launcher/src/docker/tests.rs

use super::command::{DockerCommandCtx, build_docker_command};
use super::*;
use crate::config::{DeployTarget, DistributedBackend, RecipeArgs, VenvSetup};
use std::collections::HashMap;

fn tcls() -> Cluster {
    Cluster {
        name: "t".into(),
        master_ib_ip: Some("1".into()),
        nodes: vec![
            Node {
                name: "n0".into(),
                ssh_ip: "10.0.0.1".into(),
                ib_ip: Some("1".into()),
                rank: 0,
                host_key: None,
                entity: None,
            },
            Node {
                name: "n1".into(),
                ssh_ip: "10.0.0.2".into(),
                ib_ip: Some("2".into()),
                rank: 1,
                host_key: None,
                entity: None,
            },
        ],
        extra_entities: vec![],
    }
}

fn tnccl() -> NcclConfig {
    NcclConfig {
        ib_hca: "r".into(),
        socket_ifname: "e".into(),
        ib_gid_index: Some(3),
        ib_disable: 0,
        gloo_socket_ifname: None,
        ucx_net_devices: None,
        nccl_cross_nic: None,
        nccl_proto: None,
        nccl_p2p_level: None,
    }
}

fn trec() -> Recipe {
    Recipe {
        engine: "vllm".into(),
        name: "T".into(),
        docker_image: "img".into(),
        model: "mdl".into(),
        port: 8000,
        required_nodes: Some(2),
        pipeline_parallel_size: Some(2),
        tensor_parallel_size: 1,
        args: RecipeArgs::List(vec!["--x".into(), "4096".into()]),
        env: HashMap::from([("K".into(), "V".into())]),
        cluster: None,
        entrypoint: vec![],
        distributed_backend: Some(DistributedBackend::Mp),
        before_serve: vec![],
        after_stop: vec![],
        solo: None,
        deploy: DeployTarget::Docker,
        venv: None,
        venv_setup: VenvSetup::default(),
        venv_setup_script: None,
        mods: Default::default(),
        shm_size: String::new(),
        sync_image: true,
        docker_image_dir: None,
        extra_mounts: vec![],
        build: Default::default(),
        composed_model: None,
    }
}

fn make_dm_ctx(recipe: &Recipe, nnodes: u32, pp: u32, tp: u32, keep: bool) -> DockerCommandCtx {
    DockerCommandCtx {
        cluster: tcls(),
        nccl: Some(tnccl()),
        recipe: recipe.clone(),
        recipe_name: "r".to_string(),
        pp_size: pp,
        tp_size: tp,
        nnodes,
        cache_dir: "/home/user/.cache/huggingface".to_string(),
        keep,
        init_port: 25000,
        mods_remote_path: String::new(),
        extra_mounts: vec![],
    }
}

/// Remove the single-quote wrapping `shq` adds, so existing assertions that
/// check the un-quoted command shape keep working. Quoting itself is asserted
/// separately (see `malicious_env_value_cannot_inject` etc.).
fn squash(s: &str) -> String {
    s.replace('\'', "")
}

fn run(recipe: &Recipe, node: &Node, head: bool) -> String {
    let tp = if recipe.required_nodes == Some(recipe.tensor_parallel_size) {
        2
    } else {
        recipe.tensor_parallel_size
    };
    squash(&build_docker_command(&make_dm_ctx(recipe, 2, 2, tp, false), node, head).join(" "))
}

#[test]
fn nccl_gid_unresolved_renders_auto() {
    // A cluster [nccl] block without `ib_gid_index` (auto-detect) renders
    // `NCCL_IB_GID_INDEX=auto` — the launcher resolves the real index per
    // node before the docker run; `auto` only appears on the dry-run/unresolved
    // path. Confirms the env is emitted and clearly marked, not dropped.
    let mut nccl = tnccl();
    nccl.ib_gid_index = None;
    let mut ctx = make_dm_ctx(&trec(), 2, 2, 2, false);
    ctx.nccl = Some(nccl);
    let cmd = squash(&build_docker_command(&ctx, &tcls().nodes[0], true).join(" "));
    assert!(
        cmd.contains("-e NCCL_IB_GID_INDEX=auto"),
        "unresolved GID should render 'auto' placeholder: {cmd}"
    );
}

fn run_keep(recipe: &Recipe, node: &Node, head: bool) -> String {
    let tp = if recipe.required_nodes == Some(recipe.tensor_parallel_size) {
        2
    } else {
        recipe.tensor_parallel_size
    };
    squash(&build_docker_command(&make_dm_ctx(recipe, 2, 2, tp, true), node, head).join(" "))
}

#[test]
fn mp() {
    let cmd = run(&trec(), &tcls().nodes[0], true);
    assert!(cmd.contains("docker run -d --gpus all"));
    assert!(cmd.contains("-e NCCL_IB_HCA=r"));
    assert!(cmd.contains("-e K=V"));
    assert!(cmd.contains("--name sm121_r_0"));
    assert!(
        !cmd.contains("-p "),
        "host networking: no port mapping expected"
    );
    assert!(cmd.contains("--label sm121.recipe=r"));
    assert!(cmd.contains("--label sm121.rank=0"));
    assert!(cmd.contains("img"));
    assert!(cmd.contains("mdl"));
    assert!(cmd.contains("--tensor-parallel-size 1"));
    assert!(cmd.contains("--pipeline-parallel-size 2"));
    assert!(cmd.contains("--nnodes 2"));
    assert!(cmd.contains("--node-rank 0"));
    assert!(cmd.contains("--master-addr 1"));
    assert!(cmd.contains("--master-port 25000"));
    // vLLM auto-detects mp; only explicitly pass for ray
    assert!(
        !cmd.contains("--distributed-executor-backend"),
        "mp backend should be auto-detected"
    );
    assert!(!cmd.contains("RAY_ADDRESS"));
    assert!(cmd.contains("--rm"), "default --rm should be present");
    // New: defaults & cluster env
    assert!(
        !cmd.contains("--shm-size"),
        "shm-size should not be set by default"
    );
    assert!(cmd.contains("--ulimit memlock=-1"));
    assert!(cmd.contains("--ulimit stack=67108864"));
    assert!(
        cmd.contains("-e GLOO_SOCKET_IFNAME=e"),
        "default GLOO_SOCKET_IFNAME missing"
    );
    assert!(
        cmd.contains("-e TP_SOCKET_IFNAME=e"),
        "default TP_SOCKET_IFNAME missing"
    );
    assert!(
        cmd.contains("-e OMP_NUM_THREADS=4"),
        "default OMP_NUM_THREADS missing"
    );
    assert!(
        cmd.contains("-e NCCL_IGNORE_CPU_AFFINITY=1"),
        "default NCCL_IGNORE_CPU_AFFINITY missing"
    );
    assert!(cmd.contains("-e NCCL_NET=IB"), "default NCCL_NET missing");
    assert!(
        cmd.contains("-e NCCL_DEBUG=WARN"),
        "default NCCL_DEBUG missing"
    );
    assert!(
        cmd.contains("-e MN_IF_NAME=e"),
        "cluster MN_IF_NAME missing"
    );
    assert!(
        cmd.contains("-e UCX_NET_DEVICES=e"),
        "cluster UCX_NET_DEVICES missing"
    );
    assert!(
        cmd.contains("-e OMPI_MCA_btl_tcp_if_include=e"),
        "cluster OMPI_MCA_btl_tcp_if_include missing"
    );
    // Head node: no --headless
    assert!(!cmd.contains("--headless"));
}

#[test]
fn worker_no_port() {
    let cmd = run(&trec(), &tcls().nodes[1], false);
    assert!(!cmd.contains("-p 8000:8000"));
    assert!(cmd.contains("--node-rank 1"));
    assert!(cmd.contains("--label sm121.rank=1"));
    assert!(cmd.contains("--name sm121_r_1"));
    // New: vLLM workers run --headless
    assert!(cmd.contains("--headless"), "vLLM worker should be headless");
}

#[test]
fn ray_backend() {
    let mut r = trec();
    r.distributed_backend = Some(DistributedBackend::Ray);
    let cmd = run(&r, &tcls().nodes[0], true);
    assert!(cmd.contains("-e RAY_ADDRESS=1:6379"));
    assert!(cmd.contains("-e RAY_PORT=6379"));
    assert!(cmd.contains("--node-rank 0"));
    assert!(cmd.contains("--distributed-executor-backend ray"));
}

#[test]
fn custom_tensor_parallel_size() {
    // Per-node TP=1 is the default. Custom TP (e.g. multi-node TP)
    // can be set via recipe.tensor_parallel_size.
    let mut r = trec();
    r.tensor_parallel_size = 2;
    let cmd = run(&r, &tcls().nodes[0], true);
    assert!(cmd.contains("--tensor-parallel-size 2"), "custom TP");
}

#[test]
fn auto_pp_size() {
    let cmd = run(&trec(), &tcls().nodes[0], true);
    assert!(cmd.contains("--pipeline-parallel-size 2"));
}

#[test]
fn entrypoint() {
    let mut r = trec();
    r.entrypoint = vec!["py".into(), "-m".into(), "mod".into()];
    let cmd = run(&r, &tcls().nodes[0], true);
    let ip = cmd.find("img").unwrap();
    let mp = cmd.find("mdl").unwrap();
    let ep = cmd.find("py -m mod").unwrap();
    assert!(
        ip < ep && ep < mp,
        "entrypoint should be between image and model"
    );
}

#[test]
fn keep_disables_rm() {
    let cmd = run_keep(&trec(), &tcls().nodes[0], true);
    assert!(!cmd.contains("--rm"), "--keep should suppress --rm");
    assert!(cmd.contains("docker run -d"));
    assert!(cmd.contains("--gpus all"));
    assert!(cmd.contains("-e HF_HOME=/root/.cache/huggingface"));
    assert!(cmd.contains("-v /home/user/.cache/huggingface:/root/.cache/huggingface"));
}

#[test]
fn recipe_env_overrides_cluster_defaults() {
    // Recipe-level env should win over the defaults we inject.
    let mut r = trec();
    r.env.insert("NCCL_CUMEM_ENABLE".into(), "1".into());
    r.env.insert("OMP_NUM_THREADS".into(), "16".into());
    r.env.insert("NCCL_IGNORE_CPU_AFFINITY".into(), "0".into());
    r.env.insert("NCCL_NET".into(), "SOCKET".into());
    r.env.insert("NCCL_CROSS_NIC".into(), "0".into());
    r.env.insert("NCCL_DEBUG".into(), "INFO".into());
    let cmd = run(&r, &tcls().nodes[0], true);
    // Docker uses the last -e for duplicate keys, so recipe values come after.
    // rfind returns the last occurrence, which should be the recipe override.
    let cumem_pos = cmd.rfind("NCCL_CUMEM_ENABLE=").unwrap();
    assert!(
        cmd[cumem_pos..].starts_with("NCCL_CUMEM_ENABLE=1"),
        "recipe override should be last: slice starts at {}",
        cumem_pos
    );
    let omp_pos = cmd.rfind("OMP_NUM_THREADS=").unwrap();
    assert!(
        cmd[omp_pos..].starts_with("OMP_NUM_THREADS=16"),
        "recipe override should be last"
    );
    // New NCCL defaults can also be overridden.
    let ignore_pos = cmd.rfind("NCCL_IGNORE_CPU_AFFINITY=").unwrap();
    assert!(
        cmd[ignore_pos..].starts_with("NCCL_IGNORE_CPU_AFFINITY=0"),
        "NCCL_IGNORE_CPU_AFFINITY should be overridable"
    );
    let net_pos = cmd.rfind("NCCL_NET=").unwrap();
    assert!(
        cmd[net_pos..].starts_with("NCCL_NET=SOCKET"),
        "NCCL_NET should be overridable"
    );
    let cross_pos = cmd.rfind("NCCL_CROSS_NIC=").unwrap();
    assert!(
        cmd[cross_pos..].starts_with("NCCL_CROSS_NIC=0"),
        "NCCL_CROSS_NIC should be overridable"
    );
    let debug_pos = cmd.rfind("NCCL_DEBUG=").unwrap();
    assert!(
        cmd[debug_pos..].starts_with("NCCL_DEBUG=INFO"),
        "NCCL_DEBUG should be overridable"
    );
}

#[test]
fn custom_init_port() {
    // Non-default init_port should propagate to vLLM master-port and
    // SGLang dist-init-addr.
    let r = trec();
    let tp = if r.required_nodes == Some(r.tensor_parallel_size) {
        2
    } else {
        r.tensor_parallel_size
    };
    let mut ctx = make_dm_ctx(&r, 2, 2, tp, false);
    ctx.init_port = 29500;
    let cmd = squash(&build_docker_command(&ctx, &tcls().nodes[0], true).join(" "));
    assert!(cmd.contains("--master-port 29500"));
    assert!(cmd.contains("--master-addr 1"));
}

// ── Logs command format ───────────────────────────────────────

#[test]
fn logs_tail_cmd_format() {
    let cmd = super::logs::logs_tail_cmd("test-sleep", 0, 100);
    assert!(cmd.contains("docker ps -q"));
    assert!(cmd.contains("--filter 'label=sm121.recipe=test-sleep'"));
    assert!(cmd.contains("--filter label=sm121.rank=0"));
    assert!(cmd.contains("xargs -r docker logs"));
    assert!(cmd.contains("--timestamps"));
    assert!(cmd.contains("--tail=100"));
    assert!(cmd.ends_with(" 2>&1"));
}

#[test]
fn logs_tail_cmd_per_rank() {
    let cmd0 = super::logs::logs_tail_cmd("glm51", 0, 100);
    let cmd2 = super::logs::logs_tail_cmd("glm51", 2, 100);
    assert!(cmd0.contains("label=sm121.rank=0"));
    assert!(cmd2.contains("label=sm121.rank=2"));
    assert!(!cmd0.contains("label=sm121.rank=2"));
}

#[test]
fn logs_follow_cmd_format() {
    let cmd = super::logs::logs_follow_cmd("test-sleep");
    assert!(cmd.contains("xargs -r docker logs -f"));
    assert!(cmd.contains("--timestamps"));
    // Follow targets the head node only (rank=0).
    assert!(cmd.contains("--filter label=sm121.rank=0"));
    // Follow should not include --tail (we want a live stream from now).
    assert!(!cmd.contains("--tail"));
    // Follow must redirect stderr so docker's own errors are captured
    // (mirrors the behaviour of logs_tail_cmd).
    assert!(
        cmd.contains("2>&1"),
        "follow cmd must capture stderr, got: {cmd}"
    );
}

// ── Shell quoting ─────────────────────────────────────────────

#[test]
fn shq_wraps_simple_strings() {
    assert_eq!(shq("foo"), "'foo'");
    assert_eq!(shq(""), "''");
    assert_eq!(shq("foo bar"), "'foo bar'");
}

#[test]
fn shq_escapes_inner_single_quotes() {
    // "don't" → 'don'\''t'
    assert_eq!(shq("don't"), "'don'\\''t'");
}

#[test]
fn entrypoint_script_is_shell_quoted() {
    // Every tail word — incl. the `sh -c` script — is shell-quoted.
    let mut r = trec();
    r.entrypoint = vec!["sh".into(), "-c".into(), "echo $i; sleep 1".into()];
    let cmd =
        build_docker_command(&make_dm_ctx(&r, 2, 2, 2, false), &tcls().nodes[0], true).join(" ");
    assert!(
        cmd.contains("'echo $i; sleep 1'"),
        "must be shell-quoted: {cmd}"
    );
}

#[test]
fn entrypoint_script_non_sh_passes_through() {
    // For non `sh -c` entrypoints (e.g. ["sh", "-c", "...", "extra"]) or
    // direct binaries, we leave the args alone — no false positive shq.
    let mut r = trec();
    r.entrypoint = vec!["sh".into(), "-c".into()];
    let cmd = run(&r, &tcls().nodes[0], true);
    assert!(cmd.contains("docker run"));
}

#[path = "solo_tests.rs"]
mod solo_tests;

#[path = "sglang_tests.rs"]
mod sglang_tests;