manta_server/
config.rs

1//! Typed schema for `server.toml`.
2//!
3//! The untyped `config::Config` is loaded by
4//! [`manta_shared::common::config::get_server_configuration`]; this
5//! module owns the typed deserialisation target.
6//!
7//! The shape on disk is:
8//!
9//! ```toml
10//! log = "info"
11//!
12//! [server]
13//! listen_address = "0.0.0.0"
14//! port = 8443
15//! cert = "/etc/manta/tls/server.crt"
16//! key = "/etc/manta/tls/server.key"
17//! console_inactivity_timeout_secs = 1800
18//! auth_rate_limit_per_minute = 60
19//!
20//! [sites.alps]
21//! backend = "csm"
22//! shasta_base_url = "https://api.alps.cscs.ch"
23//! root_ca_cert_file = "/etc/manta/alps-ca.pem"
24//! ```
25//!
26//! [`ServerConfiguration`] is the top-level type; [`ServerSettings`]
27//! and [`Site`] are its nested sections. There is no notion of an
28//! "active" site — the server hosts every entry in `[sites.*]`
29//! simultaneously and clients pick one per request via the
30//! `X-Manta-Site` header.
31
32use std::collections::HashMap;
33
34use crate::server::common::audit::Auditor;
35use manta_backend_dispatcher::types::K8sDetails;
36use serde::{Deserialize, Serialize};
37
38/// Which backend API this site speaks.
39#[derive(Serialize, Deserialize, Debug, Clone, PartialEq)]
40#[serde(rename_all = "lowercase")]
41pub enum BackendTechnology {
42  /// HPE Cray System Management (CSM) backend.
43  Csm,
44  /// OpenCHAMI backend.
45  Ochami,
46}
47
48impl BackendTechnology {
49  /// Return the lowercase string expected by `StaticBackendDispatcher::new`.
50  pub fn as_str(&self) -> &'static str {
51    match self {
52      Self::Csm => "csm",
53      Self::Ochami => "ochami",
54    }
55  }
56}
57
58#[derive(Serialize, Deserialize, Debug)]
59/// Connection details for a single ALPS site (CSM or OCHAMI instance).
60///
61/// The Vault URL used by handlers requiring vault (sat-file, session,
62/// console, logs) is derived at startup from
63/// `[sites.X.k8s.authentication.vault] base_url`. The vault secret path
64/// is derived from a hard-coded prefix and the site name. Neither is
65/// configured here.
66pub struct Site {
67  /// Which backend implementation this site uses (`csm` or `ochami`).
68  pub backend: BackendTechnology,
69  /// Optional per-site SOCKS5 proxy URL used by every outbound HTTP
70  /// request to this site's backend. `None` means direct connection.
71  pub socks5_proxy: Option<String>,
72  /// Base URL of the backend API (e.g. `https://api.alps.cscs.ch`).
73  pub shasta_base_url: String,
74  /// Optional Kubernetes connection details, required by handlers
75  /// that stream CFS session logs or attach to consoles.
76  pub k8s: Option<K8sDetails>,
77  /// Path (absolute or relative to the config dir) of the backend's
78  /// root CA certificate, used to verify TLS to `shasta_base_url`.
79  pub root_ca_cert_file: String,
80}
81
82/// Server-only settings — TLS, listen address, console behaviour. Lives
83/// under `[server]` in `server.toml`.
84#[derive(Serialize, Deserialize, Debug)]
85pub struct ServerSettings {
86  /// TCP listen address (e.g. "0.0.0.0"). When omitted from config
87  /// **and** no `--listen-address` flag is supplied, the server falls
88  /// back to `"0.0.0.0"`.
89  #[serde(default)]
90  pub listen_address: Option<String>,
91  /// TCP port. When omitted from config **and** no `--port` flag is
92  /// supplied, the effective default depends on whether TLS is
93  /// configured: `8443` if both `cert` and `key` are present (HTTPS),
94  /// otherwise `8080` (plain HTTP). See
95  /// [`ServerSettings::default_port`].
96  #[serde(default)]
97  pub port: Option<u16>,
98  /// Path to the TLS certificate (PEM).
99  pub cert: Option<String>,
100  /// Path to the TLS private key (PEM).
101  pub key: Option<String>,
102  /// How long a node-console WebSocket stays open without activity
103  /// before the server tears it down.
104  pub console_inactivity_timeout_secs: u64,
105  /// Per-source-IP rate limit for the `/api/v1/auth/*` endpoints,
106  /// in requests per minute. `None` disables in-process rate limiting
107  /// (operators are then expected to enforce it at the reverse proxy).
108  pub auth_rate_limit_per_minute: Option<u32>,
109  /// Global request timeout applied to every HTTP route, in seconds.
110  /// When this elapses the server returns `408 REQUEST_TIMEOUT`. All
111  /// long-running work (e.g. power transitions) now runs CLI-side,
112  /// so no endpoint needs more than the default.
113  #[serde(default = "default_request_timeout_secs")]
114  pub request_timeout_secs: u64,
115  /// Grace period (seconds) `axum_server` waits for in-flight
116  /// requests to finish after SIGTERM / Ctrl+C before force-aborting.
117  /// Matches the standard k8s `terminationGracePeriodSeconds` default
118  /// (30 s); pods that hit this without finishing get SIGKILL'd by
119  /// the kubelet.
120  #[serde(default = "default_shutdown_grace_period_secs")]
121  pub shutdown_grace_period_secs: u64,
122  /// Filesystem root that confines `POST /migrate/{backup,restore}`
123  /// file access. When set, every `destination` / `bos_file` /
124  /// `cfs_file` / `hsm_file` / `ims_file` / `image_dir` path in the
125  /// request is canonicalised and rejected unless it resolves under
126  /// this directory. When unset (default), the migrate endpoints
127  /// return `BadRequest` even for admin callers — the operator must
128  /// explicitly opt in to server-side filesystem writes.
129  #[serde(default)]
130  pub migrate_backup_root: Option<String>,
131  /// Opt in to plain-HTTP listen mode. Default `false`: when neither
132  /// `cert` nor `key` is configured the server refuses to start, so
133  /// bearer tokens can't accidentally land on the wire in cleartext.
134  /// Set to `true` only when TLS terminates upstream (reverse proxy
135  /// or sidecar); otherwise leave it off and configure both `cert`
136  /// and `key`.
137  #[serde(default)]
138  pub allow_http: bool,
139}
140
141impl ServerSettings {
142  /// Effective default listen address when neither config nor CLI flag
143  /// supplies one: bind on all interfaces.
144  pub const DEFAULT_LISTEN_ADDRESS: &'static str = "0.0.0.0";
145
146  /// Effective default port when neither config nor CLI flag supplies
147  /// one. `8443` for the HTTPS path (cert + key both present), `8080`
148  /// for plain HTTP — the latter is the typical dev / sidecar setup
149  /// where TLS is terminated upstream.
150  pub fn default_port(has_tls: bool) -> u16 {
151    if has_tls { 8443 } else { 8080 }
152  }
153}
154
155/// Default global request timeout — 600s (10 min). Bumped from 300s
156/// after `GET /cache/configuration` (and other cross-resource fan-out
157/// endpoints that hit CFS + BSS + IMS concurrently) started 408'ing
158/// against busy sites where any single upstream fetch can stall on
159/// the CSM Envoy reset window. 10 min still bounds truly hung
160/// requests while letting heavy-but-healthy ones through. Override
161/// via `request_timeout_secs` in `server.toml` if your deployment
162/// needs a different ceiling.
163fn default_shutdown_grace_period_secs() -> u64 {
164  30
165}
166
167fn default_request_timeout_secs() -> u64 {
168  600
169}
170
171/// Top-level configuration for the `manta-server` binary.
172///
173/// Persisted as TOML under `~/.config/manta/server.toml` and loaded
174/// once at startup. Has no notion of an "active" site — the server
175/// hosts every entry in [`Self::sites`] simultaneously and clients
176/// select per-request via the `X-Manta-Site` header.
177///
178/// See the module-level docs for the on-disk layout.
179#[derive(Serialize, Deserialize, Debug)]
180pub struct ServerConfiguration {
181  /// `EnvFilter` directive for the tracing subscriber.
182  pub log: String,
183  /// Network / TLS / console / rate-limit knobs for the HTTPS server.
184  pub server: ServerSettings,
185  /// Per-site backend connection details, keyed by site name. The
186  /// `X-Manta-Site` header on each request picks which one to route to.
187  pub sites: HashMap<String, Site>,
188  /// Optional Kafka audit forwarder (typically used for `/auth/*`
189  /// attempts). When `None`, the server emits no audit messages.
190  pub auditor: Option<Auditor>,
191}
192
193#[cfg(test)]
194mod tests {
195  use super::*;
196
197  #[test]
198  fn site_deserialize_missing_backend_fails() {
199    let bad_toml = r#"
200      shasta_base_url = "https://api.example.com"
201      root_ca_cert_file = "cert.pem"
202      # missing backend
203    "#;
204    let result = toml::from_str::<Site>(bad_toml);
205    assert!(result.is_err());
206  }
207
208  #[test]
209  fn backend_technology_as_str() {
210    assert_eq!(BackendTechnology::Csm.as_str(), "csm");
211    assert_eq!(BackendTechnology::Ochami.as_str(), "ochami");
212  }
213
214  #[test]
215  fn backend_technology_roundtrip_toml() {
216    // Verify TOML serializes as lowercase "csm" / "ochami"
217    #[derive(Serialize, Deserialize)]
218    struct Wrapper {
219      backend: BackendTechnology,
220    }
221    let w = Wrapper {
222      backend: BackendTechnology::Csm,
223    };
224    let s = toml::to_string(&w).unwrap();
225    assert!(s.contains("\"csm\"") || s.contains("csm"));
226    let parsed: Wrapper = toml::from_str(&s).unwrap();
227    assert_eq!(parsed.backend, BackendTechnology::Csm);
228  }
229
230  fn make_minimal_site() -> Site {
231    Site {
232      backend: BackendTechnology::Csm,
233      socks5_proxy: None,
234      shasta_base_url: "https://api.example.com".to_string(),
235      k8s: None,
236      root_ca_cert_file: "cert.pem".to_string(),
237    }
238  }
239
240  #[test]
241  fn server_configuration_roundtrip_toml_minimal() {
242    let mut sites = HashMap::new();
243    sites.insert("alps".to_string(), make_minimal_site());
244    let cfg = ServerConfiguration {
245      log: "info".to_string(),
246      server: ServerSettings {
247        listen_address: Some("0.0.0.0".to_string()),
248        port: Some(8443),
249        cert: Some("/etc/manta/tls/server.crt".to_string()),
250        key: Some("/etc/manta/tls/server.key".to_string()),
251        console_inactivity_timeout_secs: 1800,
252        auth_rate_limit_per_minute: Some(60),
253        request_timeout_secs: 300,
254        shutdown_grace_period_secs: 30,
255        migrate_backup_root: None,
256        allow_http: false,
257      },
258      sites,
259      auditor: None,
260    };
261    let toml_str = toml::to_string(&cfg).unwrap();
262    let parsed: ServerConfiguration = toml::from_str(&toml_str).unwrap();
263    assert_eq!(parsed.server.port, Some(8443));
264    assert_eq!(parsed.server.listen_address.as_deref(), Some("0.0.0.0"));
265    assert_eq!(parsed.server.console_inactivity_timeout_secs, 1800);
266    assert_eq!(parsed.server.request_timeout_secs, 300);
267    assert_eq!(
268      parsed.server.cert.as_deref(),
269      Some("/etc/manta/tls/server.crt")
270    );
271  }
272
273  /// Default port helper: 8443 when TLS is configured, 8080
274  /// otherwise. Used by `manta-server::main` when no `port` is set
275  /// in config or on the CLI.
276  #[test]
277  fn server_settings_default_port_depends_on_tls() {
278    assert_eq!(ServerSettings::default_port(true), 8443);
279    assert_eq!(ServerSettings::default_port(false), 8080);
280  }
281
282  /// power_timeout_secs is gone — confirm the surrounding
283  /// timeout-related fields still default correctly when the only
284  /// remaining knob is absent.
285  #[test]
286  fn server_settings_request_timeout_secs_defaults_to_600() {
287    let toml_str = r#"
288      listen_address = "0.0.0.0"
289      port = 8443
290      console_inactivity_timeout_secs = 1800
291    "#;
292    let parsed: ServerSettings = toml::from_str(toml_str).unwrap();
293    assert_eq!(parsed.request_timeout_secs, 600);
294  }
295
296  /// `[server]` block with neither `listen_address` nor `port`
297  /// supplied — both fields deserialise as `None`, leaving the
298  /// effective values to be filled in at startup time. Confirms the
299  /// schema-level back-compat for the new defaults.
300  #[test]
301  fn server_settings_listen_address_and_port_default_to_none() {
302    let toml_str = r#"
303      console_inactivity_timeout_secs = 1800
304    "#;
305    let parsed: ServerSettings = toml::from_str(toml_str).unwrap();
306    assert!(parsed.listen_address.is_none());
307    assert!(parsed.port.is_none());
308  }
309
310  /// Existing server.toml files that pre-date the request_timeout
311  /// field must keep working — the field falls back to its default.
312  #[test]
313  fn server_settings_request_timeout_field_defaults_when_omitted() {
314    let toml_str = r#"
315      listen_address = "0.0.0.0"
316      port = 8443
317      console_inactivity_timeout_secs = 1800
318    "#;
319    let parsed: ServerSettings = toml::from_str(toml_str).unwrap();
320    assert_eq!(parsed.request_timeout_secs, 600);
321  }
322
323  #[test]
324  fn server_configuration_deserialize_missing_server_section_fails() {
325    let bad_toml = r#"
326      log = "info"
327      [sites]
328    "#;
329    let result = toml::from_str::<ServerConfiguration>(bad_toml);
330    assert!(result.is_err());
331  }
332
333  #[test]
334  fn server_settings_optional_tls_paths() {
335    // TLS cert/key are optional in the schema — flags can supply them
336    // at runtime when the config omits them.
337    let toml_str = r#"
338      listen_address = "0.0.0.0"
339      port = 8443
340      console_inactivity_timeout_secs = 1800
341    "#;
342    let parsed: ServerSettings = toml::from_str(toml_str).unwrap();
343    assert!(parsed.cert.is_none());
344    assert!(parsed.key.is_none());
345  }
346}