]>
Commit | Line | Data |
---|---|---|
1 | use std::sync::Arc; | |
2 | use std::ffi::OsString; | |
3 | use std::path::Path; | |
4 | ||
5 | use anyhow::{bail, format_err, Error}; | |
6 | use futures::*; | |
7 | use hyper; | |
8 | use openssl::ssl::{SslMethod, SslAcceptor, SslFiletype}; | |
9 | ||
10 | use proxmox::try_block; | |
11 | use proxmox::api::RpcEnvironmentType; | |
12 | use proxmox::sys::linux::procfs::mountinfo::{Device, MountInfo}; | |
13 | ||
14 | use proxmox_backup::configdir; | |
15 | use proxmox_backup::buildcfg; | |
16 | use proxmox_backup::server; | |
17 | use proxmox_backup::tools::daemon; | |
18 | use proxmox_backup::server::{ApiConfig, rest::*}; | |
19 | use proxmox_backup::auth_helpers::*; | |
20 | use proxmox_backup::tools::disks::{ DiskManage, zfs::zfs_pool_stats }; | |
21 | ||
22 | fn main() { | |
23 | if let Err(err) = proxmox_backup::tools::runtime::main(run()) { | |
24 | eprintln!("Error: {}", err); | |
25 | std::process::exit(-1); | |
26 | } | |
27 | } | |
28 | ||
29 | async fn run() -> Result<(), Error> { | |
30 | if let Err(err) = syslog::init( | |
31 | syslog::Facility::LOG_DAEMON, | |
32 | log::LevelFilter::Info, | |
33 | Some("proxmox-backup-proxy")) { | |
34 | bail!("unable to inititialize syslog - {}", err); | |
35 | } | |
36 | ||
37 | let _ = public_auth_key(); // load with lazy_static | |
38 | let _ = csrf_secret(); // load with lazy_static | |
39 | ||
40 | let mut config = ApiConfig::new( | |
41 | buildcfg::JS_DIR, &proxmox_backup::api2::ROUTER, RpcEnvironmentType::PUBLIC)?; | |
42 | ||
43 | // add default dirs which includes jquery and bootstrap | |
44 | // my $base = '/usr/share/libpve-http-server-perl'; | |
45 | // add_dirs($self->{dirs}, '/css/' => "$base/css/"); | |
46 | // add_dirs($self->{dirs}, '/js/' => "$base/js/"); | |
47 | // add_dirs($self->{dirs}, '/fonts/' => "$base/fonts/"); | |
48 | config.add_alias("novnc", "/usr/share/novnc-pve"); | |
49 | config.add_alias("extjs", "/usr/share/javascript/extjs"); | |
50 | config.add_alias("fontawesome", "/usr/share/fonts-font-awesome"); | |
51 | config.add_alias("xtermjs", "/usr/share/pve-xtermjs"); | |
52 | config.add_alias("widgettoolkit", "/usr/share/javascript/proxmox-widget-toolkit"); | |
53 | config.add_alias("css", "/usr/share/javascript/proxmox-backup/css"); | |
54 | config.add_alias("docs", "/usr/share/doc/proxmox-backup/html"); | |
55 | ||
56 | let rest_server = RestServer::new(config); | |
57 | ||
58 | //openssl req -x509 -newkey rsa:4096 -keyout /etc/proxmox-backup/proxy.key -out /etc/proxmox-backup/proxy.pem -nodes | |
59 | let key_path = configdir!("/proxy.key"); | |
60 | let cert_path = configdir!("/proxy.pem"); | |
61 | ||
62 | let mut acceptor = SslAcceptor::mozilla_intermediate(SslMethod::tls()).unwrap(); | |
63 | acceptor.set_private_key_file(key_path, SslFiletype::PEM) | |
64 | .map_err(|err| format_err!("unable to read proxy key {} - {}", key_path, err))?; | |
65 | acceptor.set_certificate_chain_file(cert_path) | |
66 | .map_err(|err| format_err!("unable to read proxy cert {} - {}", cert_path, err))?; | |
67 | acceptor.check_private_key().unwrap(); | |
68 | ||
69 | let acceptor = Arc::new(acceptor.build()); | |
70 | ||
71 | let server = daemon::create_daemon( | |
72 | ([0,0,0,0,0,0,0,0], 8007).into(), | |
73 | |listener, ready| { | |
74 | let connections = proxmox_backup::tools::async_io::StaticIncoming::from(listener) | |
75 | .map_err(Error::from) | |
76 | .try_filter_map(move |(sock, _addr)| { | |
77 | let acceptor = Arc::clone(&acceptor); | |
78 | async move { | |
79 | sock.set_nodelay(true).unwrap(); | |
80 | sock.set_send_buffer_size(1024*1024).unwrap(); | |
81 | sock.set_recv_buffer_size(1024*1024).unwrap(); | |
82 | Ok(tokio_openssl::accept(&acceptor, sock) | |
83 | .await | |
84 | .ok() // handshake errors aren't be fatal, so return None to filter | |
85 | ) | |
86 | } | |
87 | }); | |
88 | let connections = proxmox_backup::tools::async_io::HyperAccept(connections); | |
89 | ||
90 | Ok(ready | |
91 | .and_then(|_| hyper::Server::builder(connections) | |
92 | .serve(rest_server) | |
93 | .with_graceful_shutdown(server::shutdown_future()) | |
94 | .map_err(Error::from) | |
95 | ) | |
96 | .map_err(|err| eprintln!("server error: {}", err)) | |
97 | .map(|_| ()) | |
98 | ) | |
99 | }, | |
100 | ); | |
101 | ||
102 | daemon::systemd_notify(daemon::SystemdNotify::Ready)?; | |
103 | ||
104 | let init_result: Result<(), Error> = try_block!({ | |
105 | server::create_task_control_socket()?; | |
106 | server::server_state_init()?; | |
107 | Ok(()) | |
108 | }); | |
109 | ||
110 | if let Err(err) = init_result { | |
111 | bail!("unable to start daemon - {}", err); | |
112 | } | |
113 | ||
114 | start_task_scheduler(); | |
115 | start_stat_generator(); | |
116 | ||
117 | server.await?; | |
118 | log::info!("server shutting down, waiting for active workers to complete"); | |
119 | proxmox_backup::server::last_worker_future().await?; | |
120 | log::info!("done - exit server"); | |
121 | ||
122 | Ok(()) | |
123 | } | |
124 | ||
125 | fn start_stat_generator() { | |
126 | let abort_future = server::shutdown_future(); | |
127 | let future = Box::pin(run_stat_generator()); | |
128 | let task = futures::future::select(future, abort_future); | |
129 | tokio::spawn(task.map(|_| ())); | |
130 | } | |
131 | ||
132 | fn start_task_scheduler() { | |
133 | let abort_future = server::shutdown_future(); | |
134 | let future = Box::pin(run_task_scheduler()); | |
135 | let task = futures::future::select(future, abort_future); | |
136 | tokio::spawn(task.map(|_| ())); | |
137 | } | |
138 | ||
139 | use std::time:: {Instant, Duration, SystemTime, UNIX_EPOCH}; | |
140 | ||
141 | fn next_minute() -> Result<Instant, Error> { | |
142 | let epoch_now = SystemTime::now().duration_since(UNIX_EPOCH)?; | |
143 | let epoch_next = Duration::from_secs((epoch_now.as_secs()/60 + 1)*60); | |
144 | Ok(Instant::now() + epoch_next - epoch_now) | |
145 | } | |
146 | ||
147 | async fn run_task_scheduler() { | |
148 | ||
149 | let mut count: usize = 0; | |
150 | ||
151 | loop { | |
152 | count += 1; | |
153 | ||
154 | let delay_target = match next_minute() { // try to run very minute | |
155 | Ok(d) => d, | |
156 | Err(err) => { | |
157 | eprintln!("task scheduler: compute next minute failed - {}", err); | |
158 | tokio::time::delay_until(tokio::time::Instant::from_std(Instant::now() + Duration::from_secs(60))).await; | |
159 | continue; | |
160 | } | |
161 | }; | |
162 | ||
163 | if count > 2 { // wait 1..2 minutes before starting | |
164 | match schedule_tasks().catch_unwind().await { | |
165 | Err(panic) => { | |
166 | match panic.downcast::<&str>() { | |
167 | Ok(msg) => { | |
168 | eprintln!("task scheduler panic: {}", msg); | |
169 | } | |
170 | Err(_) => { | |
171 | eprintln!("task scheduler panic - unknown type"); | |
172 | } | |
173 | } | |
174 | } | |
175 | Ok(Err(err)) => { | |
176 | eprintln!("task scheduler failed - {:?}", err); | |
177 | } | |
178 | Ok(Ok(_)) => {} | |
179 | } | |
180 | } | |
181 | ||
182 | tokio::time::delay_until(tokio::time::Instant::from_std(delay_target)).await; | |
183 | } | |
184 | } | |
185 | ||
186 | async fn schedule_tasks() -> Result<(), Error> { | |
187 | ||
188 | schedule_datastore_garbage_collection().await; | |
189 | schedule_datastore_prune().await; | |
190 | schedule_datastore_sync_jobs().await; | |
191 | ||
192 | Ok(()) | |
193 | } | |
194 | ||
195 | fn lookup_last_worker(worker_type: &str, worker_id: &str) -> Result<Option<server::UPID>, Error> { | |
196 | ||
197 | let list = proxmox_backup::server::read_task_list()?; | |
198 | ||
199 | let mut last: Option<&server::UPID> = None; | |
200 | ||
201 | for entry in list.iter() { | |
202 | if entry.upid.worker_type == worker_type { | |
203 | if let Some(ref id) = entry.upid.worker_id { | |
204 | if id == worker_id { | |
205 | match last { | |
206 | Some(ref upid) => { | |
207 | if upid.starttime < entry.upid.starttime { | |
208 | last = Some(&entry.upid) | |
209 | } | |
210 | } | |
211 | None => { | |
212 | last = Some(&entry.upid) | |
213 | } | |
214 | } | |
215 | } | |
216 | } | |
217 | } | |
218 | } | |
219 | ||
220 | Ok(last.cloned()) | |
221 | } | |
222 | ||
223 | ||
224 | async fn schedule_datastore_garbage_collection() { | |
225 | ||
226 | use proxmox_backup::backup::DataStore; | |
227 | use proxmox_backup::server::{UPID, WorkerTask}; | |
228 | use proxmox_backup::config::datastore::{self, DataStoreConfig}; | |
229 | use proxmox_backup::tools::systemd::time::{ | |
230 | parse_calendar_event, compute_next_event}; | |
231 | ||
232 | let config = match datastore::config() { | |
233 | Err(err) => { | |
234 | eprintln!("unable to read datastore config - {}", err); | |
235 | return; | |
236 | } | |
237 | Ok((config, _digest)) => config, | |
238 | }; | |
239 | ||
240 | for (store, (_, store_config)) in config.sections { | |
241 | let datastore = match DataStore::lookup_datastore(&store) { | |
242 | Ok(datastore) => datastore, | |
243 | Err(err) => { | |
244 | eprintln!("lookup_datastore failed - {}", err); | |
245 | continue; | |
246 | } | |
247 | }; | |
248 | ||
249 | let store_config: DataStoreConfig = match serde_json::from_value(store_config) { | |
250 | Ok(c) => c, | |
251 | Err(err) => { | |
252 | eprintln!("datastore config from_value failed - {}", err); | |
253 | continue; | |
254 | } | |
255 | }; | |
256 | ||
257 | let event_str = match store_config.gc_schedule { | |
258 | Some(event_str) => event_str, | |
259 | None => continue, | |
260 | }; | |
261 | ||
262 | let event = match parse_calendar_event(&event_str) { | |
263 | Ok(event) => event, | |
264 | Err(err) => { | |
265 | eprintln!("unable to parse schedule '{}' - {}", event_str, err); | |
266 | continue; | |
267 | } | |
268 | }; | |
269 | ||
270 | if datastore.garbage_collection_running() { continue; } | |
271 | ||
272 | let worker_type = "garbage_collection"; | |
273 | ||
274 | let stat = datastore.last_gc_status(); | |
275 | let last = if let Some(upid_str) = stat.upid { | |
276 | match upid_str.parse::<UPID>() { | |
277 | Ok(upid) => upid.starttime, | |
278 | Err(err) => { | |
279 | eprintln!("unable to parse upid '{}' - {}", upid_str, err); | |
280 | continue; | |
281 | } | |
282 | } | |
283 | } else { | |
284 | match lookup_last_worker(worker_type, &store) { | |
285 | Ok(Some(upid)) => upid.starttime, | |
286 | Ok(None) => 0, | |
287 | Err(err) => { | |
288 | eprintln!("lookup_last_job_start failed: {}", err); | |
289 | continue; | |
290 | } | |
291 | } | |
292 | }; | |
293 | ||
294 | let next = match compute_next_event(&event, last, false) { | |
295 | Ok(next) => next, | |
296 | Err(err) => { | |
297 | eprintln!("compute_next_event for '{}' failed - {}", event_str, err); | |
298 | continue; | |
299 | } | |
300 | }; | |
301 | let now = match SystemTime::now().duration_since(UNIX_EPOCH) { | |
302 | Ok(epoch_now) => epoch_now.as_secs() as i64, | |
303 | Err(err) => { | |
304 | eprintln!("query system time failed - {}", err); | |
305 | continue; | |
306 | } | |
307 | }; | |
308 | if next > now { continue; } | |
309 | ||
310 | let store2 = store.clone(); | |
311 | ||
312 | if let Err(err) = WorkerTask::new_thread( | |
313 | worker_type, | |
314 | Some(store.clone()), | |
315 | "backup@pam", | |
316 | false, | |
317 | move |worker| { | |
318 | worker.log(format!("starting garbage collection on store {}", store)); | |
319 | worker.log(format!("task triggered by schedule '{}'", event_str)); | |
320 | datastore.garbage_collection(&worker) | |
321 | } | |
322 | ) { | |
323 | eprintln!("unable to start garbage collection on store {} - {}", store2, err); | |
324 | } | |
325 | } | |
326 | } | |
327 | ||
328 | async fn schedule_datastore_prune() { | |
329 | ||
330 | use proxmox_backup::backup::{ | |
331 | PruneOptions, DataStore, BackupGroup, BackupDir, compute_prune_info}; | |
332 | use proxmox_backup::server::{WorkerTask}; | |
333 | use proxmox_backup::config::datastore::{self, DataStoreConfig}; | |
334 | use proxmox_backup::tools::systemd::time::{ | |
335 | parse_calendar_event, compute_next_event}; | |
336 | ||
337 | let config = match datastore::config() { | |
338 | Err(err) => { | |
339 | eprintln!("unable to read datastore config - {}", err); | |
340 | return; | |
341 | } | |
342 | Ok((config, _digest)) => config, | |
343 | }; | |
344 | ||
345 | for (store, (_, store_config)) in config.sections { | |
346 | let datastore = match DataStore::lookup_datastore(&store) { | |
347 | Ok(datastore) => datastore, | |
348 | Err(err) => { | |
349 | eprintln!("lookup_datastore '{}' failed - {}", store, err); | |
350 | continue; | |
351 | } | |
352 | }; | |
353 | ||
354 | let store_config: DataStoreConfig = match serde_json::from_value(store_config) { | |
355 | Ok(c) => c, | |
356 | Err(err) => { | |
357 | eprintln!("datastore '{}' config from_value failed - {}", store, err); | |
358 | continue; | |
359 | } | |
360 | }; | |
361 | ||
362 | let event_str = match store_config.prune_schedule { | |
363 | Some(event_str) => event_str, | |
364 | None => continue, | |
365 | }; | |
366 | ||
367 | let prune_options = PruneOptions { | |
368 | keep_last: store_config.keep_last, | |
369 | keep_hourly: store_config.keep_hourly, | |
370 | keep_daily: store_config.keep_daily, | |
371 | keep_weekly: store_config.keep_weekly, | |
372 | keep_monthly: store_config.keep_monthly, | |
373 | keep_yearly: store_config.keep_yearly, | |
374 | }; | |
375 | ||
376 | if !prune_options.keeps_something() { // no prune settings - keep all | |
377 | continue; | |
378 | } | |
379 | ||
380 | let event = match parse_calendar_event(&event_str) { | |
381 | Ok(event) => event, | |
382 | Err(err) => { | |
383 | eprintln!("unable to parse schedule '{}' - {}", event_str, err); | |
384 | continue; | |
385 | } | |
386 | }; | |
387 | ||
388 | let worker_type = "prune"; | |
389 | ||
390 | let last = match lookup_last_worker(worker_type, &store) { | |
391 | Ok(Some(upid)) => { | |
392 | if proxmox_backup::server::worker_is_active_local(&upid) { | |
393 | continue; | |
394 | } | |
395 | upid.starttime | |
396 | } | |
397 | Ok(None) => 0, | |
398 | Err(err) => { | |
399 | eprintln!("lookup_last_job_start failed: {}", err); | |
400 | continue; | |
401 | } | |
402 | }; | |
403 | ||
404 | let next = match compute_next_event(&event, last, false) { | |
405 | Ok(next) => next, | |
406 | Err(err) => { | |
407 | eprintln!("compute_next_event for '{}' failed - {}", event_str, err); | |
408 | continue; | |
409 | } | |
410 | }; | |
411 | ||
412 | let now = match SystemTime::now().duration_since(UNIX_EPOCH) { | |
413 | Ok(epoch_now) => epoch_now.as_secs() as i64, | |
414 | Err(err) => { | |
415 | eprintln!("query system time failed - {}", err); | |
416 | continue; | |
417 | } | |
418 | }; | |
419 | if next > now { continue; } | |
420 | ||
421 | let store2 = store.clone(); | |
422 | ||
423 | if let Err(err) = WorkerTask::new_thread( | |
424 | worker_type, | |
425 | Some(store.clone()), | |
426 | "backup@pam", | |
427 | false, | |
428 | move |worker| { | |
429 | worker.log(format!("Starting datastore prune on store \"{}\"", store)); | |
430 | worker.log(format!("task triggered by schedule '{}'", event_str)); | |
431 | worker.log(format!("retention options: {}", prune_options.cli_options_string())); | |
432 | ||
433 | let base_path = datastore.base_path(); | |
434 | ||
435 | let groups = BackupGroup::list_groups(&base_path)?; | |
436 | for group in groups { | |
437 | let list = group.list_backups(&base_path)?; | |
438 | let mut prune_info = compute_prune_info(list, &prune_options)?; | |
439 | prune_info.reverse(); // delete older snapshots first | |
440 | ||
441 | worker.log(format!("Starting prune on store \"{}\" group \"{}/{}\"", | |
442 | store, group.backup_type(), group.backup_id())); | |
443 | ||
444 | for (info, keep) in prune_info { | |
445 | worker.log(format!( | |
446 | "{} {}/{}/{}", | |
447 | if keep { "keep" } else { "remove" }, | |
448 | group.backup_type(), group.backup_id(), | |
449 | BackupDir::backup_time_to_string(info.backup_dir.backup_time()))); | |
450 | ||
451 | if !keep { | |
452 | datastore.remove_backup_dir(&info.backup_dir)?; | |
453 | } | |
454 | } | |
455 | } | |
456 | ||
457 | Ok(()) | |
458 | } | |
459 | ) { | |
460 | eprintln!("unable to start datastore prune on store {} - {}", store2, err); | |
461 | } | |
462 | } | |
463 | } | |
464 | ||
465 | async fn schedule_datastore_sync_jobs() { | |
466 | ||
467 | use proxmox_backup::{ | |
468 | backup::DataStore, | |
469 | client::{ HttpClient, HttpClientOptions, BackupRepository, pull::pull_store }, | |
470 | server::{ WorkerTask }, | |
471 | config::{ sync::{self, SyncJobConfig}, remote::{self, Remote} }, | |
472 | tools::systemd::time::{ parse_calendar_event, compute_next_event }, | |
473 | }; | |
474 | ||
475 | let config = match sync::config() { | |
476 | Err(err) => { | |
477 | eprintln!("unable to read sync job config - {}", err); | |
478 | return; | |
479 | } | |
480 | Ok((config, _digest)) => config, | |
481 | }; | |
482 | ||
483 | let remote_config = match remote::config() { | |
484 | Err(err) => { | |
485 | eprintln!("unable to read remote config - {}", err); | |
486 | return; | |
487 | } | |
488 | Ok((config, _digest)) => config, | |
489 | }; | |
490 | ||
491 | for (job_id, (_, job_config)) in config.sections { | |
492 | let job_config: SyncJobConfig = match serde_json::from_value(job_config) { | |
493 | Ok(c) => c, | |
494 | Err(err) => { | |
495 | eprintln!("sync job config from_value failed - {}", err); | |
496 | continue; | |
497 | } | |
498 | }; | |
499 | ||
500 | let event_str = match job_config.schedule { | |
501 | Some(ref event_str) => event_str.clone(), | |
502 | None => continue, | |
503 | }; | |
504 | ||
505 | let event = match parse_calendar_event(&event_str) { | |
506 | Ok(event) => event, | |
507 | Err(err) => { | |
508 | eprintln!("unable to parse schedule '{}' - {}", event_str, err); | |
509 | continue; | |
510 | } | |
511 | }; | |
512 | ||
513 | let worker_type = "sync"; | |
514 | ||
515 | let last = match lookup_last_worker(worker_type, &job_id) { | |
516 | Ok(Some(upid)) => { | |
517 | if proxmox_backup::server::worker_is_active_local(&upid) { | |
518 | continue; | |
519 | } | |
520 | upid.starttime | |
521 | }, | |
522 | Ok(None) => 0, | |
523 | Err(err) => { | |
524 | eprintln!("lookup_last_job_start failed: {}", err); | |
525 | continue; | |
526 | } | |
527 | }; | |
528 | ||
529 | let next = match compute_next_event(&event, last, false) { | |
530 | Ok(next) => next, | |
531 | Err(err) => { | |
532 | eprintln!("compute_next_event for '{}' failed - {}", event_str, err); | |
533 | continue; | |
534 | } | |
535 | }; | |
536 | ||
537 | let now = match SystemTime::now().duration_since(UNIX_EPOCH) { | |
538 | Ok(epoch_now) => epoch_now.as_secs() as i64, | |
539 | Err(err) => { | |
540 | eprintln!("query system time failed - {}", err); | |
541 | continue; | |
542 | } | |
543 | }; | |
544 | if next > now { continue; } | |
545 | ||
546 | ||
547 | let job_id2 = job_id.clone(); | |
548 | ||
549 | let tgt_store = match DataStore::lookup_datastore(&job_config.store) { | |
550 | Ok(datastore) => datastore, | |
551 | Err(err) => { | |
552 | eprintln!("lookup_datastore '{}' failed - {}", job_config.store, err); | |
553 | continue; | |
554 | } | |
555 | }; | |
556 | ||
557 | let remote: Remote = match remote_config.lookup("remote", &job_config.remote) { | |
558 | Ok(remote) => remote, | |
559 | Err(err) => { | |
560 | eprintln!("remote_config lookup failed: {}", err); | |
561 | continue; | |
562 | } | |
563 | }; | |
564 | ||
565 | let username = String::from("backup@pam"); | |
566 | ||
567 | let delete = job_config.remove_vanished.unwrap_or(true); | |
568 | ||
569 | if let Err(err) = WorkerTask::spawn( | |
570 | worker_type, | |
571 | Some(job_id.clone()), | |
572 | &username.clone(), | |
573 | false, | |
574 | move |worker| async move { | |
575 | worker.log(format!("Starting datastore sync job '{}'", job_id)); | |
576 | worker.log(format!("task triggered by schedule '{}'", event_str)); | |
577 | worker.log(format!("Sync datastore '{}' from '{}/{}'", | |
578 | job_config.store, job_config.remote, job_config.remote_store)); | |
579 | ||
580 | let options = HttpClientOptions::new() | |
581 | .password(Some(remote.password.clone())) | |
582 | .fingerprint(remote.fingerprint.clone()); | |
583 | ||
584 | let client = HttpClient::new(&remote.host, &remote.userid, options)?; | |
585 | let _auth_info = client.login() // make sure we can auth | |
586 | .await | |
587 | .map_err(|err| format_err!("remote connection to '{}' failed - {}", remote.host, err))?; | |
588 | ||
589 | let src_repo = BackupRepository::new(Some(remote.userid), Some(remote.host), job_config.remote_store); | |
590 | ||
591 | pull_store(&worker, &client, &src_repo, tgt_store, delete, username).await?; | |
592 | ||
593 | Ok(()) | |
594 | } | |
595 | ) { | |
596 | eprintln!("unable to start datastore sync job {} - {}", job_id2, err); | |
597 | } | |
598 | } | |
599 | } | |
600 | ||
601 | async fn run_stat_generator() { | |
602 | ||
603 | loop { | |
604 | let delay_target = Instant::now() + Duration::from_secs(10); | |
605 | ||
606 | generate_host_stats().await; | |
607 | ||
608 | tokio::time::delay_until(tokio::time::Instant::from_std(delay_target)).await; | |
609 | } | |
610 | ||
611 | } | |
612 | ||
613 | fn rrd_update_gauge(name: &str, value: f64) { | |
614 | use proxmox_backup::rrd; | |
615 | if let Err(err) = rrd::update_value(name, value, rrd::DST::Gauge) { | |
616 | eprintln!("rrd::update_value '{}' failed - {}", name, err); | |
617 | } | |
618 | } | |
619 | ||
620 | fn rrd_update_derive(name: &str, value: f64) { | |
621 | use proxmox_backup::rrd; | |
622 | if let Err(err) = rrd::update_value(name, value, rrd::DST::Derive) { | |
623 | eprintln!("rrd::update_value '{}' failed - {}", name, err); | |
624 | } | |
625 | } | |
626 | ||
627 | async fn generate_host_stats() { | |
628 | use proxmox::sys::linux::procfs::{ | |
629 | read_meminfo, read_proc_stat, read_proc_net_dev, read_loadavg}; | |
630 | use proxmox_backup::config::datastore; | |
631 | ||
632 | ||
633 | proxmox_backup::tools::runtime::block_in_place(move || { | |
634 | ||
635 | match read_proc_stat() { | |
636 | Ok(stat) => { | |
637 | rrd_update_gauge("host/cpu", stat.cpu); | |
638 | rrd_update_gauge("host/iowait", stat.iowait_percent); | |
639 | } | |
640 | Err(err) => { | |
641 | eprintln!("read_proc_stat failed - {}", err); | |
642 | } | |
643 | } | |
644 | ||
645 | match read_meminfo() { | |
646 | Ok(meminfo) => { | |
647 | rrd_update_gauge("host/memtotal", meminfo.memtotal as f64); | |
648 | rrd_update_gauge("host/memused", meminfo.memused as f64); | |
649 | rrd_update_gauge("host/swaptotal", meminfo.swaptotal as f64); | |
650 | rrd_update_gauge("host/swapused", meminfo.swapused as f64); | |
651 | } | |
652 | Err(err) => { | |
653 | eprintln!("read_meminfo failed - {}", err); | |
654 | } | |
655 | } | |
656 | ||
657 | match read_proc_net_dev() { | |
658 | Ok(netdev) => { | |
659 | use proxmox_backup::config::network::is_physical_nic; | |
660 | let mut netin = 0; | |
661 | let mut netout = 0; | |
662 | for item in netdev { | |
663 | if !is_physical_nic(&item.device) { continue; } | |
664 | netin += item.receive; | |
665 | netout += item.send; | |
666 | } | |
667 | rrd_update_derive("host/netin", netin as f64); | |
668 | rrd_update_derive("host/netout", netout as f64); | |
669 | } | |
670 | Err(err) => { | |
671 | eprintln!("read_prox_net_dev failed - {}", err); | |
672 | } | |
673 | } | |
674 | ||
675 | match read_loadavg() { | |
676 | Ok(loadavg) => { | |
677 | rrd_update_gauge("host/loadavg", loadavg.0 as f64); | |
678 | } | |
679 | Err(err) => { | |
680 | eprintln!("read_loadavg failed - {}", err); | |
681 | } | |
682 | } | |
683 | ||
684 | let disk_manager = DiskManage::new(); | |
685 | ||
686 | gather_disk_stats(disk_manager.clone(), Path::new("/"), "host"); | |
687 | ||
688 | match datastore::config() { | |
689 | Ok((config, _)) => { | |
690 | let datastore_list: Vec<datastore::DataStoreConfig> = | |
691 | config.convert_to_typed_array("datastore").unwrap_or(Vec::new()); | |
692 | ||
693 | for config in datastore_list { | |
694 | ||
695 | let rrd_prefix = format!("datastore/{}", config.name); | |
696 | let path = std::path::Path::new(&config.path); | |
697 | gather_disk_stats(disk_manager.clone(), path, &rrd_prefix); | |
698 | } | |
699 | } | |
700 | Err(err) => { | |
701 | eprintln!("read datastore config failed - {}", err); | |
702 | } | |
703 | } | |
704 | ||
705 | }); | |
706 | } | |
707 | ||
708 | ||
709 | fn gather_disk_stats(disk_manager: Arc<DiskManage>, path: &Path, rrd_prefix: &str) { | |
710 | ||
711 | match disk_usage(path) { | |
712 | Ok((total, used, _avail)) => { | |
713 | let rrd_key = format!("{}/total", rrd_prefix); | |
714 | rrd_update_gauge(&rrd_key, total as f64); | |
715 | let rrd_key = format!("{}/used", rrd_prefix); | |
716 | rrd_update_gauge(&rrd_key, used as f64); | |
717 | } | |
718 | Err(err) => { | |
719 | eprintln!("read disk_usage on {:?} failed - {}", path, err); | |
720 | } | |
721 | } | |
722 | ||
723 | match disk_manager.mount_info() { | |
724 | Ok(mountinfo) => { | |
725 | if let Some((fs_type, device, source)) = find_mounted_device(mountinfo, path) { | |
726 | let mut device_stat = None; | |
727 | match fs_type.as_str() { | |
728 | "zfs" => { | |
729 | if let Some(pool) = source { | |
730 | match zfs_pool_stats(&pool) { | |
731 | Ok(stat) => device_stat = stat, | |
732 | Err(err) => eprintln!("zfs_pool_stats({:?}) failed - {}", pool, err), | |
733 | } | |
734 | } | |
735 | } | |
736 | _ => { | |
737 | if let Ok(disk) = disk_manager.clone().disk_by_dev_num(device.into_dev_t()) { | |
738 | match disk.read_stat() { | |
739 | Ok(stat) => device_stat = stat, | |
740 | Err(err) => eprintln!("disk.read_stat {:?} failed - {}", path, err), | |
741 | } | |
742 | } | |
743 | } | |
744 | } | |
745 | if let Some(stat) = device_stat { | |
746 | let rrd_key = format!("{}/read_ios", rrd_prefix); | |
747 | rrd_update_derive(&rrd_key, stat.read_ios as f64); | |
748 | let rrd_key = format!("{}/read_bytes", rrd_prefix); | |
749 | rrd_update_derive(&rrd_key, (stat.read_sectors*512) as f64); | |
750 | ||
751 | let rrd_key = format!("{}/write_ios", rrd_prefix); | |
752 | rrd_update_derive(&rrd_key, stat.write_ios as f64); | |
753 | let rrd_key = format!("{}/write_bytes", rrd_prefix); | |
754 | rrd_update_derive(&rrd_key, (stat.write_sectors*512) as f64); | |
755 | ||
756 | let rrd_key = format!("{}/io_ticks", rrd_prefix); | |
757 | rrd_update_derive(&rrd_key, (stat.io_ticks as f64)/1000.0); | |
758 | } | |
759 | } | |
760 | } | |
761 | Err(err) => { | |
762 | eprintln!("disk_manager mount_info() failed - {}", err); | |
763 | } | |
764 | } | |
765 | } | |
766 | ||
767 | // Returns (total, used, avail) | |
768 | fn disk_usage(path: &std::path::Path) -> Result<(u64, u64, u64), Error> { | |
769 | ||
770 | let mut stat: libc::statfs64 = unsafe { std::mem::zeroed() }; | |
771 | ||
772 | use nix::NixPath; | |
773 | ||
774 | let res = path.with_nix_path(|cstr| unsafe { libc::statfs64(cstr.as_ptr(), &mut stat) })?; | |
775 | nix::errno::Errno::result(res)?; | |
776 | ||
777 | let bsize = stat.f_bsize as u64; | |
778 | ||
779 | Ok((stat.f_blocks*bsize, (stat.f_blocks-stat.f_bfree)*bsize, stat.f_bavail*bsize)) | |
780 | } | |
781 | ||
782 | // Returns (fs_type, device, mount_source) | |
783 | pub fn find_mounted_device( | |
784 | mountinfo: &MountInfo, | |
785 | path: &std::path::Path, | |
786 | ) -> Option<(String, Device, Option<OsString>)> { | |
787 | ||
788 | let mut result = None; | |
789 | let mut match_len = 0; | |
790 | ||
791 | let root_path = std::path::Path::new("/"); | |
792 | for (_id, entry) in mountinfo { | |
793 | if entry.root == root_path && path.starts_with(&entry.mount_point) { | |
794 | let len = entry.mount_point.as_path().as_os_str().len(); | |
795 | if len > match_len { | |
796 | match_len = len; | |
797 | result = Some((entry.fs_type.clone(), entry.device, entry.mount_source.clone())); | |
798 | } | |
799 | } | |
800 | } | |
801 | ||
802 | result | |
803 | } |