1
// Copyright (C) Moondance Labs Ltd.
2
// This file is part of Tanssi.
3

            
4
// Tanssi is free software: you can redistribute it and/or modify
5
// it under the terms of the GNU General Public License as published by
6
// the Free Software Foundation, either version 3 of the License, or
7
// (at your option) any later version.
8

            
9
// Tanssi is distributed in the hope that it will be useful,
10
// but WITHOUT ANY WARRANTY; without even the implied warranty of
11
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12
// GNU General Public License for more details.
13

            
14
// You should have received a copy of the GNU General Public License
15
// along with Tanssi.  If not, see <http://www.gnu.org/licenses/>.
16

            
17
//! Container Chain Spawner
18
//!
19
//! Controls the starting and stopping of container chains.
20
//!
21
//! For more information about when the database is deleted, check the
22
//! [Keep db flowchart](https://raw.githubusercontent.com/moondance-labs/tanssi/master/docs/keep_db_flowchart.png)
23

            
24
use {
25
    crate::{
26
        cli::ContainerChainCli,
27
        monitor::{SpawnedContainer, SpawnedContainersMonitor},
28
        service::{start_node_impl_container, ContainerChainClient, ParachainClient},
29
    },
30
    cumulus_primitives_core::ParaId,
31
    cumulus_relay_chain_interface::RelayChainInterface,
32
    dancebox_runtime::{opaque::Block as OpaqueBlock, Block},
33
    dc_orchestrator_chain_interface::{OrchestratorChainInterface, PHash},
34
    fs2::FileExt,
35
    futures::FutureExt,
36
    node_common::command::generate_genesis_block,
37
    pallet_author_noting_runtime_api::AuthorNotingApi,
38
    polkadot_primitives::CollatorPair,
39
    sc_cli::{Database, SyncMode},
40
    sc_network::config::MultiaddrWithPeerId,
41
    sc_service::SpawnTaskHandle,
42
    sc_transaction_pool::FullPool,
43
    sp_api::ProvideRuntimeApi,
44
    sp_core::H256,
45
    sp_keystore::KeystorePtr,
46
    sp_runtime::traits::Block as BlockT,
47
    std::{
48
        collections::{HashMap, HashSet},
49
        path::{Path, PathBuf},
50
        sync::{Arc, Mutex},
51
        time::Instant,
52
    },
53
    tokio::{
54
        sync::{mpsc, oneshot},
55
        time::{sleep, Duration},
56
    },
57
    tokio_util::sync::CancellationToken,
58
};
59

            
60
/// Timeout to wait for the database to close before starting it again, used in `wait_for_paritydb_lock`.
61
/// This is the max timeout, if the db is closed in 1 second then that function will only wait 1 second.
62
const MAX_DB_RESTART_TIMEOUT: Duration = Duration::from_secs(60);
63

            
64
/// Block diff threshold above which we decide it will be faster to delete the database and
65
/// use warp sync, rather than using full sync to download a large number of blocks.
66
/// This is only needed because warp sync does not support syncing from a state that is not
67
/// genesis, it falls back to full sync in that case.
68
/// 30_000 blocks = 50 hours at 6s/block.
69
/// Assuming a syncing speed of 100 blocks per second, this will take 5 minutes to sync.
70
const MAX_BLOCK_DIFF_FOR_FULL_SYNC: u32 = 30_000;
71

            
72
/// Task that handles spawning a stopping container chains based on assignment.
73
/// The main loop is [rx_loop](ContainerChainSpawner::rx_loop).
74
pub struct ContainerChainSpawner {
75
    /// Start container chain params
76
    pub params: ContainerChainSpawnParams,
77

            
78
    /// State
79
    pub state: Arc<Mutex<ContainerChainSpawnerState>>,
80

            
81
    /// Async callback that enables collation on the orchestrator chain
82
    pub collate_on_tanssi:
83
        Arc<dyn Fn() -> (CancellationToken, futures::channel::oneshot::Receiver<()>) + Send + Sync>,
84
    /// Stores the cancellation token used to stop the orchestrator chain collator process.
85
    /// When this is None, the orchestrator collator is not running.
86
    pub collation_cancellation_constructs:
87
        Option<(CancellationToken, futures::channel::oneshot::Receiver<()>)>,
88
}
89

            
90
/// Struct with all the params needed to start a container chain node given the CLI arguments,
91
/// and creating the ChainSpec from on-chain data from the orchestrator chain.
92
/// These params must be the same for all container chains, params that change such as the
93
/// `container_chain_para_id` should be passed as separate arguments to the [try_spawn] function.
94
///
95
/// This struct MUST NOT contain types (outside of `Option<CollationParams>`) obtained through
96
/// running an embeded orchestrator node, as this will prevent spawning a container chain in a node
97
/// connected to an orchestrator node through WebSocket.
98
#[derive(Clone)]
99
pub struct ContainerChainSpawnParams {
100
    pub orchestrator_chain_interface: Arc<dyn OrchestratorChainInterface>,
101
    pub container_chain_cli: ContainerChainCli,
102
    pub tokio_handle: tokio::runtime::Handle,
103
    pub chain_type: sc_chain_spec::ChainType,
104
    pub relay_chain: String,
105
    pub relay_chain_interface: Arc<dyn RelayChainInterface>,
106
    pub sync_keystore: KeystorePtr,
107
    pub orchestrator_para_id: ParaId,
108
    pub spawn_handle: SpawnTaskHandle,
109
    pub collation_params: Option<CollationParams>,
110
    pub data_preserver: bool,
111
}
112

            
113
/// Params specific to collation. This struct can contain types obtained through running an
114
/// embeded orchestrator node.
115
#[derive(Clone)]
116
pub struct CollationParams {
117
    pub collator_key: CollatorPair,
118
    pub orchestrator_tx_pool: Option<Arc<FullPool<OpaqueBlock, ParachainClient>>>,
119
    pub orchestrator_client: Option<Arc<ParachainClient>>,
120
    pub orchestrator_para_id: ParaId,
121
    /// If this is `false`, then `orchestrator_tx_pool` and `orchestrator_client` must be `Some`.
122
    pub solochain: bool,
123
}
124

            
125
/// Mutable state for container chain spawner. Keeps track of running chains.
126
#[derive(Default)]
127
pub struct ContainerChainSpawnerState {
128
    spawned_container_chains: HashMap<ParaId, ContainerChainState>,
129
    assigned_para_id: Option<ParaId>,
130
    next_assigned_para_id: Option<ParaId>,
131
    failed_para_ids: HashSet<ParaId>,
132
    // For debugging and detecting errors
133
    pub spawned_containers_monitor: SpawnedContainersMonitor,
134
}
135

            
136
pub struct ContainerChainState {
137
    /// Handle that can be used to stop the container chain
138
    stop_handle: StopContainerChain,
139
    /// Database path
140
    db_path: PathBuf,
141
}
142

            
143
/// Stops a container chain when signal is sent. The bool means `keep_db`, whether to keep the
144
/// container chain database (true) or remove it (false).
145
pub struct StopContainerChain {
146
    signal: oneshot::Sender<bool>,
147
    id: usize,
148
}
149

            
150
/// Messages used to control the `ContainerChainSpawner`. This is needed because one of the fields
151
/// of `ContainerChainSpawner` is not `Sync`, so we cannot simply pass an
152
/// `Arc<ContainerChainSpawner>` to other threads.
153
#[derive(Debug)]
154
pub enum CcSpawnMsg {
155
    /// Update container chain assignment
156
    UpdateAssignment {
157
        current: Option<ParaId>,
158
        next: Option<ParaId>,
159
    },
160
}
161

            
162
// Separate function to allow using `?` to return a result, and also to avoid using `self` in an
163
// async function. Mutable state should be written by locking `state`.
164
// TODO: `state` should be an async mutex
165
async fn try_spawn(
166
    try_spawn_params: ContainerChainSpawnParams,
167
    state: Arc<Mutex<ContainerChainSpawnerState>>,
168
    container_chain_para_id: ParaId,
169
    start_collation: bool,
170
) -> sc_service::error::Result<()> {
171
    let ContainerChainSpawnParams {
172
        orchestrator_chain_interface,
173
        mut container_chain_cli,
174
        tokio_handle,
175
        chain_type,
176
        relay_chain,
177
        relay_chain_interface,
178
        sync_keystore,
179
        spawn_handle,
180
        mut collation_params,
181
        data_preserver,
182
        ..
183
    } = try_spawn_params;
184
    // Preload genesis data from orchestrator chain storage.
185

            
186
    // TODO: the orchestrator chain node may not be fully synced yet,
187
    // in that case we will be reading an old state.
188
    let orchestrator_block_hash = orchestrator_chain_interface
189
        .finalized_block_hash()
190
        .await
191
        .map_err(|e| format!("Failed to get latest block hash: {e}"))?;
192

            
193
    log::info!(
194
        "Detected assignment for container chain {}",
195
        container_chain_para_id
196
    );
197

            
198
    let genesis_data = orchestrator_chain_interface
199
        .genesis_data(orchestrator_block_hash, container_chain_para_id)
200
        .await
201
        .map_err(|e| format!("Failed to call genesis_data runtime api: {}", e))?
202
        .ok_or_else(|| {
203
            format!(
204
                "No genesis data registered for container chain id {}",
205
                container_chain_para_id
206
            )
207
        })?;
208

            
209
    let boot_nodes_raw = orchestrator_chain_interface
210
        .boot_nodes(orchestrator_block_hash, container_chain_para_id)
211
        .await
212
        .map_err(|e| format!("Failed to call boot_nodes runtime api: {}", e))?;
213

            
214
    if boot_nodes_raw.is_empty() {
215
        log::warn!(
216
            "No boot nodes registered on-chain for container chain {}",
217
            container_chain_para_id
218
        );
219
    }
220
    let boot_nodes = parse_boot_nodes_ignore_invalid(boot_nodes_raw, container_chain_para_id);
221
    if boot_nodes.is_empty() {
222
        log::warn!(
223
            "No valid boot nodes for container chain {}",
224
            container_chain_para_id
225
        );
226
    }
227

            
228
    container_chain_cli
229
        .preload_chain_spec_from_genesis_data(
230
            container_chain_para_id.into(),
231
            genesis_data,
232
            chain_type.clone(),
233
            relay_chain.clone(),
234
            boot_nodes,
235
        )
236
        .map_err(|e| {
237
            format!(
238
                "failed to create container chain chain spec from on chain genesis data: {}",
239
                e
240
            )
241
        })?;
242

            
243
    log::info!(
244
        "Loaded chain spec for container chain {}",
245
        container_chain_para_id
246
    );
247

            
248
    if !data_preserver && !start_collation {
249
        log::info!("This is a syncing container chain, using random ports");
250

            
251
        collation_params = None;
252

            
253
        // Use random ports to avoid conflicts with the other running container chain
254
        let random_ports = [23456, 23457, 23458];
255

            
256
        container_chain_cli
257
            .base
258
            .base
259
            .prometheus_params
260
            .prometheus_port = Some(random_ports[0]);
261
        container_chain_cli.base.base.network_params.port = Some(random_ports[1]);
262
        container_chain_cli.base.base.rpc_port = Some(random_ports[2]);
263
    }
264

            
265
    let validator = collation_params.is_some();
266

            
267
    // Update CLI params
268
    container_chain_cli.base.para_id = Some(container_chain_para_id.into());
269
    container_chain_cli
270
        .base
271
        .base
272
        .import_params
273
        .database_params
274
        .database = Some(Database::ParityDb);
275

            
276
    let keep_db = container_chain_cli.base.keep_db;
277

            
278
    // Get a closure that checks if db_path exists.Need this to know when to use full sync instead of warp sync.
279
    let check_db_exists = {
280
        // Get db_path from config
281
        let mut container_chain_cli_config = sc_cli::SubstrateCli::create_configuration(
282
            &container_chain_cli,
283
            &container_chain_cli,
284
            tokio_handle.clone(),
285
        )
286
        .map_err(|err| format!("Container chain argument error: {}", err))?;
287

            
288
        // Change database path to make it depend on container chain para id
289
        // So instead of the usual "db/full" we have "db/full-container-2000"
290
        let mut db_path = container_chain_cli_config
291
            .database
292
            .path()
293
            .ok_or_else(|| "Failed to get database path".to_string())?
294
            .to_owned();
295
        db_path.set_file_name(format!("full-container-{}", container_chain_para_id));
296
        container_chain_cli_config.database.set_path(&db_path);
297

            
298
        // Return a closure because we may need to check if the db exists multiple times
299
        move || db_path.exists()
300
    };
301

            
302
    // Start container chain node. After starting, check if the database is good or needs to
303
    // be removed. If the db needs to be removed, this function will handle the node restart, and
304
    // return the components of a running container chain node.
305
    // This should be a separate function, but it has so many arguments that I prefer to have it as a closure for now
306
    let start_node_impl_container_with_restart = || async move {
307
        // Loop will run at most 2 times: 1 time if the db is good and 2 times if the db needs to be removed
308
        for _ in 0..2 {
309
            let db_existed_before = check_db_exists();
310
            container_chain_cli.base.base.network_params.sync = SyncMode::Warp;
311
            log::info!(
312
                "Container chain sync mode: {:?}",
313
                container_chain_cli.base.base.network_params.sync
314
            );
315

            
316
            let mut container_chain_cli_config = sc_cli::SubstrateCli::create_configuration(
317
                &container_chain_cli,
318
                &container_chain_cli,
319
                tokio_handle.clone(),
320
            )
321
            .map_err(|err| format!("Container chain argument error: {}", err))?;
322

            
323
            // Change database path to make it depend on container chain para id
324
            // So instead of the usual "db/full" we have "db/full-container-2000"
325
            let mut db_path = container_chain_cli_config
326
                .database
327
                .path()
328
                .ok_or_else(|| "Failed to get database path".to_string())?
329
                .to_owned();
330
            db_path.set_file_name(format!("full-container-{}", container_chain_para_id));
331
            container_chain_cli_config.database.set_path(&db_path);
332

            
333
            let (container_chain_task_manager, container_chain_client, container_chain_db) =
334
                start_node_impl_container(
335
                    container_chain_cli_config,
336
                    relay_chain_interface.clone(),
337
                    orchestrator_chain_interface.clone(),
338
                    sync_keystore.clone(),
339
                    container_chain_para_id,
340
                    collation_params.clone(),
341
                )
342
                .await?;
343

            
344
            // Keep all node parts in one variable to make them easier to drop
345
            let node_parts = (
346
                container_chain_task_manager,
347
                container_chain_client,
348
                container_chain_db,
349
                db_path,
350
            );
351

            
352
            if db_existed_before {
353
                // If the database already existed before, check if it can be used or it needs to be removed.
354
                // To remove the database, we restart the node, wait for the db to close to avoid a
355
                // "shutdown error" log, and then remove it.
356
                if let Some(db_removal_reason) = db_needs_removal(
357
                    &node_parts.1,
358
                    &orchestrator_chain_interface,
359
                    orchestrator_block_hash,
360
                    container_chain_para_id,
361
                    &container_chain_cli,
362
                    container_chain_cli.base.keep_db,
363
                )
364
                .await?
365
                {
366
                    let db_path = node_parts.3.clone();
367
                    // Important, drop `node_parts` before trying to `wait_for_paritydb_lock`
368
                    drop(node_parts);
369
                    // Wait here to for the database created in the previous loop iteration to close.
370
                    // Dropping is not enough because there is some background process that keeps the database open,
371
                    // so we check the paritydb lock file directly.
372
                    log::info!(
373
                        "Restarting container chain {} after db deletion. Reason: {:?}",
374
                        container_chain_para_id,
375
                        db_removal_reason,
376
                    );
377
                    wait_for_paritydb_lock(&db_path, MAX_DB_RESTART_TIMEOUT)
378
                        .await
379
                        .map_err(|e| {
380
                            log::warn!(
381
                                "Error waiting for chain {} to release db lock: {:?}",
382
                                container_chain_para_id,
383
                                e
384
                            );
385

            
386
                            e
387
                        })?;
388
                    delete_container_chain_db(&db_path);
389

            
390
                    // Recursion, will only happen once because `db_existed_before` will be false after
391
                    // removing the db. Apparently closures cannot be recursive so fake recursion by
392
                    // using a loop + continue
393
                    continue;
394
                }
395
            }
396

            
397
            // If using full sync, print a warning if the local db is at block 0 and the chain has thousands of blocks
398
            if container_chain_cli.base.base.network_params.sync == SyncMode::Full {
399
                let last_container_block_temp = node_parts.1.chain_info().best_number;
400
                let cc_block_num = get_latest_container_block_number_from_orchestrator(
401
                    &orchestrator_chain_interface,
402
                    orchestrator_block_hash,
403
                    container_chain_para_id,
404
                )
405
                .await
406
                .unwrap_or(0);
407
                if last_container_block_temp == 0 && cc_block_num > MAX_BLOCK_DIFF_FOR_FULL_SYNC {
408
                    let db_folder = format!("full-container-{}", container_chain_para_id);
409
                    log::error!("\
410
                        Existing database for container chain {} is at block 0, assuming that warp sync failed.\n\
411
                        The node will now use full sync, which has to download {} blocks.\n\
412
                        If running as collator, it may not finish syncing on time and miss block rewards.\n\
413
                        To force using warp sync, stop tanssi-node and manually remove the db folder: {:?}\n\
414
                        ", container_chain_para_id, cc_block_num, db_folder)
415
                }
416
            }
417

            
418
            return sc_service::error::Result::Ok(node_parts);
419
        }
420

            
421
        unreachable!("Above loop can run at most 2 times, and in the second iteration it is guaranteed to return")
422
    };
423

            
424
    let (mut container_chain_task_manager, container_chain_client, container_chain_db, db_path) =
425
        start_node_impl_container_with_restart().await?;
426

            
427
    // Signal that allows to gracefully stop a container chain
428
    let (signal, on_exit) = oneshot::channel::<bool>();
429

            
430
    let monitor_id;
431
    {
432
        let mut state = state.lock().expect("poison error");
433

            
434
        monitor_id = state.spawned_containers_monitor.push(SpawnedContainer {
435
            id: 0,
436
            para_id: container_chain_para_id,
437
            start_time: Instant::now(),
438
            stop_signal_time: None,
439
            stop_task_manager_time: None,
440
            stop_refcount_time: Default::default(),
441
            backend: Arc::downgrade(&container_chain_db),
442
            client: Arc::downgrade(&container_chain_client),
443
        });
444

            
445
        if state
446
            .spawned_container_chains
447
            .contains_key(&container_chain_para_id)
448
        {
449
            return Err(format!("Tried to spawn a container chain when another container chain with the same para id was already running: {:?}", container_chain_para_id).into());
450
        }
451
        state.spawned_container_chains.insert(
452
            container_chain_para_id,
453
            ContainerChainState {
454
                stop_handle: StopContainerChain {
455
                    signal,
456
                    id: monitor_id,
457
                },
458
                db_path: db_path.clone(),
459
            },
460
        );
461
    }
462

            
463
    // Add the container chain task manager as a child task to the parent task manager.
464
    // We want to stop the node if this task manager stops, but we also want to allow a
465
    // graceful shutdown using the `on_exit` future.
466
    let name = "container-chain-task-manager";
467
    spawn_handle.spawn(name, None, async move {
468
        let mut container_chain_task_manager_future =
469
            container_chain_task_manager.future().fuse();
470
        let mut on_exit_future = on_exit.fuse();
471

            
472
        futures::select! {
473
            res1 = container_chain_task_manager_future => {
474
                // An essential task failed or the task manager was stopped unexpectedly
475
                // using `.terminate()`. This should stop the container chain but not the node.
476
                if res1.is_err() {
477
                    log::error!("Essential task failed in container chain {} task manager. Shutting down container chain service", container_chain_para_id);
478
                } else {
479
                    log::error!("Unexpected shutdown in container chain {} task manager. Shutting down container chain service", container_chain_para_id);
480
                }
481
                // Mark this container chain as "failed to stop" to avoid warning in `self.stop()`
482
                let mut state = state.lock().expect("poison error");
483
                state.failed_para_ids.insert(container_chain_para_id);
484
                // Never delete db in this case because it is not a graceful shutdown
485
            }
486
            stop_unassigned = on_exit_future => {
487
                // Graceful shutdown.
488
                // `stop_unassigned` will be `Ok(keep_db)` if `.stop()` has been called, which means that the
489
                // container chain has been unassigned, and will be `Err` if the handle has been dropped,
490
                // which means that the node is stopping.
491
                // Delete existing database if running as collator
492
                if validator && stop_unassigned == Ok(false) && !keep_db {
493
                    // If this breaks after a code change, make sure that all the variables that
494
                    // may keep the chain alive are dropped before the call to `wait_for_paritydb_lock`.
495
                    drop(container_chain_task_manager_future);
496
                    drop(container_chain_task_manager);
497
                    let db_closed = wait_for_paritydb_lock(&db_path, MAX_DB_RESTART_TIMEOUT)
498
                        .await
499
                        .map_err(|e| {
500
                            log::warn!(
501
                                "Error waiting for chain {} to release db lock: {:?}",
502
                                container_chain_para_id,
503
                                e
504
                            );
505
                        }).is_ok();
506
                    // If db has not closed in 60 seconds we do not delete it.
507
                    if db_closed {
508
                        delete_container_chain_db(&db_path);
509
                    }
510
                }
511
            }
512
        }
513

            
514
        let mut state = state.lock().expect("poison error");
515
        state
516
            .spawned_containers_monitor
517
            .set_stop_task_manager_time(monitor_id, Instant::now());
518
    });
519

            
520
    Ok(())
521
}
522

            
523
/// Interface for spawning and stopping container chain embeded nodes.
524
pub trait Spawner {
525
    /// Access to the Orchestrator Chain Interface
526
    fn orchestrator_chain_interface(&self) -> Arc<dyn OrchestratorChainInterface>;
527

            
528
    /// Try to start a new container chain. In case of an error, this does not stop the node, and
529
    /// the container chain will be attempted to spawn again when the collator is reassigned to it.
530
    ///
531
    /// It is possible that we try to spawn-stop-spawn the same chain, and the second spawn fails
532
    /// because the chain has not stopped yet, because `stop` does not wait for the chain to stop,
533
    /// so before calling `spawn` make sure to call `wait_for_paritydb_lock` before, like we do in
534
    /// `handle_update_assignment`.
535
    fn spawn(
536
        &self,
537
        container_chain_para_id: ParaId,
538
        start_collation: bool,
539
    ) -> impl std::future::Future<Output = ()> + Send;
540

            
541
    /// Stop a container chain. Prints a warning if the container chain was not running.
542
    /// Returns the database path for the container chain, can be used with `wait_for_paritydb_lock`
543
    /// to ensure that the container chain has fully stopped. The database path can be `None` if the
544
    /// chain was not running.
545
    fn stop(&self, container_chain_para_id: ParaId, keep_db: bool) -> Option<PathBuf>;
546
}
547

            
548
impl Spawner for ContainerChainSpawner {
549
    /// Access to the Orchestrator Chain Interface
550
    fn orchestrator_chain_interface(&self) -> Arc<dyn OrchestratorChainInterface> {
551
        self.params.orchestrator_chain_interface.clone()
552
    }
553

            
554
    /// Try to start a new container chain. In case of an error, this does not stop the node, and
555
    /// the container chain will be attempted to spawn again when the collator is reassigned to it.
556
    ///
557
    /// It is possible that we try to spawn-stop-spawn the same chain, and the second spawn fails
558
    /// because the chain has not stopped yet, because `stop` does not wait for the chain to stop,
559
    /// so before calling `spawn` make sure to call `wait_for_paritydb_lock` before, like we do in
560
    /// `handle_update_assignment`.
561
    async fn spawn(&self, container_chain_para_id: ParaId, start_collation: bool) {
562
        let try_spawn_params = self.params.clone();
563
        let state = self.state.clone();
564
        let state2 = state.clone();
565

            
566
        match try_spawn(
567
            try_spawn_params,
568
            state,
569
            container_chain_para_id,
570
            start_collation,
571
        )
572
        .await
573
        {
574
            Ok(()) => {}
575
            Err(e) => {
576
                log::error!(
577
                    "Failed to start container chain {}: {}",
578
                    container_chain_para_id,
579
                    e
580
                );
581
                // Mark this container chain as "failed to start"
582
                let mut state = state2.lock().expect("poison error");
583
                state.failed_para_ids.insert(container_chain_para_id);
584
            }
585
        }
586
    }
587

            
588
    /// Stop a container chain. Prints a warning if the container chain was not running.
589
    /// Returns the database path for the container chain, can be used with `wait_for_paritydb_lock`
590
    /// to ensure that the container chain has fully stopped. The database path can be `None` if the
591
    /// chain was not running.
592
    fn stop(&self, container_chain_para_id: ParaId, keep_db: bool) -> Option<PathBuf> {
593
        let mut state = self.state.lock().expect("poison error");
594
        let stop_handle = state
595
            .spawned_container_chains
596
            .remove(&container_chain_para_id);
597

            
598
        match stop_handle {
599
            Some(stop_handle) => {
600
                log::info!("Stopping container chain {}", container_chain_para_id);
601

            
602
                let id = stop_handle.stop_handle.id;
603
                state
604
                    .spawned_containers_monitor
605
                    .set_stop_signal_time(id, Instant::now());
606

            
607
                // Send signal to perform graceful shutdown, which will delete the db if needed
608
                let _ = stop_handle.stop_handle.signal.send(keep_db);
609

            
610
                Some(stop_handle.db_path)
611
            }
612
            None => {
613
                // Do not print the warning message if this is a container chain that has failed to
614
                // start, because in that case it will not be running
615
                if !state.failed_para_ids.remove(&container_chain_para_id) {
616
                    log::warn!(
617
                        "Tried to stop a container chain that is not running: {}",
618
                        container_chain_para_id
619
                    );
620
                }
621

            
622
                None
623
            }
624
        }
625
    }
626
}
627

            
628
impl ContainerChainSpawner {
629
    /// Receive and process `CcSpawnMsg`s indefinitely
630
    pub async fn rx_loop(
631
        mut self,
632
        mut rx: mpsc::UnboundedReceiver<CcSpawnMsg>,
633
        validator: bool,
634
        solochain: bool,
635
    ) {
636
        // The node always starts as an orchestrator chain collator.
637
        // This is because the assignment is detected after importing a new block, so if all
638
        // collators stop at the same time, when they start again nobody will produce the new block.
639
        // So all nodes start as orchestrator chain collators, until the first block is imported,
640
        // then the real assignment is used.
641
        // Except in solochain mode, then the initial assignment is None.
642
        if validator && !solochain {
643
            self.handle_update_assignment(Some(self.params.orchestrator_para_id), None)
644
                .await;
645
        }
646

            
647
        while let Some(msg) = rx.recv().await {
648
            match msg {
649
                CcSpawnMsg::UpdateAssignment { current, next } => {
650
                    self.handle_update_assignment(current, next).await;
651
                }
652
            }
653
        }
654

            
655
        // The while loop can end if all the senders get dropped, but since this is an
656
        // essential task we don't want it to stop. So await a future that never completes.
657
        // This should only happen when starting a full node.
658
        if !validator {
659
            let () = std::future::pending().await;
660
        }
661
    }
662

            
663
    /// Handle `CcSpawnMsg::UpdateAssignment`
664
    async fn handle_update_assignment(&mut self, current: Option<ParaId>, next: Option<ParaId>) {
665
        let HandleUpdateAssignmentResult {
666
            chains_to_stop,
667
            chains_to_start,
668
            need_to_restart: _,
669
        } = handle_update_assignment_state_change(
670
            &mut self.state.lock().expect("poison error"),
671
            self.params.orchestrator_para_id,
672
            current,
673
            next,
674
        );
675

            
676
        if current != Some(self.params.orchestrator_para_id) {
677
            // If not assigned to orchestrator chain anymore, we need to stop the collator process
678
            let maybe_exit_notification_receiver = self
679
                .collation_cancellation_constructs
680
                .take()
681
                .map(|(cancellation_token, exit_notification_receiver)| {
682
                    cancellation_token.cancel();
683
                    exit_notification_receiver
684
                });
685

            
686
            if let Some(exit_notification_receiver) = maybe_exit_notification_receiver {
687
                let _ = exit_notification_receiver.await;
688
            }
689
        } else if self.collation_cancellation_constructs.is_none() {
690
            // If assigned to orchestrator chain but the collator process is not running, start it
691
            self.collation_cancellation_constructs = Some((self.collate_on_tanssi)());
692
        }
693

            
694
        // Stop all container chains that are no longer needed
695
        let mut db_paths_restart = vec![];
696
        for para_id in chains_to_stop {
697
            // Keep db if we are currently assigned to this chain
698
            let keep_db = Some(para_id) == current;
699
            let maybe_db_path = self.stop(para_id, keep_db);
700
            // If we are restarting this chain, save its db_path to check when it actually stopped
701
            if let Some(db_path) = maybe_db_path {
702
                if chains_to_start.contains(&para_id) {
703
                    db_paths_restart.push((para_id, db_path));
704
                }
705
            }
706
        }
707

            
708
        if !db_paths_restart.is_empty() {
709
            // Ensure the chains we stopped actually stopped by checking if their database is unlocked.
710
            // Using `join_all` because in one edge case we may be restarting 2 chains,
711
            // but almost always this will be only one future.
712
            let futs = db_paths_restart
713
                .into_iter()
714
                .map(|(para_id, db_path)| async move {
715
                    wait_for_paritydb_lock(&db_path, MAX_DB_RESTART_TIMEOUT)
716
                        .await
717
                        .map_err(|e| {
718
                            log::warn!(
719
                                "Error waiting for chain {} to release db lock: {:?}",
720
                                para_id,
721
                                e
722
                            );
723
                        })
724
                });
725
            futures::future::join_all(futs).await;
726
        }
727

            
728
        // Start all new container chains (usually 1)
729
        for para_id in chains_to_start {
730
            // Edge case: when starting the node it may be assigned to a container chain, so we need to
731
            // start a container chain already collating.
732
            // TODO: another edge case: if current == None, and running_chains == 0,
733
            // and chains_to_start == 1, we can start this chain as collating, and we won't need
734
            // to restart it on the next session. We need to add some extra state somewhere to
735
            // implement this properly.
736
            let start_collation = Some(para_id) == current;
737
            self.spawn(para_id, start_collation).await;
738
        }
739
    }
740
}
741

            
742
struct HandleUpdateAssignmentResult {
743
    chains_to_stop: Vec<ParaId>,
744
    chains_to_start: Vec<ParaId>,
745
    #[allow(dead_code)] // no longer used except in tests
746
    need_to_restart: bool,
747
}
748

            
749
// This is a separate function to allow testing
750
35
fn handle_update_assignment_state_change(
751
35
    state: &mut ContainerChainSpawnerState,
752
35
    orchestrator_para_id: ParaId,
753
35
    current: Option<ParaId>,
754
35
    next: Option<ParaId>,
755
35
) -> HandleUpdateAssignmentResult {
756
35
    if (state.assigned_para_id, state.next_assigned_para_id) == (current, next) {
757
        // If nothing changed there is nothing to update
758
        return HandleUpdateAssignmentResult {
759
            chains_to_stop: Default::default(),
760
            chains_to_start: Default::default(),
761
            need_to_restart: false,
762
        };
763
35
    }
764
35

            
765
35
    // Create a set with the container chains that were running before, and the container
766
35
    // chains that should be running after the updated assignment. This is used to calculate
767
35
    // the difference, and stop and start the required container chains.
768
35
    let mut running_chains_before = HashSet::new();
769
35
    let mut running_chains_after = HashSet::new();
770
35

            
771
35
    running_chains_before.extend(state.assigned_para_id);
772
35
    running_chains_before.extend(state.next_assigned_para_id);
773
35
    // Ignore orchestrator_para_id because it is handled in a special way, as it does not need to
774
35
    // start one session before in order to sync.
775
35
    running_chains_before.remove(&orchestrator_para_id);
776
35

            
777
35
    running_chains_after.extend(current);
778
35
    running_chains_after.extend(next);
779
35
    running_chains_after.remove(&orchestrator_para_id);
780
35
    let mut need_to_restart_current = false;
781
35
    let mut need_to_restart_next = false;
782
35

            
783
35
    if state.assigned_para_id != current {
784
24
        if let Some(para_id) = current {
785
            // If the assigned container chain has changed, we may need to
786
            // restart it in collation mode, unless it is the orchestrator chain.
787
16
            if para_id != orchestrator_para_id {
788
13
                need_to_restart_current = true;
789
13
            }
790
8
        }
791

            
792
24
        if let Some(para_id) = state.assigned_para_id {
793
18
            if para_id != orchestrator_para_id && Some(para_id) == next {
794
2
                need_to_restart_next = true;
795
16
            }
796
6
        }
797
11
    }
798

            
799
35
    state.assigned_para_id = current;
800
35
    state.next_assigned_para_id = next;
801
35

            
802
35
    let mut chains_to_stop: Vec<_> = running_chains_before
803
35
        .difference(&running_chains_after)
804
35
        .copied()
805
35
        .collect();
806
35
    let mut chains_to_start: Vec<_> = running_chains_after
807
35
        .difference(&running_chains_before)
808
35
        .copied()
809
35
        .collect();
810
35

            
811
35
    if need_to_restart_current {
812
        // Force restart of new assigned container chain: if it was running before it was in "syncing mode",
813
        // which doesn't use the correct ports, so start it in "collation mode".
814
13
        let id = current.unwrap();
815
13
        if running_chains_before.contains(&id) && !chains_to_stop.contains(&id) {
816
6
            chains_to_stop.push(id);
817
7
        }
818
13
        if !chains_to_start.contains(&id) {
819
6
            chains_to_start.push(id);
820
7
        }
821
22
    }
822

            
823
35
    if need_to_restart_next {
824
        // Handle edge case of going from (2000, 2001) to (2001, 2000). In that case we must restart both chains,
825
        // because previously 2000 was collating and now 2000 will only be syncing.
826
2
        let id = next.unwrap();
827
2
        if running_chains_before.contains(&id) && !chains_to_stop.contains(&id) {
828
2
            chains_to_stop.push(id);
829
2
        }
830
2
        if !chains_to_start.contains(&id) {
831
2
            chains_to_start.push(id);
832
2
        }
833
33
    }
834

            
835
    HandleUpdateAssignmentResult {
836
35
        chains_to_stop,
837
35
        chains_to_start,
838
35
        need_to_restart: need_to_restart_current || need_to_restart_next,
839
    }
840
35
}
841

            
842
/// Select [SyncMode] to use for a container chain.
843
/// We want to use warp sync unless the db still exists, or the container chain is
844
/// still at genesis block (because of a warp sync bug in that case).
845
///
846
/// Remember that warp sync doesn't work if a partially synced database already exists, it falls
847
/// back to full sync instead. The only exception is if the previous instance of the database was
848
/// interrupted before it finished downloading the state, in that case the node will use warp sync.
849
/// If it was interrupted during the block history download, the node will use full sync but also
850
/// finish the block history download in the background, even if sync mode is set to full sync.
851
pub fn select_sync_mode_using_client(
852
    db_exists: bool,
853
    orchestrator_client: &Arc<ParachainClient>,
854
    container_chain_para_id: ParaId,
855
) -> sc_service::error::Result<SyncMode> {
856
    if db_exists {
857
        // If the user wants to use warp sync, they should have already removed the database
858
        return Ok(SyncMode::Full);
859
    }
860

            
861
    // The following check is only needed because of this bug:
862
    // https://github.com/paritytech/polkadot-sdk/issues/1930
863

            
864
    let orchestrator_runtime_api = orchestrator_client.runtime_api();
865
    let orchestrator_chain_info = orchestrator_client.chain_info();
866

            
867
    // If the container chain is still at genesis block, use full sync because warp sync is broken
868
    let full_sync_needed = orchestrator_runtime_api
869
        .latest_author(orchestrator_chain_info.best_hash, container_chain_para_id)
870
        .map_err(|e| format!("Failed to read latest author: {}", e))?
871
        .is_none();
872

            
873
    if full_sync_needed {
874
        Ok(SyncMode::Full)
875
    } else {
876
        Ok(SyncMode::Warp)
877
    }
878
}
879

            
880
async fn get_latest_container_block_number_from_orchestrator(
881
    orchestrator_chain_interface: &Arc<dyn OrchestratorChainInterface>,
882
    orchestrator_block_hash: PHash,
883
    container_chain_para_id: ParaId,
884
) -> Option<u32> {
885
    // Get the container chain's latest block from orchestrator chain and compare with client's one
886

            
887
    orchestrator_chain_interface
888
        .latest_block_number(orchestrator_block_hash, container_chain_para_id)
889
        .await
890
        .unwrap_or_default()
891
}
892

            
893
#[derive(Debug)]
894
#[allow(dead_code)]
895
enum DbRemovalReason {
896
    HighBlockDiff {
897
        best_block_number_db: u32,
898
        best_block_number_onchain: u32,
899
    },
900
    GenesisHashMismatch {
901
        container_client_genesis_hash: H256,
902
        chain_spec_genesis_hash_v0: H256,
903
        chain_spec_genesis_hash_v1: H256,
904
    },
905
}
906

            
907
/// Given a container chain client, check if the database is valid. If not, returns `Some` with the
908
/// reason for db removal.
909
/// Reasons may be:
910
/// * High block diff: when the local db is outdated and it would take a long time to sync using full sync, we remove it to be able to use warp sync.
911
/// * Genesis hash mismatch, when the chain was deregistered and a different chain with the same para id was registered.
912
async fn db_needs_removal(
913
    container_chain_client: &Arc<ContainerChainClient>,
914
    orchestrator_chain_interface: &Arc<dyn OrchestratorChainInterface>,
915
    orchestrator_block_hash: PHash,
916
    container_chain_para_id: ParaId,
917
    container_chain_cli: &ContainerChainCli,
918
    keep_db: bool,
919
) -> sc_service::error::Result<Option<DbRemovalReason>> {
920
    // Check block diff, only needed if keep-db is false
921
    if !keep_db {
922
        // Get latest block number from the container chain client
923
        let last_container_block_temp = container_chain_client.chain_info().best_number;
924
        if last_container_block_temp == 0 {
925
            // Don't remove an empty database, as it may be in the process of a warp sync
926
        } else if get_latest_container_block_number_from_orchestrator(
927
            orchestrator_chain_interface,
928
            orchestrator_block_hash,
929
            container_chain_para_id,
930
        )
931
        .await
932
        .unwrap_or(0)
933
        .abs_diff(last_container_block_temp)
934
            > MAX_BLOCK_DIFF_FOR_FULL_SYNC
935
        {
936
            // if the diff is big, delete db and restart using warp sync
937
            return Ok(Some(DbRemovalReason::HighBlockDiff {
938
                best_block_number_db: last_container_block_temp,
939
                best_block_number_onchain: last_container_block_temp,
940
            }));
941
        }
942
    }
943

            
944
    // Generate genesis hash to compare against container client's genesis hash
945
    let container_preloaded_genesis = container_chain_cli.preloaded_chain_spec.as_ref().unwrap();
946

            
947
    // Check with both state versions, but first v1 which is the latest
948
    let block_v1: Block =
949
        generate_genesis_block(&**container_preloaded_genesis, sp_runtime::StateVersion::V1)
950
            .map_err(|e| format!("{:?}", e))?;
951
    let chain_spec_genesis_hash_v1 = block_v1.header().hash();
952

            
953
    let container_client_genesis_hash = container_chain_client.chain_info().genesis_hash;
954

            
955
    if container_client_genesis_hash != chain_spec_genesis_hash_v1 {
956
        let block_v0: Block =
957
            generate_genesis_block(&**container_preloaded_genesis, sp_runtime::StateVersion::V0)
958
                .map_err(|e| format!("{:?}", e))?;
959
        let chain_spec_genesis_hash_v0 = block_v0.header().hash();
960

            
961
        if container_client_genesis_hash != chain_spec_genesis_hash_v0 {
962
            log::info!("Container genesis V0: {:?}", chain_spec_genesis_hash_v0);
963
            log::info!("Container genesis V1: {:?}", chain_spec_genesis_hash_v1);
964
            log::info!(
965
                "Chain spec genesis {:?} did not match with any container genesis - Restarting...",
966
                container_client_genesis_hash
967
            );
968
            return Ok(Some(DbRemovalReason::GenesisHashMismatch {
969
                container_client_genesis_hash,
970
                chain_spec_genesis_hash_v0,
971
                chain_spec_genesis_hash_v1,
972
            }));
973
        }
974
    }
975

            
976
    Ok(None)
977
}
978

            
979
/// Remove the container chain database folder. This is called with db_path:
980
///     `Collator2002-01/data/containers/chains/simple_container_2002/paritydb/full-container-2002`
981
/// but we want to delete everything under
982
///     `Collator2002-01/data/containers/chains/simple_container_2002`
983
/// So we use `delete_empty_folders_recursive` to try to remove the parent folders as well, but only
984
/// if they are empty. This is to avoid removing any secret keys or other important data.
985
fn delete_container_chain_db(db_path: &Path) {
986
    // Remove folder `full-container-2002`
987
    let _ = std::fs::remove_dir_all(db_path);
988
    // Remove all the empty folders inside `simple_container_2002`, including self
989
    if let Some(parent) = db_path.ancestors().nth(2) {
990
        delete_empty_folders_recursive(parent);
991
    }
992
}
993

            
994
/// Removes all empty folders in `path`, recursively. Then, if `path` is empty, it removes it as well.
995
/// Ignores any IO errors.
996
fn delete_empty_folders_recursive(path: &Path) {
997
    let entry_iter = std::fs::read_dir(path);
998
    let entry_iter = match entry_iter {
999
        Ok(x) => x,
        Err(_e) => return,
    };
    for entry in entry_iter {
        let entry = match entry {
            Ok(x) => x,
            Err(_e) => continue,
        };
        let path = entry.path();
        if path.is_dir() {
            delete_empty_folders_recursive(&path);
        }
    }
    // Try to remove dir. Returns an error if the directory is not empty, but we ignore it.
    let _ = std::fs::remove_dir(path);
}
/// Parse a list of boot nodes in `Vec<u8>` format. Invalid boot nodes are filtered out.
3
fn parse_boot_nodes_ignore_invalid(
3
    boot_nodes_raw: Vec<Vec<u8>>,
3
    container_chain_para_id: ParaId,
3
) -> Vec<MultiaddrWithPeerId> {
3
    boot_nodes_raw
3
        .into_iter()
3
        .filter_map(|x| {
3
            let x = String::from_utf8(x)
3
                .map_err(|e| {
1
                    log::debug!(
                        "Invalid boot node in container chain {}: {}",
                        container_chain_para_id,
                        e
                    );
3
                })
3
                .ok()?;
2
            x.parse::<MultiaddrWithPeerId>()
2
                .map_err(|e| {
1
                    log::debug!(
                        "Invalid boot node in container chain {}: {}",
                        container_chain_para_id,
                        e
                    )
2
                })
2
                .ok()
3
        })
3
        .collect()
3
}
pub async fn wait_for_paritydb_lock(db_path: &Path, max_timeout: Duration) -> Result<(), String> {
    let now = Instant::now();
    while now.elapsed() < max_timeout {
        let lock_held = check_paritydb_lock_held(db_path)
            .map_err(|e| format!("Failed to check if lock file is held: {}", e))?;
        if !lock_held {
            return Ok(());
        }
        sleep(Duration::from_secs(1)).await;
    }
    Err("Timeout when waiting for paritydb lock".to_string())
}
/// Given a path to a paritydb database, check if its lock file is held. This indicates that a
/// background process is still using the database, so we should wait before trying to open it.
///
/// This should be kept up to date with the way paritydb handles the lock file:
/// <https://github.com/paritytech/parity-db/blob/2b6820e310a08678d4540c044f41a93d87343ac8/src/db.rs#L215>
fn check_paritydb_lock_held(db_path: &Path) -> Result<bool, std::io::Error> {
    if !db_path.is_dir() {
        // Lock file does not exist, so it is not held
        return Ok(false);
    }
    let mut lock_path: std::path::PathBuf = db_path.to_owned();
    lock_path.push("lock");
    let lock_file = std::fs::OpenOptions::new()
        .create(true)
        .read(true)
        .write(true)
        .truncate(true)
        .open(lock_path.as_path())?;
    // Check if the lock file is busy by trying to lock it.
    // Returns err if failed to adquire the lock.
    let lock_held = lock_file.try_lock_exclusive().is_err();
    Ok(lock_held)
}
#[cfg(test)]
mod tests {
    use {super::*, std::path::PathBuf};
    // Copy of ContainerChainSpawner with extra assertions for tests, and mocked spawn function.
    struct MockContainerChainSpawner {
        state: Arc<Mutex<ContainerChainSpawnerState>>,
        orchestrator_para_id: ParaId,
        collate_on_tanssi: Arc<
            dyn Fn() -> (CancellationToken, futures::channel::oneshot::Receiver<()>) + Send + Sync,
        >,
        collation_cancellation_constructs: Option<()>,
        // Keep track of the last CollateOn message, for tests
        currently_collating_on: Arc<Mutex<Option<ParaId>>>,
    }
    impl MockContainerChainSpawner {
10
        fn new() -> Self {
10
            let orchestrator_para_id = 1000.into();
10
            // The node always starts as an orchestrator chain collator
10
            let currently_collating_on = Arc::new(Mutex::new(Some(orchestrator_para_id)));
10
            let currently_collating_on2 = currently_collating_on.clone();
10
            let collate_closure = move || {
3
                let mut cco = currently_collating_on2.lock().unwrap();
3
                assert_ne!(
3
                    *cco,
3
                    Some(orchestrator_para_id),
                    "Received CollateOn message when we were already collating on this chain: {}",
                    orchestrator_para_id
                );
3
                *cco = Some(orchestrator_para_id);
3
                let (_, receiver) = futures::channel::oneshot::channel();
3
                (CancellationToken::new(), receiver)
3
            };
10
            let collate_on_tanssi: Arc<
10
                dyn Fn() -> (CancellationToken, futures::channel::oneshot::Receiver<()>)
10
                    + Send
10
                    + Sync,
10
            > = Arc::new(collate_closure);
10

            
10
            Self {
10
                state: Arc::new(Mutex::new(ContainerChainSpawnerState {
10
                    spawned_container_chains: Default::default(),
10
                    assigned_para_id: Some(orchestrator_para_id),
10
                    next_assigned_para_id: None,
10
                    failed_para_ids: Default::default(),
10
                    spawned_containers_monitor: Default::default(),
10
                })),
10
                orchestrator_para_id,
10
                collate_on_tanssi,
10
                // Some if collator starts on orchestrator chain
10
                collation_cancellation_constructs: Some(()),
10
                currently_collating_on,
10
            }
10
        }
21
        fn spawn(&self, container_chain_para_id: ParaId, start_collation: bool) {
21
            let (signal, _on_exit) = oneshot::channel();
21
            let currently_collating_on2 = self.currently_collating_on.clone();
21
            let collate_closure = move || {
13
                let mut cco = currently_collating_on2.lock().unwrap();
13
                assert_ne!(
13
                    *cco,
13
                    Some(container_chain_para_id),
                    "Received CollateOn message when we were already collating on this chain: {}",
                    container_chain_para_id
                );
13
                *cco = Some(container_chain_para_id);
13
                let (_, receiver) = futures::channel::oneshot::channel();
13
                (CancellationToken::new(), receiver)
13
            };
21
            let collate_on: Arc<
21
                dyn Fn() -> (CancellationToken, futures::channel::oneshot::Receiver<()>)
21
                    + Send
21
                    + Sync,
21
            > = Arc::new(collate_closure);
21
            // Dummy db_path for tests, is not actually used
21
            let db_path = PathBuf::from(format!("/tmp/container-{}/db", container_chain_para_id));
21

            
21
            let old = self
21
                .state
21
                .lock()
21
                .expect("poison error")
21
                .spawned_container_chains
21
                .insert(
21
                    container_chain_para_id,
21
                    ContainerChainState {
21
                        stop_handle: StopContainerChain { signal, id: 0 },
21
                        db_path,
21
                    },
21
                );
21

            
21
            assert!(
21
                old.is_none(),
                "tried to spawn a container chain that was already running: {}",
                container_chain_para_id
            );
21
            if start_collation {
13
                let (_cancellation_token, _exit_receiver) = collate_on();
13
            }
21
        }
15
        fn stop(&self, container_chain_para_id: ParaId) {
15
            let stop_handle = self
15
                .state
15
                .lock()
15
                .expect("poison error")
15
                .spawned_container_chains
15
                .remove(&container_chain_para_id);
15

            
15
            match stop_handle {
15
                Some(_stop_handle) => {
15
                    log::info!("Stopping container chain {}", container_chain_para_id);
                }
                None => {
                    panic!(
                        "Tried to stop a container chain that is not running: {}",
                        container_chain_para_id
                    );
                }
            }
            // Update currently_collating_on, if we stopped the chain we are no longer collating there
15
            let mut lco = self.currently_collating_on.lock().unwrap();
15
            if *lco == Some(container_chain_para_id) {
7
                *lco = None;
8
            }
15
        }
35
        fn handle_update_assignment(&mut self, current: Option<ParaId>, next: Option<ParaId>) {
35
            let HandleUpdateAssignmentResult {
35
                chains_to_stop,
35
                chains_to_start,
35
                need_to_restart,
35
            } = handle_update_assignment_state_change(
35
                &mut self.state.lock().unwrap(),
35
                self.orchestrator_para_id,
35
                current,
35
                next,
35
            );
35

            
35
            if current != Some(self.orchestrator_para_id) {
                // If not assigned to orchestrator chain anymore, we need to stop the collator process
27
                let mut cco = self.currently_collating_on.lock().unwrap();
27
                if *cco == Some(self.orchestrator_para_id) {
10
                    *cco = None;
17
                }
27
                self.collation_cancellation_constructs = None;
8
            } else if self.collation_cancellation_constructs.is_none() {
3
                let (_cancellation_token, _exit_notification_receiver) = (self.collate_on_tanssi)();
3
                self.collation_cancellation_constructs = Some(());
5
            }
            // Assert we never start and stop the same container chain
56
            for para_id in &chains_to_start {
21
                if !need_to_restart {
4
                    assert!(
4
                        !chains_to_stop.contains(para_id),
                        "Tried to start and stop same container chain: {}",
                        para_id
                    );
                } else {
                    // Will try to start and stop container chain with id "current" or "next", so ignore that
17
                    if Some(*para_id) != current && Some(*para_id) != next {
                        assert!(
                            !chains_to_stop.contains(para_id),
                            "Tried to start and stop same container chain: {}",
                            para_id
                        );
17
                    }
                }
            }
            // Assert we never start or stop the orchestrator chain
35
            assert!(!chains_to_start.contains(&self.orchestrator_para_id));
35
            assert!(!chains_to_stop.contains(&self.orchestrator_para_id));
            // Stop all container chains that are no longer needed
50
            for para_id in chains_to_stop {
15
                self.stop(para_id);
15
            }
            // Start all new container chains (usually 1)
56
            for para_id in chains_to_start {
21
                // Edge case: when starting the node it may be assigned to a container chain, so we need to
21
                // start a container chain already collating.
21
                let start_collation = Some(para_id) == current;
21
                self.spawn(para_id, start_collation);
21
            }
            // Assert that if we are currently assigned to a container chain, we are collating there
35
            if let Some(para_id) = current {
24
                self.assert_collating_on(Some(para_id));
24
            } else {
11
                self.assert_collating_on(None);
11
            }
35
        }
        #[track_caller]
71
        fn assert_collating_on(&self, para_id: Option<ParaId>) {
71
            let currently_collating_on = *self.currently_collating_on.lock().unwrap();
71
            assert_eq!(currently_collating_on, para_id);
71
        }
        #[track_caller]
36
        fn assert_running_chains(&self, para_ids: &[ParaId]) {
36
            let mut actually_running: Vec<ParaId> = self
36
                .state
36
                .lock()
36
                .unwrap()
36
                .spawned_container_chains
36
                .keys()
36
                .cloned()
36
                .collect();
36
            actually_running.sort();
36
            let mut should_be_running = para_ids.to_vec();
36
            should_be_running.sort();
36
            assert_eq!(actually_running, should_be_running);
36
        }
    }
    #[test]
1
    fn starts_collating_on_tanssi() {
1
        let mut m = MockContainerChainSpawner::new();
1
        m.assert_collating_on(Some(1000.into()));
1
        m.assert_running_chains(&[]);
1

            
1
        m.handle_update_assignment(None, None);
1
        m.assert_collating_on(None);
1
        m.assert_running_chains(&[]);
1
    }
    #[test]
1
    fn assigned_to_orchestrator_chain() {
1
        let mut m = MockContainerChainSpawner::new();
1

            
1
        m.handle_update_assignment(Some(1000.into()), Some(1000.into()));
1
        m.assert_collating_on(Some(1000.into()));
1
        m.assert_running_chains(&[]);
1

            
1
        m.handle_update_assignment(Some(1000.into()), None);
1
        m.assert_collating_on(Some(1000.into()));
1
        m.assert_running_chains(&[]);
1

            
1
        m.handle_update_assignment(None, None);
1
        m.assert_collating_on(None);
1
        m.assert_running_chains(&[]);
1

            
1
        m.handle_update_assignment(None, Some(1000.into()));
1
        m.assert_collating_on(None);
1
        m.assert_running_chains(&[]);
1

            
1
        m.handle_update_assignment(Some(1000.into()), Some(1000.into()));
1
        m.assert_collating_on(Some(1000.into()));
1
        m.assert_running_chains(&[]);
1
    }
    #[test]
1
    fn assigned_to_container_chain() {
1
        let mut m = MockContainerChainSpawner::new();
1

            
1
        m.handle_update_assignment(Some(2000.into()), Some(2000.into()));
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into()]);
1

            
1
        m.handle_update_assignment(Some(2000.into()), None);
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into()]);
1

            
1
        m.handle_update_assignment(None, None);
1
        m.assert_collating_on(None);
1
        m.assert_running_chains(&[]);
1

            
1
        m.handle_update_assignment(None, Some(2000.into()));
1
        m.assert_collating_on(None);
1
        m.assert_running_chains(&[2000.into()]);
1

            
1
        m.handle_update_assignment(Some(2000.into()), Some(2000.into()));
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into()]);
1
    }
    #[test]
1
    fn spawn_container_chains() {
1
        let mut m = MockContainerChainSpawner::new();
1

            
1
        m.handle_update_assignment(Some(1000.into()), Some(2000.into()));
1
        m.assert_collating_on(Some(1000.into()));
1
        m.assert_running_chains(&[2000.into()]);
1

            
1
        m.handle_update_assignment(Some(2000.into()), Some(2000.into()));
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into()]);
1

            
1
        m.handle_update_assignment(Some(2000.into()), Some(2001.into()));
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into(), 2001.into()]);
1

            
1
        m.handle_update_assignment(Some(2001.into()), Some(2001.into()));
1
        m.assert_collating_on(Some(2001.into()));
1
        m.assert_running_chains(&[2001.into()]);
1

            
1
        m.handle_update_assignment(Some(2001.into()), Some(1000.into()));
1
        m.assert_collating_on(Some(2001.into()));
1
        m.assert_running_chains(&[2001.into()]);
1

            
1
        m.handle_update_assignment(Some(1000.into()), Some(1000.into()));
1
        m.assert_collating_on(Some(1000.into()));
1
        m.assert_running_chains(&[]);
1
    }
    #[test]
1
    fn swap_current_next() {
1
        // Going from (2000, 2001) to (2001, 2000) shouldn't start or stop any container chains
1
        let mut m: MockContainerChainSpawner = MockContainerChainSpawner::new();
1

            
1
        m.handle_update_assignment(Some(2000.into()), Some(2001.into()));
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into(), 2001.into()]);
1

            
1
        m.handle_update_assignment(Some(2001.into()), Some(2000.into()));
1
        m.assert_collating_on(Some(2001.into()));
1
        m.assert_running_chains(&[2000.into(), 2001.into()]);
1
    }
    #[test]
1
    fn stop_collating_orchestrator() {
1
        let mut m: MockContainerChainSpawner = MockContainerChainSpawner::new();
1

            
1
        m.handle_update_assignment(Some(1000.into()), Some(1000.into()));
1
        m.assert_collating_on(Some(1000.into()));
1
        m.assert_running_chains(&[]);
1

            
1
        m.handle_update_assignment(Some(1000.into()), None);
1
        m.assert_collating_on(Some(1000.into()));
1
        m.assert_running_chains(&[]);
1

            
1
        m.handle_update_assignment(None, None);
1
        m.assert_collating_on(None);
1
        m.assert_running_chains(&[]);
1

            
1
        m.handle_update_assignment(Some(1000.into()), None);
1
        m.assert_collating_on(Some(1000.into()));
1
        m.assert_running_chains(&[]);
1
    }
    #[test]
1
    fn stop_collating_container() {
1
        let mut m: MockContainerChainSpawner = MockContainerChainSpawner::new();
1

            
1
        m.handle_update_assignment(Some(2000.into()), None);
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into()]);
1

            
1
        m.handle_update_assignment(None, None);
1
        m.assert_collating_on(None);
1
        m.assert_running_chains(&[]);
1

            
1
        m.handle_update_assignment(None, Some(2000.into()));
1
        m.assert_collating_on(None);
1
        m.assert_running_chains(&[2000.into()]);
1

            
1
        // This will send a CollateOn message to the same chain as the last CollateOn,
1
        // but this is needed because that chain has been stopped
1
        m.handle_update_assignment(Some(2000.into()), Some(2000.into()));
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into()]);
1
    }
    #[test]
1
    fn stop_collating_container_start_immediately() {
1
        let mut m: MockContainerChainSpawner = MockContainerChainSpawner::new();
1

            
1
        m.handle_update_assignment(Some(2000.into()), None);
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into()]);
1

            
1
        m.handle_update_assignment(None, None);
1
        m.assert_collating_on(None);
1
        m.assert_running_chains(&[]);
1

            
1
        // This will start the chain already collating
1
        m.handle_update_assignment(Some(2000.into()), Some(2000.into()));
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into()]);
1
    }
    #[test]
1
    fn stop_all_chains() {
1
        let mut m: MockContainerChainSpawner = MockContainerChainSpawner::new();
1

            
1
        m.handle_update_assignment(Some(2000.into()), Some(2001.into()));
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into(), 2001.into()]);
1

            
1
        m.handle_update_assignment(None, None);
1
        m.assert_collating_on(None);
1
        m.assert_running_chains(&[]);
1
    }
    #[test]
1
    fn keep_collating_on_container() {
1
        let mut m: MockContainerChainSpawner = MockContainerChainSpawner::new();
1

            
1
        m.handle_update_assignment(Some(2000.into()), None);
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into()]);
1

            
1
        m.handle_update_assignment(None, Some(2000.into()));
1
        m.assert_collating_on(None);
1
        m.assert_running_chains(&[2000.into()]);
1

            
1
        m.handle_update_assignment(Some(2000.into()), Some(2000.into()));
1
        m.assert_collating_on(Some(2000.into()));
1
        m.assert_running_chains(&[2000.into()]);
1
    }
    #[test]
1
    fn invalid_boot_nodes_are_ignored() {
1
        let para_id = 100.into();
1
        let bootnode1 =
1
            b"/ip4/127.0.0.1/tcp/33049/ws/p2p/12D3KooWHVMhQDHBpj9vQmssgyfspYecgV6e3hH1dQVDUkUbCYC9"
1
                .to_vec();
1
        assert_eq!(
1
            parse_boot_nodes_ignore_invalid(vec![b"A".to_vec()], para_id),
1
            vec![]
1
        );
1
        assert_eq!(
1
            parse_boot_nodes_ignore_invalid(vec![b"\xff".to_vec()], para_id),
1
            vec![]
1
        );
        // Valid boot nodes are not ignored
1
        assert_eq!(
1
            parse_boot_nodes_ignore_invalid(vec![bootnode1], para_id).len(),
1
            1
1
        );
1
    }
    #[test]
1
    fn path_ancestors() {
1
        // Test the implementation of `delete_container_chain_db`
1
        let db_path = PathBuf::from("/tmp/zombienet/Collator2002-01/data/containers/chains/simple_container_2002/paritydb/full-container-2002");
1
        let parent = db_path.ancestors().nth(2).unwrap();
1

            
1
        assert_eq!(
1
            parent,
1
            PathBuf::from(
1
                "/tmp/zombienet/Collator2002-01/data/containers/chains/simple_container_2002"
1
            )
1
        )
1
    }
}