| 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460 |
- use crate::config::*;
- use crate::pre_serve::*;
- use std::collections::{HashMap, HashSet};
- fn tcls() -> Cluster {
- Cluster {
- name: "test-cluster".to_string(),
- master_ib_ip: Some("192.168.1.1".to_string()),
- nodes: vec![
- Node {
- name: "node0".to_string(),
- ssh_ip: "10.0.0.1".to_string(),
- ib_ip: Some("192.168.1.1".to_string()),
- rank: 0,
- host_key: None,
- entity: None,
- },
- Node {
- name: "node1".to_string(),
- ssh_ip: "10.0.0.2".to_string(),
- ib_ip: Some("192.168.1.2".to_string()),
- rank: 1,
- host_key: None,
- entity: None,
- },
- ],
- extra_entities: vec![],
- }
- }
- fn tcfg() -> Config {
- Config {
- ssh_user: "u".to_string(),
- ssh_key_path: "/k".into(),
- clusters: vec![tcls()],
- nccl: HashMap::new(),
- hf_cache_dir: None,
- init_port: 25000,
- before_serve: HashMap::new(),
- exec_aliases: HashMap::new(),
- home_assistant: None,
- global_mods_dir: None,
- drop_caches: true,
- mount_aliases: HashMap::new(),
- recipe_sources: Vec::new(),
- docker_image_sources: Vec::new(),
- cluster_sources: Vec::new(),
- }
- }
- fn tnccl() -> NcclConfig {
- NcclConfig {
- ib_hca: "rocep1s0f0".to_string(),
- socket_ifname: "enp1s0f0np0".to_string(),
- 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,
- }
- }
- #[test]
- fn test_recipe_deserialize_minimal() {
- let r: Recipe = toml::from_str("\nname = \"M\"\nmodel = \"t/m\"\n").unwrap();
- assert_eq!(r.name, "M");
- assert_eq!(r.engine, "vllm");
- assert_eq!(r.port, 30000);
- }
- #[test]
- fn test_recipe_deserialize_full() {
- let r: Recipe = toml::from_str("engine = \"sglang\"\nname = \"Full\"\ndocker_image = \"o/i:t\"\nmodel = \"o/m\"\nport = 30000\nrequired_nodes = 4\npipeline_parallel_size = 2\ndistributed_backend = \"ray\"\ncluster = \"c\"\nargs = [\"--a\", \"v\"]\n[env]\nK = \"V\"\n").unwrap();
- assert_eq!(r.engine, "sglang");
- assert_eq!(r.required_nodes, Some(4));
- assert_eq!(r.distributed_backend, Some(DistributedBackend::Ray));
- }
- #[test]
- fn test_recipe_deserialize_no_required_nodes() {
- let r: Recipe = toml::from_str("name = \"F\"\nmodel = \"t/m\"\n").unwrap();
- assert_eq!(r.required_nodes, None);
- assert_eq!(r.pipeline_parallel_size, None);
- }
- #[test]
- fn test_composed_source_exclude_deserialize() {
- let toml = r#"
- name = "C"
- model = "t/m"
- [composed_model]
- mount = "/root/model"
- [[composed_model.sources]]
- repo = "base"
- [[composed_model.sources]]
- repo = "overlay"
- glob = "L*.safetensors"
- exclude = ["L5.safetensors", "L8.safetensors"]
- [[composed_model.files]]
- name = "cfg.json"
- content = "{}"
- "#;
- let r: Recipe = toml::from_str(toml).unwrap();
- let composed = r.composed_model.unwrap();
- assert_eq!(composed.sources.len(), 2);
- let overlay = &composed.sources[1];
- assert_eq!(overlay.repo, "overlay");
- assert_eq!(overlay.glob.as_deref(), Some("L*.safetensors"));
- let exclude = overlay
- .exclude
- .as_ref()
- .expect("exclude should deserialize");
- assert_eq!(exclude.len(), 2);
- assert!(exclude.contains(&"L5.safetensors".to_string()));
- assert!(exclude.contains(&"L8.safetensors".to_string()));
- }
- #[test]
- fn test_config_deserialize() {
- let c: Config = toml::from_str("ssh_user = \"a\"\nssh_key_path = \"/p\"\n[[clusters]]\nname = \"p\"\nmaster_ib_ip = \"1\"\nnodes = [{name=\"n1\",ssh_ip=\"10.0.0.10\",ib_ip=\"1\",rank=0},{name=\"n2\",ssh_ip=\"10.0.0.11\",ib_ip=\"2\",rank=1}]\n[nccl.\"p\"]\nib_hca = \"m\"\nsocket_ifname = \"e\"\nib_gid_index = 3\nib_disable = 0\n").unwrap();
- assert_eq!(c.clusters.len(), 1);
- assert_eq!(c.clusters[0].nodes.len(), 2);
- }
- #[test]
- fn test_config_deserialize_with_host_key() {
- let c: Config = toml::from_str("ssh_user = \"a\"\nssh_key_path = \"/k\"\n[[clusters]]\nname = \"s\"\nmaster_ib_ip = \"1\"\nnodes = [\n { name = \"n\", ssh_ip = \"2\", ib_ip = \"1\", rank = 0, host_key = \"SHA256:abc\" },\n]\n").unwrap();
- assert_eq!(c.clusters[0].nodes[0].host_key, Some("SHA256:abc".into()));
- }
- #[test]
- fn test_select_cluster_by_name() {
- let mut cfg = tcfg();
- cfg.nccl.insert("test-cluster".into(), tnccl());
- let (c, n) = select_cluster(&cfg, "test-cluster", None, &HashSet::new()).unwrap();
- assert_eq!(c.name, "test-cluster");
- assert_eq!(n.unwrap().ib_hca, "rocep1s0f0");
- }
- #[test]
- fn test_select_cluster_auto() {
- let mut cfg = tcfg();
- cfg.nccl.insert("test-cluster".into(), tnccl());
- let (c, _n) = select_cluster(&cfg, "auto", None, &HashSet::new()).unwrap();
- assert_eq!(c.name, "test-cluster");
- }
- #[test]
- fn test_select_cluster_auto_no_nccl_allows_single_node() {
- // With required_nodes = None, NCCL config is not required — the cluster
- // is selected even without NCCL config (the node count is unknown at
- // selection time, so we can't require it).
- let cfg = tcfg();
- let (c, n) = select_cluster(&cfg, "auto", None, &HashSet::new()).unwrap();
- assert_eq!(c.name, "test-cluster");
- assert!(n.is_none());
- }
- #[test]
- fn test_select_cluster_multi_node_requires_nccl() {
- // Multi-node (required_nodes > 1) without NCCL config → error.
- assert!(select_cluster(&tcfg(), "auto", Some(2), &HashSet::new()).is_err());
- }
- #[test]
- fn test_select_cluster_not_found() {
- let mut cfg = tcfg();
- cfg.clusters = vec![];
- assert!(select_cluster(&cfg, "x", None, &HashSet::new()).is_err());
- }
- #[test]
- fn test_select_cluster_by_name_found() {
- let mut nccl = HashMap::new();
- nccl.insert("test-cluster".into(), tnccl());
- let mut cfg = tcfg();
- cfg.nccl = nccl;
- let c = select_cluster_by_name(&cfg, "test-cluster").unwrap();
- assert_eq!(c.name, "test-cluster");
- }
- #[test]
- fn test_select_cluster_by_name_not_found() {
- assert!(select_cluster_by_name(&tcfg(), "x").is_err());
- }
- #[test]
- fn test_select_cluster_by_prefix() {
- let mut cfg = tcfg();
- cfg.nccl.insert("test-cluster".into(), tnccl());
- let c = select_cluster_by_name(&cfg, "test-c").unwrap();
- assert_eq!(c.name, "test-cluster");
- assert!(select_cluster_by_name(&cfg, "no-such").is_err());
- }
- #[test]
- fn test_default_engine_is_vllm() {
- let r: Recipe = toml::from_str("name = \"x\"\nmodel = \"m\"\n").unwrap();
- assert_eq!(r.engine, "vllm");
- }
- #[test]
- fn test_default_distributed_backend_is_none() {
- let r: Recipe = toml::from_str("name = \"x\"\nmodel = \"m\"\n").unwrap();
- assert_eq!(r.distributed_backend, None);
- }
- #[test]
- fn test_deploy_defaults_to_docker() {
- let r: Recipe = toml::from_str("name = \"x\"\nmodel = \"m\"\n").unwrap();
- assert_eq!(r.deploy, DeployTarget::Docker);
- }
- #[test]
- fn test_deploy_explicit_tmux() {
- let r: Recipe = toml::from_str("name = \"x\"\nmodel = \"m\"\ndeploy = \"tmux\"\n").unwrap();
- assert_eq!(r.deploy, DeployTarget::Tmux);
- }
- #[test]
- fn test_deploy_explicit_docker() {
- let r: Recipe = toml::from_str("name = \"x\"\nmodel = \"m\"\ndeploy = \"docker\"\n").unwrap();
- assert_eq!(r.deploy, DeployTarget::Docker);
- }
- #[test]
- fn test_venv_field() {
- let r: Recipe = toml::from_str(
- "name = \"x\"\nmodel = \"m\"\ndeploy = \"tmux\"\nvenv = \"~/venvs/vllm025\"\n",
- )
- .unwrap();
- assert_eq!(r.deploy, DeployTarget::Tmux);
- assert_eq!(r.venv.as_deref(), Some("~/venvs/vllm025"));
- }
- #[test]
- fn test_venv_defaults_to_none() {
- let r: Recipe = toml::from_str("name = \"x\"\nmodel = \"m\"\n").unwrap();
- assert_eq!(r.venv, None);
- }
- #[test]
- fn test_venv_setup_list_deserializes() {
- let r: Recipe = toml::from_str(
- r#"name = "x"
- model = "m"
- deploy = "tmux"
- venv = "~/venvs/vllm025"
- venv_setup = [
- "uv venv ~/venvs/vllm025 -p 3.12",
- "uv pip install vllm==0.25.1 --torch-backend=cu130",
- ]
- "#,
- )
- .unwrap();
- assert_eq!(r.venv.as_deref(), Some("~/venvs/vllm025"));
- let setup = r.venv_setup.to_vec();
- assert_eq!(setup.len(), 2);
- assert_eq!(setup[0], "uv venv ~/venvs/vllm025 -p 3.12");
- assert_eq!(
- setup[1],
- "uv pip install vllm==0.25.1 --torch-backend=cu130"
- );
- }
- #[test]
- fn test_venv_setup_inline_deserializes() {
- let r: Recipe = toml::from_str(
- r#"name = "x"
- model = "m"
- deploy = "tmux"
- venv = "~/venvs/vllm025"
- venv_setup = """
- uv venv ~/venvs/vllm025 -p 3.12
- uv pip install vllm==0.25.1 --torch-backend=cu130
- """
- "#,
- )
- .unwrap();
- let setup = r.venv_setup.to_vec();
- assert_eq!(setup.len(), 2);
- assert_eq!(setup[0], "uv venv ~/venvs/vllm025 -p 3.12");
- assert_eq!(
- setup[1],
- "uv pip install vllm==0.25.1 --torch-backend=cu130"
- );
- }
- #[test]
- fn test_venv_setup_inline_skips_comments_and_blanks() {
- let r: Recipe = toml::from_str(
- r#"name = "x"
- model = "m"
- deploy = "tmux"
- venv = "~/venvs/vllm025"
- venv_setup = """
- # create the venv
- uv venv ~/venvs/vllm025 -p 3.12
- # install deps
- uv pip install vllm==0.25.1
- """
- "#,
- )
- .unwrap();
- let setup = r.venv_setup.to_vec();
- assert_eq!(setup.len(), 2);
- assert_eq!(setup[0], "uv venv ~/venvs/vllm025 -p 3.12");
- assert_eq!(setup[1], "uv pip install vllm==0.25.1");
- }
- #[test]
- fn test_venv_setup_defaults_to_empty() {
- let r: Recipe = toml::from_str("name = \"x\"\nmodel = \"m\"\n").unwrap();
- assert!(r.venv_setup.to_vec().is_empty());
- }
- #[test]
- fn default_config_dir_appends_sm121() {
- let Some(d) = default_config_dir() else {
- return;
- };
- assert!(d.ends_with("sm121") && d.parent().is_some());
- }
- #[tokio::test]
- async fn test_merge_cluster_sources_adds_new_keeps_main() {
- let dir = std::env::temp_dir().join(format!("sm121-csrc-{}", std::process::id()));
- std::fs::create_dir_all(&dir).unwrap();
- let src_path = dir.join("clusters-extra.toml");
- // "extra" is new; "test-cluster" collides with tcfg's cluster (main wins).
- std::fs::write(
- &src_path,
- r#"
- ssh_user = "srcuser"
- ssh_key_path = "/src/key"
- [[clusters]]
- name = "extra"
- nodes = [{ name = "e0", ssh_ip = "10.9.0.1", rank = 0 }]
- [[clusters]]
- name = "test-cluster"
- nodes = [{ name = "intruder", ssh_ip = "10.66.0.1", rank = 0 }]
- [nccl."extra"]
- ib_hca = "mlx5_0"
- socket_ifname = "eth0"
- ib_disable = 0
- [before_serve."extra"]
- commands = ["echo hi"]
- [home_assistant]
- url = "http://ha.lan:8123"
- token = "sufficiently-long-test-token"
- shutdown_delay_seconds = 30
- "#,
- )
- .unwrap();
- let mut cfg = tcfg();
- cfg.cluster_sources = vec![src_path.to_string_lossy().into_owned()];
- cfg.ssh_user = String::new();
- cfg.ssh_key_path = std::path::PathBuf::new();
- merge_cluster_sources(&mut cfg).await.unwrap();
- assert_eq!(cfg.ssh_user, "srcuser", "empty ssh_user filled from source");
- assert_eq!(cfg.ssh_key_path, std::path::PathBuf::from("/src/key"));
- assert_eq!(
- cfg.clusters.len(),
- 2,
- "extra added, conflicting name not duplicated"
- );
- let main = cfg
- .clusters
- .iter()
- .find(|c| c.name == "test-cluster")
- .unwrap();
- assert_eq!(
- main.nodes[0].name, "node0",
- "main-config cluster wins on conflict"
- );
- assert!(cfg.clusters.iter().any(|c| c.name == "extra"));
- assert!(cfg.nccl.contains_key("extra"));
- assert!(cfg.before_serve.contains_key("extra"));
- // HA config merges from the source when main has none (power on from
- // other cwds depends on this).
- assert!(
- cfg.home_assistant.is_some(),
- "home_assistant filled from source"
- );
- assert_eq!(
- cfg.home_assistant.as_ref().unwrap().url,
- "http://ha.lan:8123"
- );
- // A main-config ssh_user is never overridden by a source; same rule
- // for home_assistant.
- let mut cfg2 = tcfg();
- cfg2.cluster_sources = vec![src_path.to_string_lossy().into_owned()];
- cfg2.home_assistant = Some(crate::config::HomeAssistantConfig {
- url: "http://main-ha:8123".into(),
- token: "main".into(),
- shutdown_delay_seconds: 5,
- });
- merge_cluster_sources(&mut cfg2).await.unwrap();
- assert_eq!(cfg2.ssh_user, "u", "main ssh_user wins over source");
- assert_eq!(
- cfg2.home_assistant.as_ref().unwrap().url,
- "http://main-ha:8123",
- "main home_assistant wins over source"
- );
- }
- #[tokio::test]
- async fn test_load_recipe_from_recipe_sources() {
- let dir = std::env::temp_dir().join(format!("sm121-rsrc-{}", std::process::id()));
- let rdir = dir.join("remote-recipes").join("remote-recipe");
- std::fs::create_dir_all(&rdir).unwrap();
- std::fs::write(rdir.join("recipe.toml"), "name = \"remote\"\n").unwrap();
- let recipe = load_recipe(
- "remote-recipe",
- &[dir.join("remote-recipes").to_string_lossy().into_owned()],
- )
- .await
- .unwrap();
- assert_eq!(recipe.name, "remote");
- }
- #[test]
- fn effective_master_ip_falls_back_to_rank0_ib_ip() {
- let mut c = tcls();
- c.master_ib_ip = None;
- assert_eq!(c.effective_master_ip(), "192.168.1.1"); // rank-0 ib_ip
- }
- #[test]
- fn effective_master_ip_falls_back_to_ssh_ip_when_ib_missing() {
- let mut c = tcls();
- c.master_ib_ip = None;
- for n in &mut c.nodes {
- n.ib_ip = None;
- }
- assert_eq!(c.effective_master_ip(), "10.0.0.1"); // rank-0 ssh_ip
- }
- #[test]
- fn effective_master_ip_uses_first_node_without_rank0() {
- let mut c = tcls();
- c.master_ib_ip = None;
- for n in &mut c.nodes {
- n.rank = 3; // no rank 0 → first node wins
- }
- assert_eq!(c.effective_master_ip(), "192.168.1.1");
- }
- #[test]
- fn effective_master_ip_prefers_explicit_config() {
- assert_eq!(tcls().effective_master_ip(), "192.168.1.1");
- }
|