]>
Commit | Line | Data |
---|---|---|
1 | use std::sync::Arc; | |
2 | use std::path::{Path, PathBuf}; | |
3 | use std::os::unix::io::AsRawFd; | |
4 | ||
5 | use anyhow::{bail, format_err, Error}; | |
6 | use futures::*; | |
7 | use hyper; | |
8 | use openssl::ssl::{SslMethod, SslAcceptor, SslFiletype}; | |
9 | use tokio_stream::wrappers::ReceiverStream; | |
10 | ||
11 | use proxmox::try_block; | |
12 | use proxmox::api::RpcEnvironmentType; | |
13 | ||
14 | use proxmox_backup::{ | |
15 | backup::DataStore, | |
16 | server::{ | |
17 | WorkerTask, | |
18 | ApiConfig, | |
19 | rest::*, | |
20 | jobstate::{ | |
21 | self, | |
22 | Job, | |
23 | }, | |
24 | rotate_task_log_archive, | |
25 | }, | |
26 | tools::systemd::time::{ | |
27 | parse_calendar_event, | |
28 | compute_next_event, | |
29 | }, | |
30 | }; | |
31 | ||
32 | ||
33 | use proxmox_backup::api2::types::Authid; | |
34 | use proxmox_backup::configdir; | |
35 | use proxmox_backup::buildcfg; | |
36 | use proxmox_backup::server; | |
37 | use proxmox_backup::auth_helpers::*; | |
38 | use proxmox_backup::tools::{ | |
39 | daemon, | |
40 | disks::{ | |
41 | DiskManage, | |
42 | zfs_pool_stats, | |
43 | }, | |
44 | logrotate::LogRotate, | |
45 | socket::{ | |
46 | set_tcp_keepalive, | |
47 | PROXMOX_BACKUP_TCP_KEEPALIVE_TIME, | |
48 | }, | |
49 | }; | |
50 | ||
51 | use proxmox_backup::api2::pull::do_sync_job; | |
52 | use proxmox_backup::server::do_verification_job; | |
53 | use proxmox_backup::server::do_prune_job; | |
54 | ||
55 | fn main() -> Result<(), Error> { | |
56 | proxmox_backup::tools::setup_safe_path_env(); | |
57 | ||
58 | let backup_uid = proxmox_backup::backup::backup_user()?.uid; | |
59 | let backup_gid = proxmox_backup::backup::backup_group()?.gid; | |
60 | let running_uid = nix::unistd::Uid::effective(); | |
61 | let running_gid = nix::unistd::Gid::effective(); | |
62 | ||
63 | if running_uid != backup_uid || running_gid != backup_gid { | |
64 | bail!("proxy not running as backup user or group (got uid {} gid {})", running_uid, running_gid); | |
65 | } | |
66 | ||
67 | proxmox_backup::tools::runtime::main(run()) | |
68 | } | |
69 | ||
70 | async fn run() -> Result<(), Error> { | |
71 | if let Err(err) = syslog::init( | |
72 | syslog::Facility::LOG_DAEMON, | |
73 | log::LevelFilter::Info, | |
74 | Some("proxmox-backup-proxy")) { | |
75 | bail!("unable to inititialize syslog - {}", err); | |
76 | } | |
77 | ||
78 | // Note: To debug early connection error use | |
79 | // PROXMOX_DEBUG=1 ./target/release/proxmox-backup-proxy | |
80 | let debug = std::env::var("PROXMOX_DEBUG").is_ok(); | |
81 | ||
82 | let _ = public_auth_key(); // load with lazy_static | |
83 | let _ = csrf_secret(); // load with lazy_static | |
84 | ||
85 | let mut config = ApiConfig::new( | |
86 | buildcfg::JS_DIR, &proxmox_backup::api2::ROUTER, RpcEnvironmentType::PUBLIC)?; | |
87 | ||
88 | config.add_alias("novnc", "/usr/share/novnc-pve"); | |
89 | config.add_alias("extjs", "/usr/share/javascript/extjs"); | |
90 | config.add_alias("qrcodejs", "/usr/share/javascript/qrcodejs"); | |
91 | config.add_alias("fontawesome", "/usr/share/fonts-font-awesome"); | |
92 | config.add_alias("xtermjs", "/usr/share/pve-xtermjs"); | |
93 | config.add_alias("locale", "/usr/share/pbs-i18n"); | |
94 | config.add_alias("widgettoolkit", "/usr/share/javascript/proxmox-widget-toolkit"); | |
95 | config.add_alias("docs", "/usr/share/doc/proxmox-backup/html"); | |
96 | ||
97 | let mut indexpath = PathBuf::from(buildcfg::JS_DIR); | |
98 | indexpath.push("index.hbs"); | |
99 | config.register_template("index", &indexpath)?; | |
100 | config.register_template("console", "/usr/share/pve-xtermjs/index.html.hbs")?; | |
101 | ||
102 | let mut commando_sock = server::CommandoSocket::new(server::our_ctrl_sock()); | |
103 | ||
104 | config.enable_file_log(buildcfg::API_ACCESS_LOG_FN, &mut commando_sock)?; | |
105 | ||
106 | let rest_server = RestServer::new(config); | |
107 | ||
108 | //openssl req -x509 -newkey rsa:4096 -keyout /etc/proxmox-backup/proxy.key -out /etc/proxmox-backup/proxy.pem -nodes | |
109 | let key_path = configdir!("/proxy.key"); | |
110 | let cert_path = configdir!("/proxy.pem"); | |
111 | ||
112 | let mut acceptor = SslAcceptor::mozilla_intermediate_v5(SslMethod::tls()).unwrap(); | |
113 | acceptor.set_private_key_file(key_path, SslFiletype::PEM) | |
114 | .map_err(|err| format_err!("unable to read proxy key {} - {}", key_path, err))?; | |
115 | acceptor.set_certificate_chain_file(cert_path) | |
116 | .map_err(|err| format_err!("unable to read proxy cert {} - {}", cert_path, err))?; | |
117 | acceptor.check_private_key().unwrap(); | |
118 | ||
119 | let acceptor = Arc::new(acceptor.build()); | |
120 | ||
121 | let server = daemon::create_daemon( | |
122 | ([0,0,0,0,0,0,0,0], 8007).into(), | |
123 | |listener, ready| { | |
124 | ||
125 | let connections = accept_connections(listener, acceptor, debug); | |
126 | let connections = hyper::server::accept::from_stream(ReceiverStream::new(connections)); | |
127 | ||
128 | Ok(ready | |
129 | .and_then(|_| hyper::Server::builder(connections) | |
130 | .serve(rest_server) | |
131 | .with_graceful_shutdown(server::shutdown_future()) | |
132 | .map_err(Error::from) | |
133 | ) | |
134 | .map_err(|err| eprintln!("server error: {}", err)) | |
135 | .map(|_| ()) | |
136 | ) | |
137 | }, | |
138 | "proxmox-backup-proxy.service", | |
139 | ); | |
140 | ||
141 | server::write_pid(buildcfg::PROXMOX_BACKUP_PROXY_PID_FN)?; | |
142 | daemon::systemd_notify(daemon::SystemdNotify::Ready)?; | |
143 | ||
144 | let init_result: Result<(), Error> = try_block!({ | |
145 | server::register_task_control_commands(&mut commando_sock)?; | |
146 | commando_sock.spawn()?; | |
147 | server::server_state_init()?; | |
148 | Ok(()) | |
149 | }); | |
150 | ||
151 | if let Err(err) = init_result { | |
152 | bail!("unable to start daemon - {}", err); | |
153 | } | |
154 | ||
155 | start_task_scheduler(); | |
156 | start_stat_generator(); | |
157 | ||
158 | server.await?; | |
159 | log::info!("server shutting down, waiting for active workers to complete"); | |
160 | proxmox_backup::server::last_worker_future().await?; | |
161 | log::info!("done - exit server"); | |
162 | ||
163 | Ok(()) | |
164 | } | |
165 | ||
166 | fn accept_connections( | |
167 | listener: tokio::net::TcpListener, | |
168 | acceptor: Arc<openssl::ssl::SslAcceptor>, | |
169 | debug: bool, | |
170 | ) -> tokio::sync::mpsc::Receiver<Result<std::pin::Pin<Box<tokio_openssl::SslStream<tokio::net::TcpStream>>>, Error>> { | |
171 | ||
172 | const MAX_PENDING_ACCEPTS: usize = 1024; | |
173 | ||
174 | let (sender, receiver) = tokio::sync::mpsc::channel(MAX_PENDING_ACCEPTS); | |
175 | ||
176 | let accept_counter = Arc::new(()); | |
177 | ||
178 | tokio::spawn(async move { | |
179 | loop { | |
180 | match listener.accept().await { | |
181 | Err(err) => { | |
182 | eprintln!("error accepting tcp connection: {}", err); | |
183 | } | |
184 | Ok((sock, _addr)) => { | |
185 | sock.set_nodelay(true).unwrap(); | |
186 | let _ = set_tcp_keepalive(sock.as_raw_fd(), PROXMOX_BACKUP_TCP_KEEPALIVE_TIME); | |
187 | let acceptor = Arc::clone(&acceptor); | |
188 | ||
189 | let ssl = match openssl::ssl::Ssl::new(acceptor.context()) { | |
190 | Ok(ssl) => ssl, | |
191 | Err(err) => { | |
192 | eprintln!("failed to create Ssl object from Acceptor context - {}", err); | |
193 | continue; | |
194 | }, | |
195 | }; | |
196 | let stream = match tokio_openssl::SslStream::new(ssl, sock) { | |
197 | Ok(stream) => stream, | |
198 | Err(err) => { | |
199 | eprintln!("failed to create SslStream using ssl and connection socket - {}", err); | |
200 | continue; | |
201 | }, | |
202 | }; | |
203 | ||
204 | let mut stream = Box::pin(stream); | |
205 | let sender = sender.clone(); | |
206 | ||
207 | if Arc::strong_count(&accept_counter) > MAX_PENDING_ACCEPTS { | |
208 | eprintln!("connection rejected - to many open connections"); | |
209 | continue; | |
210 | } | |
211 | ||
212 | let accept_counter = accept_counter.clone(); | |
213 | tokio::spawn(async move { | |
214 | let accept_future = tokio::time::timeout( | |
215 | Duration::new(10, 0), stream.as_mut().accept()); | |
216 | ||
217 | let result = accept_future.await; | |
218 | ||
219 | match result { | |
220 | Ok(Ok(())) => { | |
221 | if sender.send(Ok(stream)).await.is_err() && debug { | |
222 | eprintln!("detect closed connection channel"); | |
223 | } | |
224 | } | |
225 | Ok(Err(err)) => { | |
226 | if debug { | |
227 | eprintln!("https handshake failed - {}", err); | |
228 | } | |
229 | } | |
230 | Err(_) => { | |
231 | if debug { | |
232 | eprintln!("https handshake timeout"); | |
233 | } | |
234 | } | |
235 | } | |
236 | ||
237 | drop(accept_counter); // decrease reference count | |
238 | }); | |
239 | } | |
240 | } | |
241 | } | |
242 | }); | |
243 | ||
244 | receiver | |
245 | } | |
246 | ||
247 | fn start_stat_generator() { | |
248 | let abort_future = server::shutdown_future(); | |
249 | let future = Box::pin(run_stat_generator()); | |
250 | let task = futures::future::select(future, abort_future); | |
251 | tokio::spawn(task.map(|_| ())); | |
252 | } | |
253 | ||
254 | fn start_task_scheduler() { | |
255 | let abort_future = server::shutdown_future(); | |
256 | let future = Box::pin(run_task_scheduler()); | |
257 | let task = futures::future::select(future, abort_future); | |
258 | tokio::spawn(task.map(|_| ())); | |
259 | } | |
260 | ||
261 | use std::time::{SystemTime, Instant, Duration, UNIX_EPOCH}; | |
262 | ||
263 | fn next_minute() -> Result<Instant, Error> { | |
264 | let now = SystemTime::now(); | |
265 | let epoch_now = now.duration_since(UNIX_EPOCH)?; | |
266 | let epoch_next = Duration::from_secs((epoch_now.as_secs()/60 + 1)*60); | |
267 | Ok(Instant::now() + epoch_next - epoch_now) | |
268 | } | |
269 | ||
270 | async fn run_task_scheduler() { | |
271 | ||
272 | let mut count: usize = 0; | |
273 | ||
274 | loop { | |
275 | count += 1; | |
276 | ||
277 | let delay_target = match next_minute() { // try to run very minute | |
278 | Ok(d) => d, | |
279 | Err(err) => { | |
280 | eprintln!("task scheduler: compute next minute failed - {}", err); | |
281 | tokio::time::sleep_until(tokio::time::Instant::from_std(Instant::now() + Duration::from_secs(60))).await; | |
282 | continue; | |
283 | } | |
284 | }; | |
285 | ||
286 | if count > 2 { // wait 1..2 minutes before starting | |
287 | match schedule_tasks().catch_unwind().await { | |
288 | Err(panic) => { | |
289 | match panic.downcast::<&str>() { | |
290 | Ok(msg) => { | |
291 | eprintln!("task scheduler panic: {}", msg); | |
292 | } | |
293 | Err(_) => { | |
294 | eprintln!("task scheduler panic - unknown type"); | |
295 | } | |
296 | } | |
297 | } | |
298 | Ok(Err(err)) => { | |
299 | eprintln!("task scheduler failed - {:?}", err); | |
300 | } | |
301 | Ok(Ok(_)) => {} | |
302 | } | |
303 | } | |
304 | ||
305 | tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await; | |
306 | } | |
307 | } | |
308 | ||
309 | async fn schedule_tasks() -> Result<(), Error> { | |
310 | ||
311 | schedule_datastore_garbage_collection().await; | |
312 | schedule_datastore_prune().await; | |
313 | schedule_datastore_sync_jobs().await; | |
314 | schedule_datastore_verify_jobs().await; | |
315 | schedule_task_log_rotate().await; | |
316 | ||
317 | Ok(()) | |
318 | } | |
319 | ||
320 | async fn schedule_datastore_garbage_collection() { | |
321 | ||
322 | use proxmox_backup::config::{ | |
323 | datastore::{ | |
324 | self, | |
325 | DataStoreConfig, | |
326 | }, | |
327 | }; | |
328 | ||
329 | let config = match datastore::config() { | |
330 | Err(err) => { | |
331 | eprintln!("unable to read datastore config - {}", err); | |
332 | return; | |
333 | } | |
334 | Ok((config, _digest)) => config, | |
335 | }; | |
336 | ||
337 | for (store, (_, store_config)) in config.sections { | |
338 | let datastore = match DataStore::lookup_datastore(&store) { | |
339 | Ok(datastore) => datastore, | |
340 | Err(err) => { | |
341 | eprintln!("lookup_datastore failed - {}", err); | |
342 | continue; | |
343 | } | |
344 | }; | |
345 | ||
346 | let store_config: DataStoreConfig = match serde_json::from_value(store_config) { | |
347 | Ok(c) => c, | |
348 | Err(err) => { | |
349 | eprintln!("datastore config from_value failed - {}", err); | |
350 | continue; | |
351 | } | |
352 | }; | |
353 | ||
354 | let event_str = match store_config.gc_schedule { | |
355 | Some(event_str) => event_str, | |
356 | None => continue, | |
357 | }; | |
358 | ||
359 | let event = match parse_calendar_event(&event_str) { | |
360 | Ok(event) => event, | |
361 | Err(err) => { | |
362 | eprintln!("unable to parse schedule '{}' - {}", event_str, err); | |
363 | continue; | |
364 | } | |
365 | }; | |
366 | ||
367 | if datastore.garbage_collection_running() { continue; } | |
368 | ||
369 | let worker_type = "garbage_collection"; | |
370 | ||
371 | let last = match jobstate::last_run_time(worker_type, &store) { | |
372 | Ok(time) => time, | |
373 | Err(err) => { | |
374 | eprintln!("could not get last run time of {} {}: {}", worker_type, store, err); | |
375 | continue; | |
376 | } | |
377 | }; | |
378 | ||
379 | let next = match compute_next_event(&event, last, false) { | |
380 | Ok(Some(next)) => next, | |
381 | Ok(None) => continue, | |
382 | Err(err) => { | |
383 | eprintln!("compute_next_event for '{}' failed - {}", event_str, err); | |
384 | continue; | |
385 | } | |
386 | }; | |
387 | ||
388 | let now = proxmox::tools::time::epoch_i64(); | |
389 | ||
390 | if next > now { continue; } | |
391 | ||
392 | let job = match Job::new(worker_type, &store) { | |
393 | Ok(job) => job, | |
394 | Err(_) => continue, // could not get lock | |
395 | }; | |
396 | ||
397 | let auth_id = Authid::root_auth_id(); | |
398 | ||
399 | if let Err(err) = crate::server::do_garbage_collection_job(job, datastore, auth_id, Some(event_str), false) { | |
400 | eprintln!("unable to start garbage collection job on datastore {} - {}", store, err); | |
401 | } | |
402 | } | |
403 | } | |
404 | ||
405 | async fn schedule_datastore_prune() { | |
406 | ||
407 | use proxmox_backup::{ | |
408 | backup::{ | |
409 | PruneOptions, | |
410 | }, | |
411 | config::datastore::{ | |
412 | self, | |
413 | DataStoreConfig, | |
414 | }, | |
415 | }; | |
416 | ||
417 | let config = match datastore::config() { | |
418 | Err(err) => { | |
419 | eprintln!("unable to read datastore config - {}", err); | |
420 | return; | |
421 | } | |
422 | Ok((config, _digest)) => config, | |
423 | }; | |
424 | ||
425 | for (store, (_, store_config)) in config.sections { | |
426 | ||
427 | let store_config: DataStoreConfig = match serde_json::from_value(store_config) { | |
428 | Ok(c) => c, | |
429 | Err(err) => { | |
430 | eprintln!("datastore '{}' config from_value failed - {}", store, err); | |
431 | continue; | |
432 | } | |
433 | }; | |
434 | ||
435 | let event_str = match store_config.prune_schedule { | |
436 | Some(event_str) => event_str, | |
437 | None => continue, | |
438 | }; | |
439 | ||
440 | let prune_options = PruneOptions { | |
441 | keep_last: store_config.keep_last, | |
442 | keep_hourly: store_config.keep_hourly, | |
443 | keep_daily: store_config.keep_daily, | |
444 | keep_weekly: store_config.keep_weekly, | |
445 | keep_monthly: store_config.keep_monthly, | |
446 | keep_yearly: store_config.keep_yearly, | |
447 | }; | |
448 | ||
449 | if !prune_options.keeps_something() { // no prune settings - keep all | |
450 | continue; | |
451 | } | |
452 | ||
453 | let worker_type = "prune"; | |
454 | if check_schedule(worker_type, &event_str, &store) { | |
455 | let job = match Job::new(worker_type, &store) { | |
456 | Ok(job) => job, | |
457 | Err(_) => continue, // could not get lock | |
458 | }; | |
459 | ||
460 | let auth_id = Authid::root_auth_id().clone(); | |
461 | if let Err(err) = do_prune_job(job, prune_options, store.clone(), &auth_id, Some(event_str)) { | |
462 | eprintln!("unable to start datastore prune job {} - {}", &store, err); | |
463 | } | |
464 | }; | |
465 | } | |
466 | } | |
467 | ||
468 | async fn schedule_datastore_sync_jobs() { | |
469 | ||
470 | use proxmox_backup::config::sync::{ | |
471 | self, | |
472 | SyncJobConfig, | |
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 | for (job_id, (_, job_config)) in config.sections { | |
484 | let job_config: SyncJobConfig = match serde_json::from_value(job_config) { | |
485 | Ok(c) => c, | |
486 | Err(err) => { | |
487 | eprintln!("sync job config from_value failed - {}", err); | |
488 | continue; | |
489 | } | |
490 | }; | |
491 | ||
492 | let event_str = match job_config.schedule { | |
493 | Some(ref event_str) => event_str.clone(), | |
494 | None => continue, | |
495 | }; | |
496 | ||
497 | let worker_type = "syncjob"; | |
498 | if check_schedule(worker_type, &event_str, &job_id) { | |
499 | let job = match Job::new(worker_type, &job_id) { | |
500 | Ok(job) => job, | |
501 | Err(_) => continue, // could not get lock | |
502 | }; | |
503 | ||
504 | let auth_id = Authid::root_auth_id().clone(); | |
505 | if let Err(err) = do_sync_job(job, job_config, &auth_id, Some(event_str)) { | |
506 | eprintln!("unable to start datastore sync job {} - {}", &job_id, err); | |
507 | } | |
508 | }; | |
509 | } | |
510 | } | |
511 | ||
512 | async fn schedule_datastore_verify_jobs() { | |
513 | ||
514 | use proxmox_backup::config::verify::{ | |
515 | self, | |
516 | VerificationJobConfig, | |
517 | }; | |
518 | ||
519 | let config = match verify::config() { | |
520 | Err(err) => { | |
521 | eprintln!("unable to read verification job config - {}", err); | |
522 | return; | |
523 | } | |
524 | Ok((config, _digest)) => config, | |
525 | }; | |
526 | for (job_id, (_, job_config)) in config.sections { | |
527 | let job_config: VerificationJobConfig = match serde_json::from_value(job_config) { | |
528 | Ok(c) => c, | |
529 | Err(err) => { | |
530 | eprintln!("verification job config from_value failed - {}", err); | |
531 | continue; | |
532 | } | |
533 | }; | |
534 | let event_str = match job_config.schedule { | |
535 | Some(ref event_str) => event_str.clone(), | |
536 | None => continue, | |
537 | }; | |
538 | ||
539 | let worker_type = "verificationjob"; | |
540 | let auth_id = Authid::root_auth_id().clone(); | |
541 | if check_schedule(worker_type, &event_str, &job_id) { | |
542 | let job = match Job::new(&worker_type, &job_id) { | |
543 | Ok(job) => job, | |
544 | Err(_) => continue, // could not get lock | |
545 | }; | |
546 | if let Err(err) = do_verification_job(job, job_config, &auth_id, Some(event_str)) { | |
547 | eprintln!("unable to start datastore verification job {} - {}", &job_id, err); | |
548 | } | |
549 | }; | |
550 | } | |
551 | } | |
552 | ||
553 | async fn schedule_task_log_rotate() { | |
554 | ||
555 | let worker_type = "logrotate"; | |
556 | let job_id = "access-log_and_task-archive"; | |
557 | ||
558 | // schedule daily at 00:00 like normal logrotate | |
559 | let schedule = "00:00"; | |
560 | ||
561 | if !check_schedule(worker_type, schedule, job_id) { | |
562 | // if we never ran the rotation, schedule instantly | |
563 | match jobstate::JobState::load(worker_type, job_id) { | |
564 | Ok(state) => match state { | |
565 | jobstate::JobState::Created { .. } => {}, | |
566 | _ => return, | |
567 | }, | |
568 | _ => return, | |
569 | } | |
570 | } | |
571 | ||
572 | let mut job = match Job::new(worker_type, job_id) { | |
573 | Ok(job) => job, | |
574 | Err(_) => return, // could not get lock | |
575 | }; | |
576 | ||
577 | if let Err(err) = WorkerTask::new_thread( | |
578 | worker_type, | |
579 | None, | |
580 | Authid::root_auth_id().clone(), | |
581 | false, | |
582 | move |worker| { | |
583 | job.start(&worker.upid().to_string())?; | |
584 | worker.log("starting task log rotation".to_string()); | |
585 | ||
586 | let result = try_block!({ | |
587 | let max_size = 512 * 1024 - 1; // an entry has ~ 100b, so > 5000 entries/file | |
588 | let max_files = 20; // times twenty files gives > 100000 task entries | |
589 | let has_rotated = rotate_task_log_archive(max_size, true, Some(max_files))?; | |
590 | if has_rotated { | |
591 | worker.log("task log archive was rotated".to_string()); | |
592 | } else { | |
593 | worker.log("task log archive was not rotated".to_string()); | |
594 | } | |
595 | ||
596 | let max_size = 32 * 1024 * 1024 - 1; | |
597 | let max_files = 14; | |
598 | let mut logrotate = LogRotate::new(buildcfg::API_ACCESS_LOG_FN, true) | |
599 | .ok_or_else(|| format_err!("could not get API access log file names"))?; | |
600 | ||
601 | if logrotate.rotate(max_size, None, Some(max_files))? { | |
602 | println!("rotated access log, telling daemons to re-open log file"); | |
603 | proxmox_backup::tools::runtime::block_on(command_reopen_logfiles())?; | |
604 | worker.log("API access log was rotated".to_string()); | |
605 | } else { | |
606 | worker.log("API access log was not rotated".to_string()); | |
607 | } | |
608 | ||
609 | let mut logrotate = LogRotate::new(buildcfg::API_AUTH_LOG_FN, true) | |
610 | .ok_or_else(|| format_err!("could not get API auth log file names"))?; | |
611 | ||
612 | if logrotate.rotate(max_size, None, Some(max_files))? { | |
613 | worker.log("API authentication log was rotated".to_string()); | |
614 | } else { | |
615 | worker.log("API authentication log was not rotated".to_string()); | |
616 | } | |
617 | ||
618 | Ok(()) | |
619 | }); | |
620 | ||
621 | let status = worker.create_state(&result); | |
622 | ||
623 | if let Err(err) = job.finish(status) { | |
624 | eprintln!("could not finish job state for {}: {}", worker_type, err); | |
625 | } | |
626 | ||
627 | result | |
628 | }, | |
629 | ) { | |
630 | eprintln!("unable to start task log rotation: {}", err); | |
631 | } | |
632 | ||
633 | } | |
634 | ||
635 | async fn command_reopen_logfiles() -> Result<(), Error> { | |
636 | // only care about the most recent daemon instance for each, proxy & api, as other older ones | |
637 | // should not respond to new requests anyway, but only finish their current one and then exit. | |
638 | let sock = server::our_ctrl_sock(); | |
639 | let f1 = server::send_command(sock, serde_json::json!({ | |
640 | "command": "api-access-log-reopen", | |
641 | })); | |
642 | ||
643 | let pid = server::read_pid(buildcfg::PROXMOX_BACKUP_API_PID_FN)?; | |
644 | let sock = server::ctrl_sock_from_pid(pid); | |
645 | let f2 = server::send_command(sock, serde_json::json!({ | |
646 | "command": "api-access-log-reopen", | |
647 | })); | |
648 | ||
649 | match futures::join!(f1, f2) { | |
650 | (Err(e1), Err(e2)) => Err(format_err!("reopen commands failed, proxy: {}; api: {}", e1, e2)), | |
651 | (Err(e1), Ok(_)) => Err(format_err!("reopen commands failed, proxy: {}", e1)), | |
652 | (Ok(_), Err(e2)) => Err(format_err!("reopen commands failed, api: {}", e2)), | |
653 | _ => Ok(()), | |
654 | } | |
655 | } | |
656 | ||
657 | async fn run_stat_generator() { | |
658 | ||
659 | let mut count = 0; | |
660 | loop { | |
661 | count += 1; | |
662 | let save = if count >= 6 { count = 0; true } else { false }; | |
663 | ||
664 | let delay_target = Instant::now() + Duration::from_secs(10); | |
665 | ||
666 | generate_host_stats(save).await; | |
667 | ||
668 | tokio::time::sleep_until(tokio::time::Instant::from_std(delay_target)).await; | |
669 | ||
670 | } | |
671 | ||
672 | } | |
673 | ||
674 | fn rrd_update_gauge(name: &str, value: f64, save: bool) { | |
675 | use proxmox_backup::rrd; | |
676 | if let Err(err) = rrd::update_value(name, value, rrd::DST::Gauge, save) { | |
677 | eprintln!("rrd::update_value '{}' failed - {}", name, err); | |
678 | } | |
679 | } | |
680 | ||
681 | fn rrd_update_derive(name: &str, value: f64, save: bool) { | |
682 | use proxmox_backup::rrd; | |
683 | if let Err(err) = rrd::update_value(name, value, rrd::DST::Derive, save) { | |
684 | eprintln!("rrd::update_value '{}' failed - {}", name, err); | |
685 | } | |
686 | } | |
687 | ||
688 | async fn generate_host_stats(save: bool) { | |
689 | use proxmox::sys::linux::procfs::{ | |
690 | read_meminfo, read_proc_stat, read_proc_net_dev, read_loadavg}; | |
691 | use proxmox_backup::config::datastore; | |
692 | ||
693 | ||
694 | proxmox_backup::tools::runtime::block_in_place(move || { | |
695 | ||
696 | match read_proc_stat() { | |
697 | Ok(stat) => { | |
698 | rrd_update_gauge("host/cpu", stat.cpu, save); | |
699 | rrd_update_gauge("host/iowait", stat.iowait_percent, save); | |
700 | } | |
701 | Err(err) => { | |
702 | eprintln!("read_proc_stat failed - {}", err); | |
703 | } | |
704 | } | |
705 | ||
706 | match read_meminfo() { | |
707 | Ok(meminfo) => { | |
708 | rrd_update_gauge("host/memtotal", meminfo.memtotal as f64, save); | |
709 | rrd_update_gauge("host/memused", meminfo.memused as f64, save); | |
710 | rrd_update_gauge("host/swaptotal", meminfo.swaptotal as f64, save); | |
711 | rrd_update_gauge("host/swapused", meminfo.swapused as f64, save); | |
712 | } | |
713 | Err(err) => { | |
714 | eprintln!("read_meminfo failed - {}", err); | |
715 | } | |
716 | } | |
717 | ||
718 | match read_proc_net_dev() { | |
719 | Ok(netdev) => { | |
720 | use proxmox_backup::config::network::is_physical_nic; | |
721 | let mut netin = 0; | |
722 | let mut netout = 0; | |
723 | for item in netdev { | |
724 | if !is_physical_nic(&item.device) { continue; } | |
725 | netin += item.receive; | |
726 | netout += item.send; | |
727 | } | |
728 | rrd_update_derive("host/netin", netin as f64, save); | |
729 | rrd_update_derive("host/netout", netout as f64, save); | |
730 | } | |
731 | Err(err) => { | |
732 | eprintln!("read_prox_net_dev failed - {}", err); | |
733 | } | |
734 | } | |
735 | ||
736 | match read_loadavg() { | |
737 | Ok(loadavg) => { | |
738 | rrd_update_gauge("host/loadavg", loadavg.0 as f64, save); | |
739 | } | |
740 | Err(err) => { | |
741 | eprintln!("read_loadavg failed - {}", err); | |
742 | } | |
743 | } | |
744 | ||
745 | let disk_manager = DiskManage::new(); | |
746 | ||
747 | gather_disk_stats(disk_manager.clone(), Path::new("/"), "host", save); | |
748 | ||
749 | match datastore::config() { | |
750 | Ok((config, _)) => { | |
751 | let datastore_list: Vec<datastore::DataStoreConfig> = | |
752 | config.convert_to_typed_array("datastore").unwrap_or_default(); | |
753 | ||
754 | for config in datastore_list { | |
755 | ||
756 | let rrd_prefix = format!("datastore/{}", config.name); | |
757 | let path = std::path::Path::new(&config.path); | |
758 | gather_disk_stats(disk_manager.clone(), path, &rrd_prefix, save); | |
759 | } | |
760 | } | |
761 | Err(err) => { | |
762 | eprintln!("read datastore config failed - {}", err); | |
763 | } | |
764 | } | |
765 | ||
766 | }); | |
767 | } | |
768 | ||
769 | fn check_schedule(worker_type: &str, event_str: &str, id: &str) -> bool { | |
770 | let event = match parse_calendar_event(event_str) { | |
771 | Ok(event) => event, | |
772 | Err(err) => { | |
773 | eprintln!("unable to parse schedule '{}' - {}", event_str, err); | |
774 | return false; | |
775 | } | |
776 | }; | |
777 | ||
778 | let last = match jobstate::last_run_time(worker_type, &id) { | |
779 | Ok(time) => time, | |
780 | Err(err) => { | |
781 | eprintln!("could not get last run time of {} {}: {}", worker_type, id, err); | |
782 | return false; | |
783 | } | |
784 | }; | |
785 | ||
786 | let next = match compute_next_event(&event, last, false) { | |
787 | Ok(Some(next)) => next, | |
788 | Ok(None) => return false, | |
789 | Err(err) => { | |
790 | eprintln!("compute_next_event for '{}' failed - {}", event_str, err); | |
791 | return false; | |
792 | } | |
793 | }; | |
794 | ||
795 | let now = proxmox::tools::time::epoch_i64(); | |
796 | next <= now | |
797 | } | |
798 | ||
799 | fn gather_disk_stats(disk_manager: Arc<DiskManage>, path: &Path, rrd_prefix: &str, save: bool) { | |
800 | ||
801 | match proxmox_backup::tools::disks::disk_usage(path) { | |
802 | Ok(status) => { | |
803 | let rrd_key = format!("{}/total", rrd_prefix); | |
804 | rrd_update_gauge(&rrd_key, status.total as f64, save); | |
805 | let rrd_key = format!("{}/used", rrd_prefix); | |
806 | rrd_update_gauge(&rrd_key, status.used as f64, save); | |
807 | } | |
808 | Err(err) => { | |
809 | eprintln!("read disk_usage on {:?} failed - {}", path, err); | |
810 | } | |
811 | } | |
812 | ||
813 | match disk_manager.find_mounted_device(path) { | |
814 | Ok(None) => {}, | |
815 | Ok(Some((fs_type, device, source))) => { | |
816 | let mut device_stat = None; | |
817 | match fs_type.as_str() { | |
818 | "zfs" => { | |
819 | if let Some(pool) = source { | |
820 | match zfs_pool_stats(&pool) { | |
821 | Ok(stat) => device_stat = stat, | |
822 | Err(err) => eprintln!("zfs_pool_stats({:?}) failed - {}", pool, err), | |
823 | } | |
824 | } | |
825 | } | |
826 | _ => { | |
827 | if let Ok(disk) = disk_manager.clone().disk_by_dev_num(device.into_dev_t()) { | |
828 | match disk.read_stat() { | |
829 | Ok(stat) => device_stat = stat, | |
830 | Err(err) => eprintln!("disk.read_stat {:?} failed - {}", path, err), | |
831 | } | |
832 | } | |
833 | } | |
834 | } | |
835 | if let Some(stat) = device_stat { | |
836 | let rrd_key = format!("{}/read_ios", rrd_prefix); | |
837 | rrd_update_derive(&rrd_key, stat.read_ios as f64, save); | |
838 | let rrd_key = format!("{}/read_bytes", rrd_prefix); | |
839 | rrd_update_derive(&rrd_key, (stat.read_sectors*512) as f64, save); | |
840 | ||
841 | let rrd_key = format!("{}/write_ios", rrd_prefix); | |
842 | rrd_update_derive(&rrd_key, stat.write_ios as f64, save); | |
843 | let rrd_key = format!("{}/write_bytes", rrd_prefix); | |
844 | rrd_update_derive(&rrd_key, (stat.write_sectors*512) as f64, save); | |
845 | ||
846 | let rrd_key = format!("{}/io_ticks", rrd_prefix); | |
847 | rrd_update_derive(&rrd_key, (stat.io_ticks as f64)/1000.0, save); | |
848 | } | |
849 | } | |
850 | Err(err) => { | |
851 | eprintln!("find_mounted_device failed - {}", err); | |
852 | } | |
853 | } | |
854 | } |