *same* store the daemon
1751 /// serves from (notably the `POST /images/import` handler) must reuse these
1752 /// shared `Arc`s rather than opening a second handle, which would fail with
1753 /// "database is locked by another process".
1754 ///
1755 /// Returns `Some((registry, blob_cache))` for runtimes that own an on-disk
1756 /// `zlayer-registry` store (youki, and a composite whose primary owns one).
1757 /// The default returns `None` for runtimes that delegate image storage to
1758 /// an external daemon (Docker/WSL/HCS) or have no store (WASM, mocks); such
1759 /// runtimes keep the import handler on its runtime-backed fallback.
1760 fn image_store_handles(&self) -> Option<ImageStoreHandles> {
1761 None
1762 }
1763
1764 /// Stream a tar archive of the container's filesystem.
1765 ///
1766 /// Mirrors Docker's `GET /containers/{id}/export`. The default
1767 /// implementation returns [`AgentError::Unsupported`].
1768 async fn export_container_fs(&self, _id: &ContainerId) -> Result<ImageExportStream> {
1769 Err(AgentError::Unsupported(
1770 "export_container_fs is not supported by this runtime".into(),
1771 ))
1772 }
1773
1774 /// Commit a container's filesystem state to a new image.
1775 ///
1776 /// Mirrors Docker's `POST /commit?container=...`. `opts` carries the
1777 /// optional repo/tag/comment/author/pause/changes parameters. The
1778 /// default implementation returns [`AgentError::Unsupported`].
1779 async fn commit_container(
1780 &self,
1781 _id: &ContainerId,
1782 _opts: &CommitOptions,
1783 ) -> Result<CommitOutcome> {
1784 Err(AgentError::Unsupported(
1785 "commit_container is not supported by this runtime".into(),
1786 ))
1787 }
1788
1789 /// Return rich inspect details for a container: published ports, attached
1790 /// networks, first IPv4, health, and most-recent exit code.
1791 ///
1792 /// Runtimes implement this by translating the backend's native inspect
1793 /// response (bollard's `ContainerInspectResponse` for Docker) into the
1794 /// runtime-level [`ContainerInspectDetails`] struct. The API layer merges
1795 /// these fields into `ContainerInfo` on `GET /api/v1/containers` and
1796 /// `GET /api/v1/containers/{id}` (§3.15 of `ZLAYER_SDK_FIXES.md`).
1797 ///
1798 /// The default implementation returns [`ContainerInspectDetails::default`]
1799 /// — an all-empty record, which the API layer treats as "this runtime
1800 /// doesn't support rich inspect; skip all the extra fields". This keeps
1801 /// non-Docker runtimes (Youki, WASM, Mock) backwards compatible; they can
1802 /// override this later if they gain equivalent inspect capability.
1803 async fn inspect_detailed(&self, _id: &ContainerId) -> Result<ContainerInspectDetails> {
1804 Ok(ContainerInspectDetails::default())
1805 }
1806
1807 /// Pause all processes in the container by freezing its cgroup.
1808 ///
1809 /// Mirrors Docker's `POST /containers/{id}/pause`. After pause, the
1810 /// container's processes are suspended in the kernel via the cgroup
1811 /// freezer; calls to [`Runtime::container_state`] still report
1812 /// `Running` but no instructions execute until [`Runtime::unpause_container`].
1813 ///
1814 /// The default implementation returns [`AgentError::Unsupported`].
1815 /// Backends override this with their native pause API (bollard's
1816 /// `pause_container` for Docker, libcontainer's `Container::pause` for
1817 /// Youki).
1818 async fn pause_container(&self, _id: &ContainerId) -> Result<()> {
1819 Err(AgentError::Unsupported(
1820 "pause_container is not supported by this runtime".into(),
1821 ))
1822 }
1823
1824 /// Resume a previously-paused container by thawing its cgroup freezer.
1825 ///
1826 /// Mirrors Docker's `POST /containers/{id}/unpause`. The default
1827 /// implementation returns [`AgentError::Unsupported`].
1828 async fn unpause_container(&self, _id: &ContainerId) -> Result<()> {
1829 Err(AgentError::Unsupported(
1830 "unpause_container is not supported by this runtime".into(),
1831 ))
1832 }
1833
1834 /// Update a running container's resource limits and restart policy.
1835 ///
1836 /// Mirrors Docker's `POST /containers/{id}/update`. The fields on
1837 /// [`ContainerResourceUpdate`] are individually optional: backends
1838 /// apply only the fields that are `Some` and leave the rest of the
1839 /// container's runtime configuration untouched. A fully-empty update
1840 /// short-circuits to a no-op (no cgroup writes, no warnings).
1841 ///
1842 /// Returns a [`ContainerUpdateOutcome`] whose `warnings` vector
1843 /// surfaces non-fatal issues (e.g. "kernel memory limit is
1844 /// deprecated", or "real-time scheduling not supported on this
1845 /// kernel"). Empty `warnings` ⇒ every requested field was applied.
1846 ///
1847 /// The default implementation returns [`AgentError::Unsupported`].
1848 /// Backends override this:
1849 ///
1850 /// * Docker — calls bollard's `update_container` with a
1851 /// `ContainerUpdateBody` populated from the input.
1852 /// * Youki — writes the corresponding cgroup v2 files
1853 /// (`cpu.weight`, `memory.max`, `pids.max`, `io.weight`,
1854 /// `cpuset.cpus`, `cpuset.mems`) under
1855 /// `<container_root>/cgroup` and persists the new restart policy
1856 /// in the on-disk supervisor state.
1857 /// * Other backends (WASM, mocks) inherit the `Unsupported` default.
1858 async fn update_container_resources(
1859 &self,
1860 _id: &ContainerId,
1861 _update: &ContainerResourceUpdate,
1862 ) -> Result<ContainerUpdateOutcome> {
1863 Err(AgentError::Unsupported(
1864 "update_container_resources is not supported by this runtime".into(),
1865 ))
1866 }
1867
1868 /// List the processes running inside a container (`docker top`).
1869 ///
1870 /// `ps_args` is forwarded to the runtime as the `ps(1)` argument list when
1871 /// supported; an empty slice means "use the runtime's default columns".
1872 /// Mirrors Docker's `GET /containers/{id}/top?ps_args=<...>`.
1873 ///
1874 /// The default implementation returns [`AgentError::Unsupported`].
1875 async fn top_container(
1876 &self,
1877 _id: &ContainerId,
1878 _ps_args: &[String],
1879 ) -> Result<ContainerTopOutput> {
1880 Err(AgentError::Unsupported(
1881 "top_container is not supported by this runtime".into(),
1882 ))
1883 }
1884
1885 /// Report changes to a container's filesystem since it was created.
1886 ///
1887 /// Mirrors Docker's `GET /containers/{id}/changes`. Returns one
1888 /// [`FilesystemChangeEntry`] per added / modified / deleted path in the
1889 /// container's writable layer. Runtimes that don't compute layer diffs
1890 /// (e.g. youki, which uses raw bundle rootfs without a layered FS) return
1891 /// [`AgentError::Unsupported`].
1892 async fn changes_container(&self, _id: &ContainerId) -> Result<Vec<FilesystemChangeEntry>> {
1893 Err(AgentError::Unsupported(
1894 "changes_container is not supported by this runtime".into(),
1895 ))
1896 }
1897
1898 /// Report the published port mappings for a container.
1899 ///
1900 /// Mirrors Docker's `GET /containers/{id}/port`. Returns one
1901 /// [`PortMappingEntry`] per (container-port, protocol, host-binding)
1902 /// triple. Containers with no published ports return an empty vector.
1903 ///
1904 /// The default implementation returns [`AgentError::Unsupported`].
1905 async fn port_mappings_container(&self, _id: &ContainerId) -> Result<Vec<PortMappingEntry>> {
1906 Err(AgentError::Unsupported(
1907 "port_mappings_container is not supported by this runtime".into(),
1908 ))
1909 }
1910
1911 /// Prune stopped containers from the runtime.
1912 ///
1913 /// Mirrors Docker's `POST /containers/prune`. Returns the IDs of
1914 /// containers that were removed plus the bytes reclaimed. The default
1915 /// implementation returns [`AgentError::Unsupported`].
1916 async fn prune_containers(&self) -> Result<ContainerPruneResult> {
1917 Err(AgentError::Unsupported(
1918 "prune_containers is not supported by this runtime".into(),
1919 ))
1920 }
1921
1922 /// Enumerate all containers known to this runtime, including stopped /
1923 /// exited ones (Docker's `list_containers(all=true)` semantics).
1924 ///
1925 /// Used by `zlayer-api::handlers::standalone_reconcile` on daemon boot
1926 /// to match persisted standalone-container records against the
1927 /// runtime's actual inventory: entries the runtime no longer reports
1928 /// are pruned, surviving entries are re-registered in the
1929 /// `ContainerIdMap`, and runtime containers carrying a
1930 /// `com.zlayer.container_id` label that has no storage match are
1931 /// counted as orphans (logged but otherwise left alone).
1932 ///
1933 /// The default implementation returns an empty list, which makes
1934 /// reconcile degrade to a label-blind pass: storage entries can still
1935 /// be probed individually via [`Runtime::container_state`], but orphan
1936 /// detection is disabled until the backend overrides this method
1937 /// (Docker via `bollard::list_containers`, youki via state-dir walk).
1938 async fn list_containers(&self) -> Result<Vec<RuntimeContainerSummary>> {
1939 Ok(Vec::new())
1940 }
1941
1942 /// Stream a TAR archive of the file or directory at `path` inside the
1943 /// container.
1944 ///
1945 /// Mirrors Docker's `GET /containers/{id}/archive?path=<...>`. The
1946 /// returned [`ArchiveStream`] yields raw `application/x-tar` bytes.
1947 /// Backends decide whether to materialize the archive in memory or
1948 /// stream it on the fly:
1949 ///
1950 /// * Docker — bollard's `download_from_container` produces a chunked
1951 /// stream of TAR bytes; we forward it verbatim.
1952 /// * Youki — a rootfs walk under `<bundle>/rootfs<path>` produces a
1953 /// TAR archive in a worker task and streams it through an mpsc.
1954 /// * Other backends (WASM, mocks) inherit the `Unsupported` default.
1955 ///
1956 /// The default implementation returns [`AgentError::Unsupported`].
1957 async fn archive_get(&self, _id: &ContainerId, _path: &str) -> Result<ArchiveStream> {
1958 Err(AgentError::Unsupported(
1959 "archive_get is not supported by this runtime".into(),
1960 ))
1961 }
1962
1963 /// Extract a TAR archive into the container at `path`.
1964 ///
1965 /// Mirrors Docker's `PUT /containers/{id}/archive?path=<...>`. The
1966 /// runtime must extract `tar_bytes` (an uncompressed TAR archive) into
1967 /// `path` inside the container, honouring [`ArchivePutOptions`].
1968 ///
1969 /// `path` must already exist inside the container and must be a
1970 /// directory; if it does not exist the runtime returns
1971 /// [`AgentError::NotFound`]. Mismatched directory/non-directory
1972 /// replacements with `no_overwrite_dir_non_dir=true` return
1973 /// [`AgentError::InvalidSpec`].
1974 ///
1975 /// The default implementation returns [`AgentError::Unsupported`].
1976 async fn archive_put(
1977 &self,
1978 _id: &ContainerId,
1979 _path: &str,
1980 _tar_bytes: bytes::Bytes,
1981 _opts: ArchivePutOptions,
1982 ) -> Result<()> {
1983 Err(AgentError::Unsupported(
1984 "archive_put is not supported by this runtime".into(),
1985 ))
1986 }
1987
1988 /// Return stat metadata for the file or directory at `path` inside the
1989 /// container.
1990 ///
1991 /// Mirrors Docker's `HEAD /containers/{id}/archive?path=<...>`, which
1992 /// answers with the metadata that `GET /archive` *would* expose without
1993 /// materializing the TAR. Used by `docker cp` and the API layer to
1994 /// short-circuit on missing paths.
1995 ///
1996 /// The default implementation returns [`AgentError::Unsupported`].
1997 async fn archive_head(&self, _id: &ContainerId, _path: &str) -> Result<PathStat> {
1998 Err(AgentError::Unsupported(
1999 "archive_head is not supported by this runtime".into(),
2000 ))
2001 }
2002}
2003
2004/// Validate a signal name for [`Runtime::kill_container`].
2005///
2006/// Accepts both the `SIG`-prefixed form (`"SIGKILL"`) and the bare form
2007/// (`"KILL"`). Returns the canonical uppercase `SIG`-prefixed name on success.
2008///
2009/// # Errors
2010///
2011/// Returns [`AgentError::InvalidSpec`] when `signal` is not one of the
2012/// supported signals: `SIGKILL`, `SIGTERM`, `SIGINT`, `SIGHUP`, `SIGUSR1`,
2013/// `SIGUSR2`.
2014pub fn validate_signal(signal: &str) -> Result<String> {
2015 let trimmed = signal.trim();
2016 if trimmed.is_empty() {
2017 return Err(AgentError::InvalidSpec(
2018 "signal must not be empty".to_string(),
2019 ));
2020 }
2021 let upper = trimmed.to_ascii_uppercase();
2022 let canonical = if upper.starts_with("SIG") {
2023 upper
2024 } else {
2025 format!("SIG{upper}")
2026 };
2027 match canonical.as_str() {
2028 "SIGKILL" | "SIGTERM" | "SIGINT" | "SIGHUP" | "SIGUSR1" | "SIGUSR2" => Ok(canonical),
2029 other => Err(AgentError::InvalidSpec(format!(
2030 "unsupported signal '{other}'; allowed: SIGKILL, SIGTERM, SIGINT, SIGHUP, SIGUSR1, SIGUSR2"
2031 ))),
2032 }
2033}
2034
2035/// Auth context injected into every container so it can talk back to the host
2036/// API without needing external credentials.
2037#[derive(Debug, Clone)]
2038pub struct ContainerAuthContext {
2039 /// Base URL of the `ZLayer` API, e.g. `"http://127.0.0.1:3669"`.
2040 pub api_url: String,
2041 /// JWT signing secret — used to mint per-container tokens at start time.
2042 pub jwt_secret: String,
2043 /// Absolute path of the Unix socket on the host (bind-mounted into Linux
2044 /// containers; added to `writable_dirs` for macOS sandbox).
2045 pub socket_path: String,
2046 /// Sink for persisting/revoking the per-container scoped token. `None`
2047 /// disables persistence (token minted without a `jti`, not revocable).
2048 pub token_sink: Option<std::sync::Arc<dyn crate::auth::ContainerTokenSink>>,
2049 /// Spawns/tears down the per-container Docker Engine API socket for
2050 /// containers that opt into `zlayer.io/docker-socket`. `None` disables the
2051 /// feature (no socket provisioned). Wired in the bin (the agent can't depend
2052 /// on `zlayer-docker`).
2053 pub docker_socket_spawner: Option<std::sync::Arc<dyn crate::auth::DockerSocketSpawner>>,
2054}
2055
2056/// In-memory mock runtime for testing and development.
2057///
2058/// In addition to tracking container lifecycle in memory, the mock exposes
2059/// per-container queues for streaming method outputs so unit tests can
2060/// pre-script the events that [`Runtime::logs_stream`],
2061/// [`Runtime::stats_stream`], and [`Runtime::pull_image_stream`] should
2062/// yield. See [`MockRuntime::enqueue_log_chunk`],
2063/// [`MockRuntime::enqueue_stats_sample`], and
2064/// [`MockRuntime::enqueue_pull_progress`].
2065pub struct MockRuntime {
2066 containers: tokio::sync::RwLock<std::collections::HashMap<ContainerId, Container>>,
2067 /// Pre-scripted log chunks per container. Each call to
2068 /// [`Runtime::logs_stream`] drains this queue in order; once empty, the
2069 /// stream either terminates (when `follow=false`) or hangs forever
2070 /// (when `follow=true`) so tests can exercise both branches.
2071 pub logs_to_yield:
2072 Arc<Mutex<std::collections::HashMap<ContainerId, VecDeque<Result<LogChunk>>>>>,
2073 /// Pre-scripted stats samples per container. Drained in order by
2074 /// [`Runtime::stats_stream`].
2075 pub stats_to_yield:
2076 Arc<Mutex<std::collections::HashMap<ContainerId, VecDeque<Result<StatsSample>>>>>,
2077 /// Pre-scripted pull progress events keyed by image reference. Drained
2078 /// in order by [`Runtime::pull_image_stream`].
2079 pub pull_progress_to_yield:
2080 Arc<Mutex<std::collections::HashMap<String, VecDeque<Result<PullProgress>>>>>,
2081}
2082
2083impl MockRuntime {
2084 #[must_use]
2085 pub fn new() -> Self {
2086 Self {
2087 containers: tokio::sync::RwLock::new(std::collections::HashMap::new()),
2088 logs_to_yield: Arc::new(Mutex::new(std::collections::HashMap::new())),
2089 stats_to_yield: Arc::new(Mutex::new(std::collections::HashMap::new())),
2090 pull_progress_to_yield: Arc::new(Mutex::new(std::collections::HashMap::new())),
2091 }
2092 }
2093
2094 /// Push a single [`LogChunk`] onto the queue for `id`. Subsequent calls to
2095 /// [`Runtime::logs_stream`] will yield enqueued chunks in FIFO order.
2096 pub async fn enqueue_log_chunk(&self, id: &ContainerId, chunk: LogChunk) {
2097 self.logs_to_yield
2098 .lock()
2099 .await
2100 .entry(id.clone())
2101 .or_default()
2102 .push_back(Ok(chunk));
2103 }
2104
2105 /// Push a pre-built error onto the log queue for `id`. The next
2106 /// [`Runtime::logs_stream`] call drains this as the next yielded item.
2107 pub async fn enqueue_log_error(&self, id: &ContainerId, err: AgentError) {
2108 self.logs_to_yield
2109 .lock()
2110 .await
2111 .entry(id.clone())
2112 .or_default()
2113 .push_back(Err(err));
2114 }
2115
2116 /// Push a single [`StatsSample`] onto the queue for `id`. Subsequent calls
2117 /// to [`Runtime::stats_stream`] will yield enqueued samples in FIFO order.
2118 pub async fn enqueue_stats_sample(&self, id: &ContainerId, sample: StatsSample) {
2119 self.stats_to_yield
2120 .lock()
2121 .await
2122 .entry(id.clone())
2123 .or_default()
2124 .push_back(Ok(sample));
2125 }
2126
2127 /// Push a single [`PullProgress`] event onto the queue for `image`.
2128 /// Subsequent calls to [`Runtime::pull_image_stream`] for the same image
2129 /// reference will yield enqueued events in FIFO order.
2130 pub async fn enqueue_pull_progress(&self, image: &str, progress: PullProgress) {
2131 self.pull_progress_to_yield
2132 .lock()
2133 .await
2134 .entry(image.to_string())
2135 .or_default()
2136 .push_back(Ok(progress));
2137 }
2138}
2139
2140impl Default for MockRuntime {
2141 fn default() -> Self {
2142 Self::new()
2143 }
2144}
2145
2146#[async_trait::async_trait]
2147impl Runtime for MockRuntime {
2148 async fn pull_image(&self, _image: &str) -> Result<()> {
2149 self.pull_image_with_policy(
2150 _image,
2151 PullPolicy::IfNotPresent,
2152 None,
2153 zlayer_spec::SourcePolicy::default(),
2154 )
2155 .await
2156 }
2157
2158 async fn pull_image_with_policy(
2159 &self,
2160 _image: &str,
2161 _policy: PullPolicy,
2162 _auth: Option<&RegistryAuth>,
2163 _source: zlayer_spec::SourcePolicy,
2164 ) -> Result<()> {
2165 // Mock: always succeeds
2166 tokio::time::sleep(Duration::from_millis(100)).await;
2167 Ok(())
2168 }
2169
2170 async fn create_container(&self, id: &ContainerId, spec: &ServiceSpec) -> Result<()> {
2171 let mut containers = self.containers.write().await;
2172 containers.insert(
2173 id.clone(),
2174 Container {
2175 id: id.clone(),
2176 image: spec.image.name.to_string(),
2177 state: ContainerState::Pending,
2178 pid: None,
2179 task: None,
2180 overlay_ip: None,
2181 health_monitor: None,
2182 port_override: None,
2183 },
2184 );
2185 Ok(())
2186 }
2187
2188 async fn start_container(&self, id: &ContainerId) -> Result<()> {
2189 let mut containers = self.containers.write().await;
2190 if let Some(container) = containers.get_mut(id) {
2191 container.state = ContainerState::Running;
2192 container.pid = Some(std::process::id()); // Mock PID
2193 }
2194 Ok(())
2195 }
2196
2197 async fn stop_container(&self, id: &ContainerId, _timeout: Duration) -> Result<()> {
2198 let mut containers = self.containers.write().await;
2199 if let Some(container) = containers.get_mut(id) {
2200 container.state = ContainerState::Exited { code: 0 };
2201 }
2202 Ok(())
2203 }
2204
2205 async fn remove_container(&self, id: &ContainerId) -> Result<()> {
2206 let mut containers = self.containers.write().await;
2207 containers.remove(id);
2208 Ok(())
2209 }
2210
2211 async fn container_state(&self, id: &ContainerId) -> Result<ContainerState> {
2212 let containers = self.containers.read().await;
2213 containers
2214 .get(id)
2215 .map(|c| c.state.clone())
2216 .ok_or_else(|| AgentError::NotFound {
2217 container: id.to_string(),
2218 reason: "container not found".to_string(),
2219 })
2220 }
2221
2222 async fn container_logs(&self, _id: &ContainerId, tail: usize) -> Result<Vec<LogEntry>> {
2223 let entries = vec![
2224 LogEntry {
2225 timestamp: chrono::Utc::now(),
2226 stream: LogStream::Stdout,
2227 message: "mock log line 1".to_string(),
2228 source: LogSource::Container("mock".to_string()),
2229 service: None,
2230 deployment: None,
2231 },
2232 LogEntry {
2233 timestamp: chrono::Utc::now(),
2234 stream: LogStream::Stderr,
2235 message: "mock error line".to_string(),
2236 source: LogSource::Container("mock".to_string()),
2237 service: None,
2238 deployment: None,
2239 },
2240 ];
2241 let skip = entries.len().saturating_sub(tail);
2242 Ok(entries.into_iter().skip(skip).collect())
2243 }
2244
2245 async fn exec(&self, _id: &ContainerId, cmd: &[String]) -> Result<(i32, String, String)> {
2246 Ok((0, cmd.join(" "), String::new()))
2247 }
2248
2249 async fn get_container_stats(&self, id: &ContainerId) -> Result<ContainerStats> {
2250 // Mock: return dummy stats
2251 let containers = self.containers.read().await;
2252 if containers.contains_key(id) {
2253 Ok(ContainerStats {
2254 cpu_usage_usec: 1_000_000, // 1 second
2255 memory_bytes: 50 * 1024 * 1024, // 50 MB
2256 memory_limit: 256 * 1024 * 1024, // 256 MB
2257 timestamp: std::time::Instant::now(),
2258 })
2259 } else {
2260 Err(AgentError::NotFound {
2261 container: id.to_string(),
2262 reason: "container not found".to_string(),
2263 })
2264 }
2265 }
2266
2267 async fn wait_container(&self, id: &ContainerId) -> Result<i32> {
2268 // Mock: simulate waiting for container to exit
2269 let containers = self.containers.read().await;
2270 if let Some(container) = containers.get(id) {
2271 match &container.state {
2272 ContainerState::Exited { code } => Ok(*code),
2273 ContainerState::Failed { .. } => Ok(1),
2274 _ => {
2275 // Simulate a brief wait and then return success
2276 drop(containers);
2277 tokio::time::sleep(Duration::from_millis(50)).await;
2278 Ok(0)
2279 }
2280 }
2281 } else {
2282 Err(AgentError::NotFound {
2283 container: id.to_string(),
2284 reason: "container not found".to_string(),
2285 })
2286 }
2287 }
2288
2289 async fn get_logs(&self, id: &ContainerId) -> Result<Vec<LogEntry>> {
2290 // Mock: return dummy structured log entries
2291 let containers = self.containers.read().await;
2292 if containers.contains_key(id) {
2293 let container_name = id.to_string();
2294 Ok(vec![
2295 LogEntry {
2296 timestamp: chrono::Utc::now(),
2297 stream: LogStream::Stdout,
2298 message: format!("[{container_name}] Container started"),
2299 source: LogSource::Container(container_name.clone()),
2300 service: None,
2301 deployment: None,
2302 },
2303 LogEntry {
2304 timestamp: chrono::Utc::now(),
2305 stream: LogStream::Stdout,
2306 message: format!("[{container_name}] Executing command..."),
2307 source: LogSource::Container(container_name.clone()),
2308 service: None,
2309 deployment: None,
2310 },
2311 LogEntry {
2312 timestamp: chrono::Utc::now(),
2313 stream: LogStream::Stdout,
2314 message: format!("[{container_name}] Command completed successfully"),
2315 source: LogSource::Container(container_name),
2316 service: None,
2317 deployment: None,
2318 },
2319 ])
2320 } else {
2321 Err(AgentError::NotFound {
2322 container: id.to_string(),
2323 reason: "container not found".to_string(),
2324 })
2325 }
2326 }
2327
2328 async fn get_container_pid(&self, id: &ContainerId) -> Result<Option<u32>> {
2329 let containers = self.containers.read().await;
2330 if let Some(container) = containers.get(id) {
2331 Ok(container.pid)
2332 } else {
2333 Err(AgentError::NotFound {
2334 container: id.to_string(),
2335 reason: "container not found".to_string(),
2336 })
2337 }
2338 }
2339
2340 async fn get_container_ip(&self, id: &ContainerId) -> Result<Option<IpAddr>> {
2341 let containers = self.containers.read().await;
2342 if containers.contains_key(id) {
2343 // Mock: deterministic IP based on replica number (172.17.0.{replica+2})
2344 #[allow(clippy::cast_possible_truncation)]
2345 let last_octet = (id.replica + 2) as u8;
2346 Ok(Some(IpAddr::V4(std::net::Ipv4Addr::new(
2347 172, 17, 0, last_octet,
2348 ))))
2349 } else {
2350 Err(AgentError::NotFound {
2351 container: id.to_string(),
2352 reason: "container not found".to_string(),
2353 })
2354 }
2355 }
2356
2357 async fn list_images(&self) -> Result<Vec<ImageInfo>> {
2358 Ok(Vec::new())
2359 }
2360
2361 async fn remove_image(&self, _image: &str, _force: bool) -> Result<()> {
2362 Ok(())
2363 }
2364
2365 async fn prune_images(&self) -> Result<PruneResult> {
2366 Ok(PruneResult::default())
2367 }
2368
2369 async fn kill_container(&self, id: &ContainerId, signal: Option<&str>) -> Result<()> {
2370 // Validate signal even in the mock so callers exercise the same error
2371 // path. Default to SIGKILL when omitted.
2372 let _canonical = validate_signal(signal.unwrap_or("SIGKILL"))?;
2373 let mut containers = self.containers.write().await;
2374 let container = containers.get_mut(id).ok_or_else(|| AgentError::NotFound {
2375 container: id.to_string(),
2376 reason: "container not found".to_string(),
2377 })?;
2378 container.state = ContainerState::Exited { code: 137 };
2379 Ok(())
2380 }
2381
2382 async fn tag_image(&self, _source: &str, _target: &str) -> Result<()> {
2383 // The in-memory mock doesn't store images; treat tag as a no-op success.
2384 Ok(())
2385 }
2386
2387 async fn logs_stream(&self, id: &ContainerId, opts: LogsStreamOptions) -> Result<LogsStream> {
2388 use futures_util::StreamExt;
2389
2390 // Drain the per-container queue once at call time; the resulting
2391 // iterator is owned by the stream so the lock isn't held while the
2392 // consumer reads.
2393 let queued: Vec<Result<LogChunk>> = {
2394 let mut guard = self.logs_to_yield.lock().await;
2395 guard.remove(id).map(Vec::from).unwrap_or_default()
2396 };
2397 let head = futures_util::stream::iter(queued);
2398 if opts.follow {
2399 // After the queued items are drained, hang forever so callers can
2400 // exercise the "still-following, no more data" branch and cancel
2401 // by dropping the stream.
2402 let tail = futures_util::stream::pending::<Result<LogChunk>>();
2403 Ok(Box::pin(head.chain(tail)))
2404 } else {
2405 Ok(Box::pin(head))
2406 }
2407 }
2408
2409 async fn stats_stream(&self, id: &ContainerId) -> Result<StatsStream> {
2410 let queued: Vec<Result<StatsSample>> = {
2411 let mut guard = self.stats_to_yield.lock().await;
2412 guard.remove(id).map(Vec::from).unwrap_or_default()
2413 };
2414 // `stats_stream` has no `follow` flag on the trait. The mock yields
2415 // exactly what was pre-loaded and then closes — tests that want a
2416 // forever-pending stream can simulate it by holding the receiver and
2417 // never enqueueing more. Closing on drain keeps tests bounded so a
2418 // forgotten consumer never deadlocks the test runner.
2419 Ok(Box::pin(futures_util::stream::iter(queued)))
2420 }
2421
2422 async fn pull_image_stream(
2423 &self,
2424 image: &str,
2425 _auth: Option<&RegistryAuth>,
2426 ) -> Result<PullProgressStream> {
2427 let queued: Vec<Result<PullProgress>> = {
2428 let mut guard = self.pull_progress_to_yield.lock().await;
2429 guard.remove(image).map(Vec::from).unwrap_or_default()
2430 };
2431 // Pulls are inherently bounded: the real backends emit a final `Done`
2432 // event and close the stream. The mock follows the same shape — it
2433 // just yields whatever the test pre-loaded and ends.
2434 Ok(Box::pin(futures_util::stream::iter(queued)))
2435 }
2436}
2437
2438#[cfg(test)]
2439mod tests {
2440 use super::*;
2441
2442 #[tokio::test]
2443 async fn test_mock_runtime() {
2444 let runtime = MockRuntime::new();
2445 let id = ContainerId::new("test".to_string(), 1);
2446
2447 runtime.pull_image("test:latest").await.unwrap();
2448 runtime.create_container(&id, &mock_spec()).await.unwrap();
2449 runtime.start_container(&id).await.unwrap();
2450
2451 let state = runtime.container_state(&id).await.unwrap();
2452 assert_eq!(state, ContainerState::Running);
2453 }
2454
2455 #[test]
2456 fn parse_display_round_trips_cluster_shape() {
2457 // The exact id the deployment-run path surfaces: service=alpine,
2458 // role=default, replica=1, node_id=1 → `alpine-default-1-on-1`.
2459 let cid = ContainerId::with_role_and_node("alpine", 1, "default", 1);
2460 let s = cid.to_string();
2461 assert_eq!(s, "alpine-default-1-on-1");
2462 assert_eq!(ContainerId::parse_display(&s), Some(cid));
2463 }
2464
2465 #[test]
2466 fn parse_display_round_trips_legacy_shape() {
2467 let cid = ContainerId::new("alpine", 3);
2468 let s = cid.to_string();
2469 assert_eq!(s, "alpine-rep-3");
2470 assert_eq!(ContainerId::parse_display(&s), Some(cid));
2471 }
2472
2473 #[test]
2474 fn parse_display_handles_hyphenated_service() {
2475 // A service name containing `-` must round-trip: the parser anchors on
2476 // the rightmost structural markers, not a left-to-right split.
2477 let cid = ContainerId::with_role_and_node("my-web-app", 2, "read", 7);
2478 let s = cid.to_string();
2479 assert_eq!(s, "my-web-app-read-2-on-7");
2480 assert_eq!(ContainerId::parse_display(&s), Some(cid));
2481 }
2482
2483 #[test]
2484 fn parse_display_rejects_non_ids() {
2485 // A 64-char hex id, a bare name, and obvious garbage must not parse.
2486 assert_eq!(ContainerId::parse_display("alpine"), None);
2487 assert_eq!(
2488 ContainerId::parse_display(
2489 "0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef"
2490 ),
2491 None
2492 );
2493 assert_eq!(ContainerId::parse_display("alpine-default-x-on-1"), None);
2494 assert_eq!(ContainerId::parse_display("alpine-default-1-on-y"), None);
2495 assert_eq!(ContainerId::parse_display(""), None);
2496 }
2497
2498 #[test]
2499 fn validate_signal_accepts_known_signals() {
2500 // SIG-prefixed form
2501 assert_eq!(validate_signal("SIGKILL").unwrap(), "SIGKILL");
2502 assert_eq!(validate_signal("SIGTERM").unwrap(), "SIGTERM");
2503 assert_eq!(validate_signal("SIGINT").unwrap(), "SIGINT");
2504 assert_eq!(validate_signal("SIGHUP").unwrap(), "SIGHUP");
2505 assert_eq!(validate_signal("SIGUSR1").unwrap(), "SIGUSR1");
2506 assert_eq!(validate_signal("SIGUSR2").unwrap(), "SIGUSR2");
2507
2508 // Bare form (no "SIG" prefix) should be canonicalised.
2509 assert_eq!(validate_signal("KILL").unwrap(), "SIGKILL");
2510 assert_eq!(validate_signal("term").unwrap(), "SIGTERM");
2511 // Whitespace around the name is tolerated.
2512 assert_eq!(validate_signal(" INT ").unwrap(), "SIGINT");
2513 }
2514
2515 #[test]
2516 fn validate_signal_rejects_unknown_or_empty() {
2517 assert!(matches!(
2518 validate_signal(""),
2519 Err(AgentError::InvalidSpec(_))
2520 ));
2521 assert!(matches!(
2522 validate_signal(" "),
2523 Err(AgentError::InvalidSpec(_))
2524 ));
2525 assert!(matches!(
2526 validate_signal("SIGSEGV"),
2527 Err(AgentError::InvalidSpec(_))
2528 ));
2529 assert!(matches!(
2530 validate_signal("NOPE"),
2531 Err(AgentError::InvalidSpec(_))
2532 ));
2533 // Signals outside the POSIX allowlist are rejected even if real.
2534 assert!(matches!(
2535 validate_signal("SIGPIPE"),
2536 Err(AgentError::InvalidSpec(_))
2537 ));
2538 }
2539
2540 #[tokio::test]
2541 async fn mock_kill_container_defaults_to_sigkill() {
2542 let runtime = MockRuntime::new();
2543 let id = ContainerId::new("kill-me".to_string(), 0);
2544 runtime.create_container(&id, &mock_spec()).await.unwrap();
2545 runtime.start_container(&id).await.unwrap();
2546
2547 // `None` -> defaults to SIGKILL; returns Ok and marks the container
2548 // as exited.
2549 runtime.kill_container(&id, None).await.unwrap();
2550 let state = runtime.container_state(&id).await.unwrap();
2551 assert!(
2552 matches!(state, ContainerState::Exited { code: 137 }),
2553 "expected Exited(137), got {state:?}"
2554 );
2555 }
2556
2557 #[test]
2558 fn wait_reason_serializes_as_snake_case() {
2559 assert_eq!(
2560 serde_json::to_string(&WaitReason::Exited).unwrap(),
2561 "\"exited\""
2562 );
2563 assert_eq!(
2564 serde_json::to_string(&WaitReason::Signal).unwrap(),
2565 "\"signal\""
2566 );
2567 assert_eq!(
2568 serde_json::to_string(&WaitReason::OomKilled).unwrap(),
2569 "\"oom_killed\""
2570 );
2571 assert_eq!(
2572 serde_json::to_string(&WaitReason::RuntimeError).unwrap(),
2573 "\"runtime_error\""
2574 );
2575 }
2576
2577 #[test]
2578 fn wait_reason_deserialize_roundtrip() {
2579 for variant in [
2580 WaitReason::Exited,
2581 WaitReason::Signal,
2582 WaitReason::OomKilled,
2583 WaitReason::RuntimeError,
2584 ] {
2585 let s = serde_json::to_string(&variant).unwrap();
2586 let back: WaitReason = serde_json::from_str(&s).unwrap();
2587 assert_eq!(variant, back, "roundtrip failed for {variant:?}");
2588 }
2589 }
2590
2591 #[test]
2592 fn signal_name_from_exit_code_known_signals() {
2593 assert_eq!(signal_name_from_exit_code(137).as_deref(), Some("SIGKILL"));
2594 assert_eq!(signal_name_from_exit_code(143).as_deref(), Some("SIGTERM"));
2595 assert_eq!(signal_name_from_exit_code(130).as_deref(), Some("SIGINT"));
2596 assert_eq!(signal_name_from_exit_code(129).as_deref(), Some("SIGHUP"));
2597 assert_eq!(signal_name_from_exit_code(139).as_deref(), Some("SIGSEGV"));
2598 }
2599
2600 #[test]
2601 fn signal_name_from_exit_code_handles_unknown_and_normal() {
2602 // Normal exits (<= 128) return None.
2603 assert_eq!(signal_name_from_exit_code(0), None);
2604 assert_eq!(signal_name_from_exit_code(1), None);
2605 assert_eq!(signal_name_from_exit_code(128), None);
2606
2607 // Unknown signals produce a stable string form.
2608 assert_eq!(
2609 signal_name_from_exit_code(128 + 99).as_deref(),
2610 Some("signal_99")
2611 );
2612 }
2613
2614 #[tokio::test]
2615 async fn default_wait_outcome_delegates_to_wait_container() {
2616 let runtime = MockRuntime::new();
2617 let id = ContainerId::new("wait-test".to_string(), 0);
2618 runtime.create_container(&id, &mock_spec()).await.unwrap();
2619 runtime.start_container(&id).await.unwrap();
2620
2621 let outcome = runtime.wait_outcome(&id).await.unwrap();
2622 // MockRuntime::wait_container returns 0 for running containers.
2623 assert_eq!(outcome.exit_code, 0);
2624 assert_eq!(outcome.reason, WaitReason::Exited);
2625 assert!(outcome.signal.is_none());
2626 assert!(outcome.finished_at.is_none());
2627 }
2628
2629 #[tokio::test]
2630 async fn mock_kill_container_rejects_bogus_signal() {
2631 let runtime = MockRuntime::new();
2632 let id = ContainerId::new("kill-me".to_string(), 0);
2633 runtime.create_container(&id, &mock_spec()).await.unwrap();
2634 runtime.start_container(&id).await.unwrap();
2635
2636 let err = runtime
2637 .kill_container(&id, Some("SIGFOO"))
2638 .await
2639 .unwrap_err();
2640 assert!(
2641 matches!(err, AgentError::InvalidSpec(_)),
2642 "expected InvalidSpec, got {err:?}"
2643 );
2644 }
2645
2646 // The default trait impls of `logs_stream` and `stats_stream` still return
2647 // `AgentError::Unsupported`; `pull_image_stream` now performs a blocking
2648 // pull and synthesizes a Status+Done progress stream. `MockRuntime`
2649 // overrides all three so tests can pre-script stream output (see
2650 // `mock_logs_stream_yields_queued_items_in_order` and friends below).
2651 // A trivial `BareRuntime` exercises the default trait impls without
2652 // dragging in MockRuntime's overrides.
2653
2654 /// Minimal `Runtime` implementation used to exercise the default trait
2655 /// impls of `logs_stream` / `stats_stream` / `pull_image_stream`. Most
2656 /// methods panic, but `pull_image_with_policy` returns `Ok(())` so the
2657 /// default `pull_image_stream` (which delegates to it) can be exercised.
2658 struct BareRuntime;
2659
2660 #[async_trait::async_trait]
2661 impl Runtime for BareRuntime {
2662 async fn pull_image(&self, _image: &str) -> Result<()> {
2663 unimplemented!()
2664 }
2665 async fn pull_image_with_policy(
2666 &self,
2667 _image: &str,
2668 _policy: PullPolicy,
2669 _auth: Option<&RegistryAuth>,
2670 _source: zlayer_spec::SourcePolicy,
2671 ) -> Result<()> {
2672 // The default `pull_image_stream` delegates here; return Ok so the
2673 // streaming default can be exercised without a real registry.
2674 Ok(())
2675 }
2676 async fn create_container(&self, _id: &ContainerId, _spec: &ServiceSpec) -> Result<()> {
2677 unimplemented!()
2678 }
2679 async fn start_container(&self, _id: &ContainerId) -> Result<()> {
2680 unimplemented!()
2681 }
2682 async fn stop_container(&self, _id: &ContainerId, _timeout: Duration) -> Result<()> {
2683 unimplemented!()
2684 }
2685 async fn remove_container(&self, _id: &ContainerId) -> Result<()> {
2686 unimplemented!()
2687 }
2688 async fn container_state(&self, _id: &ContainerId) -> Result<ContainerState> {
2689 unimplemented!()
2690 }
2691 async fn container_logs(&self, _id: &ContainerId, _tail: usize) -> Result<Vec<LogEntry>> {
2692 unimplemented!()
2693 }
2694 async fn exec(&self, _id: &ContainerId, _cmd: &[String]) -> Result<(i32, String, String)> {
2695 unimplemented!()
2696 }
2697 async fn get_container_stats(&self, _id: &ContainerId) -> Result<ContainerStats> {
2698 unimplemented!()
2699 }
2700 async fn wait_container(&self, _id: &ContainerId) -> Result<i32> {
2701 unimplemented!()
2702 }
2703 async fn get_logs(&self, _id: &ContainerId) -> Result<Vec<LogEntry>> {
2704 unimplemented!()
2705 }
2706 async fn get_container_pid(&self, _id: &ContainerId) -> Result<Option<u32>> {
2707 unimplemented!()
2708 }
2709 async fn get_container_ip(&self, _id: &ContainerId) -> Result<Option<IpAddr>> {
2710 unimplemented!()
2711 }
2712 }
2713
2714 #[tokio::test]
2715 async fn default_logs_stream_is_unsupported() {
2716 let runtime = BareRuntime;
2717 let id = ContainerId::new("stream-test".to_string(), 0);
2718 // The success-side `LogsStream` is not `Debug`, so we can't call
2719 // `unwrap_err`; pattern-match on the Result directly instead.
2720 match runtime.logs_stream(&id, LogsStreamOptions::default()).await {
2721 Err(AgentError::Unsupported(_)) => {}
2722 Err(other) => panic!("expected Unsupported, got {other:?}"),
2723 Ok(_) => panic!("expected Err(Unsupported), got Ok"),
2724 }
2725 }
2726
2727 #[tokio::test]
2728 async fn default_stats_stream_is_unsupported() {
2729 let runtime = BareRuntime;
2730 let id = ContainerId::new("stream-test".to_string(), 0);
2731 match runtime.stats_stream(&id).await {
2732 Err(AgentError::Unsupported(_)) => {}
2733 Err(other) => panic!("expected Unsupported, got {other:?}"),
2734 Ok(_) => panic!("expected Err(Unsupported), got Ok"),
2735 }
2736 }
2737
2738 #[tokio::test]
2739 async fn default_pull_image_stream_synthesizes_progress() {
2740 use futures_util::StreamExt as _;
2741
2742 // The default `pull_image_stream` now performs a blocking pull (via
2743 // `pull_image_with_policy`, which `BareRuntime` answers with `Ok`) and
2744 // synthesizes a Status + Done progress stream.
2745 let runtime = BareRuntime;
2746 let stream = runtime
2747 .pull_image_stream("alpine:latest", None)
2748 .await
2749 .expect("default pull_image_stream should succeed when the pull succeeds");
2750 let events: Vec<_> = stream.collect().await;
2751 assert_eq!(events.len(), 2, "expected a Status then a Done event");
2752 assert!(
2753 matches!(events[0], Ok(PullProgress::Status { .. })),
2754 "first event should be a Status line, got {:?}",
2755 events[0]
2756 );
2757 assert!(
2758 matches!(events[1], Ok(PullProgress::Done { .. })),
2759 "second event should be the terminal Done, got {:?}",
2760 events[1]
2761 );
2762 }
2763
2764 #[tokio::test]
2765 async fn default_archive_get_is_unsupported() {
2766 let runtime = BareRuntime;
2767 let id = ContainerId::new("archive-test".to_string(), 0);
2768 match runtime.archive_get(&id, "/etc/hosts").await {
2769 Err(AgentError::Unsupported(_)) => {}
2770 Err(other) => panic!("expected Unsupported, got {other:?}"),
2771 Ok(_) => panic!("expected Err(Unsupported), got Ok"),
2772 }
2773 }
2774
2775 #[tokio::test]
2776 async fn default_archive_put_is_unsupported() {
2777 let runtime = BareRuntime;
2778 let id = ContainerId::new("archive-test".to_string(), 0);
2779 let err = runtime
2780 .archive_put(
2781 &id,
2782 "/tmp",
2783 bytes::Bytes::from_static(&[]),
2784 ArchivePutOptions::default(),
2785 )
2786 .await
2787 .unwrap_err();
2788 assert!(matches!(err, AgentError::Unsupported(_)));
2789 }
2790
2791 #[tokio::test]
2792 async fn default_archive_head_is_unsupported() {
2793 let runtime = BareRuntime;
2794 let id = ContainerId::new("archive-test".to_string(), 0);
2795 let err = runtime.archive_head(&id, "/etc/hosts").await.unwrap_err();
2796 assert!(matches!(err, AgentError::Unsupported(_)));
2797 }
2798
2799 #[tokio::test]
2800 async fn default_exec_pty_is_unsupported() {
2801 let runtime = BareRuntime;
2802 let id = ContainerId::new("exec-pty".to_string(), 0);
2803 // The success-side `ExecHandle` is not `Debug`, so we can't call
2804 // `unwrap_err`; pattern-match on the Result directly instead.
2805 match runtime.exec_pty(&id, ExecOptions::default()).await {
2806 Err(AgentError::Unsupported(_)) => {}
2807 Err(other) => panic!("expected Unsupported, got {other:?}"),
2808 Ok(_) => panic!("expected Err(Unsupported), got Ok"),
2809 }
2810 }
2811
2812 #[tokio::test]
2813 async fn default_inspect_image_native_is_unsupported() {
2814 let runtime = BareRuntime;
2815 let err = runtime.inspect_image_native("alpine").await.unwrap_err();
2816 assert!(matches!(err, AgentError::Unsupported(_)));
2817 }
2818
2819 #[tokio::test]
2820 async fn default_image_history_is_unsupported() {
2821 let runtime = BareRuntime;
2822 let err = runtime.image_history("alpine").await.unwrap_err();
2823 assert!(matches!(err, AgentError::Unsupported(_)));
2824 }
2825
2826 #[tokio::test]
2827 async fn default_search_images_is_unsupported() {
2828 let runtime = BareRuntime;
2829 let err = runtime.search_images("nginx", 10).await.unwrap_err();
2830 assert!(matches!(err, AgentError::Unsupported(_)));
2831 }
2832
2833 #[tokio::test]
2834 async fn default_save_images_is_unsupported() {
2835 let runtime = BareRuntime;
2836 // The success-side stream isn't `Debug`, so pattern-match the Result.
2837 match runtime.save_images(&["alpine".to_string()]).await {
2838 Err(AgentError::Unsupported(_)) => {}
2839 Err(other) => panic!("expected Unsupported, got {other:?}"),
2840 Ok(_) => panic!("expected Err(Unsupported), got Ok"),
2841 }
2842 }
2843
2844 #[tokio::test]
2845 async fn default_load_images_is_unsupported() {
2846 let runtime = BareRuntime;
2847 match runtime
2848 .load_images(bytes::Bytes::from_static(&[]), false)
2849 .await
2850 {
2851 Err(AgentError::Unsupported(_)) => {}
2852 Err(other) => panic!("expected Unsupported, got {other:?}"),
2853 Ok(_) => panic!("expected Err(Unsupported), got Ok"),
2854 }
2855 }
2856
2857 #[tokio::test]
2858 async fn default_import_image_is_unsupported() {
2859 let runtime = BareRuntime;
2860 let err = runtime
2861 .import_image(bytes::Bytes::from_static(&[]), None, None)
2862 .await
2863 .unwrap_err();
2864 assert!(matches!(err, AgentError::Unsupported(_)));
2865 }
2866
2867 #[tokio::test]
2868 async fn default_export_container_fs_is_unsupported() {
2869 let runtime = BareRuntime;
2870 let id = ContainerId::new("export".to_string(), 0);
2871 match runtime.export_container_fs(&id).await {
2872 Err(AgentError::Unsupported(_)) => {}
2873 Err(other) => panic!("expected Unsupported, got {other:?}"),
2874 Ok(_) => panic!("expected Err(Unsupported), got Ok"),
2875 }
2876 }
2877
2878 #[tokio::test]
2879 async fn default_commit_container_is_unsupported() {
2880 let runtime = BareRuntime;
2881 let id = ContainerId::new("commit".to_string(), 0);
2882 let err = runtime
2883 .commit_container(&id, &CommitOptions::default())
2884 .await
2885 .unwrap_err();
2886 assert!(matches!(err, AgentError::Unsupported(_)));
2887 }
2888
2889 #[test]
2890 fn load_progress_serializes_with_kind_discriminator() {
2891 let status = LoadProgress::Status {
2892 id: Some("abc".to_string()),
2893 status: "Loading layer".to_string(),
2894 };
2895 let json = serde_json::to_value(&status).unwrap();
2896 assert_eq!(json["kind"], "status");
2897 assert_eq!(json["status"], "Loading layer");
2898
2899 let done = LoadProgress::Done {
2900 references: vec!["alpine:latest".to_string()],
2901 };
2902 let json = serde_json::to_value(&done).unwrap();
2903 assert_eq!(json["kind"], "done");
2904 assert_eq!(json["references"], serde_json::json!(["alpine:latest"]));
2905 }
2906
2907 #[test]
2908 fn commit_options_default_is_no_op_pause_false() {
2909 let opts = CommitOptions::default();
2910 assert!(opts.repo.is_none());
2911 assert!(opts.tag.is_none());
2912 assert!(opts.comment.is_none());
2913 assert!(opts.author.is_none());
2914 assert!(!opts.pause);
2915 assert!(opts.changes.is_none());
2916 }
2917
2918 #[test]
2919 fn image_inspect_info_default_round_trips_via_serde() {
2920 let info = ImageInspectInfo::default();
2921 let json = serde_json::to_string(&info).unwrap();
2922 let back: ImageInspectInfo = serde_json::from_str(&json).unwrap();
2923 assert_eq!(info, back);
2924 }
2925
2926 #[tokio::test]
2927 async fn mock_logs_stream_yields_queued_items_in_order() {
2928 use futures_util::StreamExt;
2929
2930 let runtime = MockRuntime::new();
2931 let id = ContainerId::new("logs-order".to_string(), 0);
2932
2933 let make_chunk = |s: &str, ch: LogChannel| LogChunk {
2934 stream: ch,
2935 bytes: bytes::Bytes::copy_from_slice(s.as_bytes()),
2936 timestamp: None,
2937 };
2938
2939 runtime
2940 .enqueue_log_chunk(&id, make_chunk("first", LogChannel::Stdout))
2941 .await;
2942 runtime
2943 .enqueue_log_chunk(&id, make_chunk("second", LogChannel::Stderr))
2944 .await;
2945 runtime
2946 .enqueue_log_chunk(&id, make_chunk("third", LogChannel::Stdout))
2947 .await;
2948
2949 // `follow=false` so the stream ends once the queue is drained.
2950 let opts = LogsStreamOptions {
2951 follow: false,
2952 ..LogsStreamOptions::default()
2953 };
2954 let mut stream = runtime.logs_stream(&id, opts).await.unwrap();
2955
2956 let mut got = Vec::new();
2957 while let Some(item) = stream.next().await {
2958 let chunk = item.unwrap();
2959 got.push((
2960 chunk.stream,
2961 String::from_utf8(chunk.bytes.to_vec()).unwrap(),
2962 ));
2963 }
2964 assert_eq!(
2965 got,
2966 vec![
2967 (LogChannel::Stdout, "first".to_string()),
2968 (LogChannel::Stderr, "second".to_string()),
2969 (LogChannel::Stdout, "third".to_string()),
2970 ]
2971 );
2972 }
2973
2974 #[tokio::test]
2975 async fn mock_logs_stream_empty_queue_ends_immediately_when_not_follow() {
2976 use futures_util::StreamExt;
2977
2978 let runtime = MockRuntime::new();
2979 let id = ContainerId::new("logs-empty".to_string(), 0);
2980
2981 let opts = LogsStreamOptions {
2982 follow: false,
2983 ..LogsStreamOptions::default()
2984 };
2985 let mut stream = runtime.logs_stream(&id, opts).await.unwrap();
2986
2987 // Empty queue + follow=false => stream is closed on first poll.
2988 // Wrap in a short timeout so a regression that hangs would surface
2989 // as a test failure rather than the test runner stalling.
2990 let next = tokio::time::timeout(Duration::from_millis(500), stream.next())
2991 .await
2992 .expect("stream did not terminate; expected immediate close on empty queue");
2993 assert!(
2994 next.is_none(),
2995 "expected stream to be exhausted, got Some(_)"
2996 );
2997 }
2998
2999 #[tokio::test]
3000 async fn mock_stats_stream_yields_queued_samples_in_order() {
3001 use futures_util::StreamExt;
3002
3003 let runtime = MockRuntime::new();
3004 let id = ContainerId::new("stats-order".to_string(), 0);
3005
3006 let now = chrono::Utc::now();
3007 let mk = |cpu: u64| StatsSample {
3008 cpu_total_ns: cpu,
3009 cpu_system_ns: 0,
3010 online_cpus: 1,
3011 mem_used_bytes: 0,
3012 mem_limit_bytes: 0,
3013 net_rx_bytes: 0,
3014 net_tx_bytes: 0,
3015 blkio_read_bytes: 0,
3016 blkio_write_bytes: 0,
3017 pids_current: 0,
3018 pids_limit: None,
3019 timestamp: now,
3020 };
3021
3022 runtime.enqueue_stats_sample(&id, mk(100)).await;
3023 runtime.enqueue_stats_sample(&id, mk(200)).await;
3024 runtime.enqueue_stats_sample(&id, mk(300)).await;
3025
3026 let mut stream = runtime.stats_stream(&id).await.unwrap();
3027
3028 let mut cpus = Vec::new();
3029 while let Some(item) = stream.next().await {
3030 cpus.push(item.unwrap().cpu_total_ns);
3031 }
3032 assert_eq!(cpus, vec![100, 200, 300]);
3033 }
3034
3035 #[tokio::test]
3036 async fn mock_pull_image_stream_yields_queued_progress_in_order() {
3037 use futures_util::StreamExt;
3038
3039 let runtime = MockRuntime::new();
3040 let image = "alpine:latest";
3041
3042 runtime
3043 .enqueue_pull_progress(
3044 image,
3045 PullProgress::Status {
3046 id: Some("layer-1".to_string()),
3047 status: "Pulling fs layer".to_string(),
3048 progress: None,
3049 current: None,
3050 total: None,
3051 },
3052 )
3053 .await;
3054 runtime
3055 .enqueue_pull_progress(
3056 image,
3057 PullProgress::Status {
3058 id: Some("layer-1".to_string()),
3059 status: "Downloading".to_string(),
3060 progress: Some("[==> ] 1MB/4MB".to_string()),
3061 current: Some(1024 * 1024),
3062 total: Some(4 * 1024 * 1024),
3063 },
3064 )
3065 .await;
3066 runtime
3067 .enqueue_pull_progress(
3068 image,
3069 PullProgress::Done {
3070 reference: image.to_string(),
3071 digest: Some("sha256:deadbeef".to_string()),
3072 },
3073 )
3074 .await;
3075
3076 let mut stream = runtime.pull_image_stream(image, None).await.unwrap();
3077 let mut events = Vec::new();
3078 while let Some(item) = stream.next().await {
3079 events.push(item.unwrap());
3080 }
3081
3082 assert_eq!(events.len(), 3);
3083 match &events[0] {
3084 PullProgress::Status { status, .. } => assert_eq!(status, "Pulling fs layer"),
3085 done @ PullProgress::Done { .. } => panic!("expected Status, got {done:?}"),
3086 }
3087 match &events[1] {
3088 PullProgress::Status {
3089 status,
3090 current,
3091 total,
3092 ..
3093 } => {
3094 assert_eq!(status, "Downloading");
3095 assert_eq!(*current, Some(1024 * 1024));
3096 assert_eq!(*total, Some(4 * 1024 * 1024));
3097 }
3098 done @ PullProgress::Done { .. } => panic!("expected Status, got {done:?}"),
3099 }
3100 match &events[2] {
3101 PullProgress::Done { reference, digest } => {
3102 assert_eq!(reference, image);
3103 assert_eq!(digest.as_deref(), Some("sha256:deadbeef"));
3104 }
3105 status @ PullProgress::Status { .. } => panic!("expected Done, got {status:?}"),
3106 }
3107 }
3108
3109 #[test]
3110 fn log_channel_serializes_as_snake_case() {
3111 assert_eq!(
3112 serde_json::to_string(&LogChannel::Stdin).unwrap(),
3113 "\"stdin\""
3114 );
3115 assert_eq!(
3116 serde_json::to_string(&LogChannel::Stdout).unwrap(),
3117 "\"stdout\""
3118 );
3119 assert_eq!(
3120 serde_json::to_string(&LogChannel::Stderr).unwrap(),
3121 "\"stderr\""
3122 );
3123 }
3124
3125 fn mock_spec() -> ServiceSpec {
3126 use zlayer_spec::*;
3127 serde_yaml::from_str::<DeploymentSpec>(
3128 r"
3129version: v1
3130deployment: test
3131services:
3132 test:
3133 rtype: service
3134 image:
3135 name: test:latest
3136 endpoints:
3137 - name: http
3138 protocol: http
3139 port: 8080
3140",
3141 )
3142 .unwrap()
3143 .services
3144 .remove("test")
3145 .unwrap()
3146 }
3147}