fedimint_client/module/init/
recovery.rs

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
use std::time::Duration;
use std::{cmp, ops};

use fedimint_api_client::api::DynGlobalApi;
use fedimint_core::db::DatabaseTransaction;
use fedimint_core::encoding::{Decodable, Encodable};
use fedimint_core::epoch::ConsensusItem;
use fedimint_core::module::registry::ModuleDecoderRegistry;
use fedimint_core::module::{ApiVersion, ModuleCommon};
use fedimint_core::session_outcome::{AcceptedItem, SessionStatus};
use fedimint_core::task::{MaybeSend, MaybeSync, ShuttingDownError, TaskGroup};
use fedimint_core::transaction::Transaction;
use fedimint_core::{apply, async_trait_maybe_send, OutPoint};
use fedimint_logging::LOG_CLIENT_RECOVERY;
use futures::{Stream, StreamExt as _};
use rand::{thread_rng, Rng as _};
use serde::{Deserialize, Serialize};
use tracing::{debug, info, trace, warn};

use super::{ClientModuleInit, ClientModuleRecoverArgs};
use crate::module::recovery::RecoveryProgress;
use crate::module::{ClientContext, ClientModule};

#[derive(Debug, Clone, Eq, PartialEq, Encodable, Decodable, Serialize, Deserialize)]
/// Common state tracked during recovery from history
pub struct RecoveryFromHistoryCommon {
    start_session: u64,
    next_session: u64,
    end_session: u64,
}

impl RecoveryFromHistoryCommon {
    pub fn new(start_session: u64, next_session: u64, end_session: u64) -> Self {
        Self {
            start_session,
            next_session,
            end_session,
        }
    }
}

/// Module specific logic for [`ClientModuleRecoverArgs::recover_from_history`]
///
/// See [`ClientModuleRecoverArgs::recover_from_history`] for more information.
#[apply(async_trait_maybe_send!)]
pub trait RecoveryFromHistory: std::fmt::Debug + MaybeSend + MaybeSync + Clone {
    /// [`ClientModuleInit`] of this recovery logic.
    type Init: ClientModuleInit;

    /// New empty state to start recovery from, and session number to start from
    async fn new(
        init: &Self::Init,
        args: &ClientModuleRecoverArgs<Self::Init>,
        snapshot: Option<&<<Self::Init as ClientModuleInit>::Module as ClientModule>::Backup>,
    ) -> anyhow::Result<(Self, u64)>;

    /// Try to load the existing state previously stored with
    /// [`RecoveryFromHistory::store_dbtx`].
    ///
    /// Storing and restoring progress is used to save progress and
    /// continue recovery if it was previously terminated before completion.
    async fn load_dbtx(
        init: &Self::Init,
        dbtx: &mut DatabaseTransaction<'_>,
        args: &ClientModuleRecoverArgs<Self::Init>,
    ) -> anyhow::Result<Option<(Self, RecoveryFromHistoryCommon)>>;

    /// Store the current recovery state in the database
    ///
    /// See [`Self::load_dbtx`].
    async fn store_dbtx(
        &self,
        dbtx: &mut DatabaseTransaction<'_>,
        common: &RecoveryFromHistoryCommon,
    );

    /// Delete the recovery state from the database
    ///
    /// See [`Self::load_dbtx`].
    async fn delete_dbtx(&self, dbtx: &mut DatabaseTransaction<'_>);

    /// Read the finalization status
    ///
    /// See [`Self::load_dbtx`].
    async fn load_finalized(dbtx: &mut DatabaseTransaction<'_>) -> Option<bool>;

    /// Store finalization status
    ///
    /// See [`Self::load_finalized`].
    async fn store_finalized(dbtx: &mut DatabaseTransaction<'_>, state: bool);

    /// Handle session outcome, adjusting the current state
    ///
    /// It is expected that most implementations don't need to override this
    /// function, and override more granular ones instead (e.g.
    /// [`Self::handle_input`] and/or [`Self::handle_output`]).
    ///
    /// The default implementation will loop through items in the
    /// `session.items` and forward them one by one to respective functions
    /// (see [`Self::handle_transaction`]).
    async fn handle_session(
        &mut self,
        client_ctx: &ClientContext<<Self::Init as ClientModuleInit>::Module>,
        session_idx: u64,
        session_items: &Vec<AcceptedItem>,
    ) -> anyhow::Result<()> {
        for accepted_item in session_items {
            if let ConsensusItem::Transaction(ref transaction) = accepted_item.item {
                self.handle_transaction(client_ctx, transaction, session_idx)
                    .await?;
            }
        }
        Ok(())
    }

    /// Handle session outcome, adjusting the current state
    ///
    /// It is expected that most implementations don't need to override this
    /// function, and override more granular ones instead (e.g.
    /// [`Self::handle_input`] and/or [`Self::handle_output`]).
    ///
    /// The default implementation will loop through inputs and outputs
    /// of the transaction, filter and downcast ones matching current module
    /// and forward them one by one to respective functions
    /// (e.g. [`Self::handle_input`], [`Self::handle_output`]).
    async fn handle_transaction(
        &mut self,
        client_ctx: &ClientContext<<Self::Init as ClientModuleInit>::Module>,
        transaction: &Transaction,
        session_idx: u64,
    ) -> anyhow::Result<()> {
        trace!(
            target: LOG_CLIENT_RECOVERY,
            tx_hash = %transaction.tx_hash(),
            input_num = transaction.inputs.len(),
            output_num = transaction.outputs.len(),
            "processing transaction"
        );

        for (idx, input) in transaction.inputs.iter().enumerate() {
            trace!(
                target: LOG_CLIENT_RECOVERY,
                tx_hash = %transaction.tx_hash(),
                idx,
                module_id = input.module_instance_id(),
                "found transaction input"
            );

            if let Some(own_input) = client_ctx.input_from_dyn(input) {
                self.handle_input(client_ctx, idx, own_input, session_idx)
                    .await?;
            }
        }

        for (out_idx, output) in transaction.outputs.iter().enumerate() {
            trace!(
                target: LOG_CLIENT_RECOVERY,
                tx_hash = %transaction.tx_hash(),
                idx = out_idx,
                module_id = output.module_instance_id(),
                "found transaction output"
            );

            if let Some(own_output) = client_ctx.output_from_dyn(output) {
                let out_point = OutPoint {
                    txid: transaction.tx_hash(),
                    out_idx: out_idx as u64,
                };

                self.handle_output(client_ctx, out_point, own_output, session_idx)
                    .await?;
            }
        }

        Ok(())
    }

    /// Handle transaction input, adjusting the current state
    ///
    /// Default implementation does nothing.
    async fn handle_input(
        &mut self,
        _client_ctx: &ClientContext<<Self::Init as ClientModuleInit>::Module>,
        _idx: usize,
        _input: &<<<Self::Init as ClientModuleInit>::Module as ClientModule>::Common as ModuleCommon>::Input,
        _session_idx: u64,
    ) -> anyhow::Result<()> {
        Ok(())
    }

    /// Handle transaction output, adjusting the current state
    ///
    /// Default implementation does nothing.
    async fn handle_output(
        &mut self,
        _client_ctx: &ClientContext<<Self::Init as ClientModuleInit>::Module>,
        _out_point: OutPoint,
        _output: &<<<Self::Init as ClientModuleInit>::Module as ClientModule>::Common as ModuleCommon>::Output,
        _session_idx: u64,
    ) -> anyhow::Result<()> {
        Ok(())
    }

    /// Called before `finalize_dbtx`, to allow final state changes outside
    /// of retriable database transaction.
    async fn pre_finalize(&mut self) -> anyhow::Result<()> {
        Ok(())
    }

    /// Finalize the recovery converting the tracked state to final
    /// changes in the database.
    ///
    /// This is the only place during recovery where module gets a chance to
    /// create state machines, etc.
    ///
    /// Notably this function is running in a database-autocommit wrapper, so
    /// might be called again on database commit failure.
    async fn finalize_dbtx(&self, dbtx: &mut DatabaseTransaction<'_>) -> anyhow::Result<()>;
}

impl<Init> ClientModuleRecoverArgs<Init>
where
    Init: ClientModuleInit,
{
    /// Run recover of a module from federation consensus history
    ///
    /// It is expected that most modules will implement their recovery
    /// by following Federation consensus history to restore their
    /// state. This function implement such a recovery by being generic
    /// over [`RecoveryFromHistory`] trait, which provides module-specific
    /// parts of recovery logic.
    pub async fn recover_from_history<Recovery>(
        &self,
        init: &Init,
        snapshot: Option<&<<Init as ClientModuleInit>::Module as ClientModule>::Backup>,
    ) -> anyhow::Result<()>
    where
        Recovery: RecoveryFromHistory<Init = Init> + std::fmt::Debug,
    {
        /// Fetch epochs in a given range and send them over `sender`
        ///
        /// Since WASM's `spawn` does not support join handles, we indicate
        /// errors via `sender` itself.
        fn fetch_block_stream<'a>(
            api: DynGlobalApi,
            core_api_version: ApiVersion,
            decoders: ModuleDecoderRegistry,
            epoch_range: ops::Range<u64>,
            task_group: TaskGroup,
        ) -> impl futures::Stream<Item = Result<(u64, Vec<AcceptedItem>), ShuttingDownError>> + 'a
        {
            // How many request for blocks to run in parallel (streaming).
            const PARALLISM_LEVEL: usize = 64;
            const VERSION_THAT_INTRODUCED_GET_SESSION_STATUS: ApiVersion =
                ApiVersion { major: 0, minor: 1 };

            futures::stream::iter(epoch_range.clone())
                .map(move |session_idx| {
                    let api = api.clone();
                    let decoders = decoders.clone();
                    let task_group = task_group.clone();

                    Box::pin(async move {
                        // NOTE: Each block is fetched in a spawned task. This avoids a footgun
                        // of stuff in streams not making any progress when the stream itself
                        // is not being polled, and possibly can increase the fetching performance.
                        task_group.spawn_cancellable("recovery fetch block", async move {

                            info!(session_idx, "Fetching epoch");

                            let mut retry_sleep = Duration::from_millis(10);
                            let block = loop {
                                trace!(target: LOG_CLIENT_RECOVERY, session_idx, "Awaiting signed block");

                                let items_res = if core_api_version < VERSION_THAT_INTRODUCED_GET_SESSION_STATUS {
                                    api.await_block(session_idx, &decoders).await.map(|s| s.items)
                                } else {
                                    api.get_session_status(session_idx, &decoders).await.map(|s| match s {
                                        SessionStatus::Initial => panic!("Federation missing session that existed when we started recovery"),
                                        SessionStatus::Pending(items) => items,
                                        SessionStatus::Complete(s) => s.items,
                                    })
                                };

                                match items_res {
                                    Ok(block) => {
                                        debug!(target: LOG_CLIENT_RECOVERY, session_idx, "Got signed session");
                                        break block
                                    },
                                    Err(e) => {
                                        const MAX_SLEEP: Duration = Duration::from_secs(120);

                                        warn!(target: LOG_CLIENT_RECOVERY, e = %e, session_idx, "Error trying to fetch signed block");
                                        // We don't want PARALLISM_LEVEL tasks hammering Federation
                                        // with requests, so max sleep is significant
                                        if retry_sleep <= MAX_SLEEP {
                                            retry_sleep = retry_sleep
                                                + thread_rng().gen_range(Duration::ZERO..=retry_sleep);
                                        }
                                        fedimint_core::runtime::sleep(cmp::min(retry_sleep, MAX_SLEEP))
                                            .await;
                                    }
                                }
                            };

                            (session_idx, block)
                        }).await.expect("Can't fail")
                    })
                })
                .buffered(PARALLISM_LEVEL)
        }

        /// Make enough progress to justify saving a state snapshot

        async fn make_progress<'a, Init, Recovery: RecoveryFromHistory<Init = Init>>(
            client_ctx: &ClientContext<<Init as ClientModuleInit>::Module>,
            common_state: &mut RecoveryFromHistoryCommon,
            state: &mut Recovery,
            block_stream: &mut (impl Stream<Item = Result<(u64, Vec<AcceptedItem>), ShuttingDownError>>
                      + Unpin),
        ) -> anyhow::Result<()>
        where
            Init: ClientModuleInit,
        {
            /// the amount of blocks after which we save progress in the
            /// database (return from this function)
            ///
            /// TODO: Instead of a fixed range of session
            /// indexes, make the loop time-based, so the amount of
            /// progress we can loose on termination is time-bound,
            /// and thus more adaptive.
            const PROGRESS_SNAPSHOT_BLOCKS: u64 = 10;

            let block_range = common_state.next_session
                ..cmp::min(
                    common_state
                        .next_session
                        .wrapping_add(PROGRESS_SNAPSHOT_BLOCKS),
                    common_state.end_session,
                );

            debug!(
                target: LOG_CLIENT_RECOVERY,
                ?block_range,
                "Processing blocks"
            );

            for _ in block_range {
                let Some(res) = block_stream.next().await else {
                    break;
                };

                let (session_idx, accepted_items) = res?;

                assert_eq!(common_state.next_session, session_idx);
                state
                    .handle_session(client_ctx, session_idx, &accepted_items)
                    .await?;

                common_state.next_session += 1;
            }

            Ok(())
        }

        let db = self.db();
        let client_ctx = self.context();

        if Recovery::load_finalized(&mut db.begin_transaction_nc().await)
            .await
            .unwrap_or_default()
        {
            // In rare circumstances, the finalization could complete, yet the completion
            // of `recover` function not yet persisted in the database. So
            // it's possible that `recovery` would be called again on an
            // already finalized state. Because of this we store a
            // finalization marker in the same dbtx as the finalization itself, detect this
            // here and exit early.
            //
            // Example sequence how this happens (if `finalize_dbtx` didn't exist):
            //
            // 0. module recovery is complete and progress saved to the db
            // 1. `dbtx` with finalization commits, progress deleted, completing recovery on
            //    the client module side
            // 2. client crashes/gets terminated (tricky corner case)
            // 3. client starts again
            // 4. client never observed/persisted that the module finished recovery, so
            //    calls module recovery again
            // 5. module doesn't see progress, starts recovery again, eventually completes
            //    again and moves to finalization
            // 6. module runs finalization again and probably fails because it's actually
            //    not idempotent and doesn't expect the already existing state.
            warn!("Previously finalized, exiting");
            return Ok(());
        }
        let current_session_count = client_ctx.global_api().session_count().await?;
        debug!(target: LOG_CLIENT_RECOVERY, session_count = current_session_count, "Current session count");

        let (mut state, mut common_state) =
            // TODO: if load fails (e.g. module didn't migrate an existing recovery state and failed to decode it),
            // we could just ... start from scratch? at least being able to force this behavior might be useful
            if let Some((state, common_state)) = Recovery::load_dbtx(init, &mut db.begin_transaction_nc().await, self).await? {
                (state, common_state)
            } else {
                let (state, start_session) = Recovery::new(init, self, snapshot).await?;

                debug!(target: LOG_CLIENT_RECOVERY, start_session, "Recovery start session");
                (state,
                RecoveryFromHistoryCommon {
                    start_session,
                    next_session: start_session,
                    end_session: current_session_count + 1,
                })
            };

        let block_stream_session_range = common_state.next_session..common_state.end_session;
        debug!(target: LOG_CLIENT_RECOVERY, range = ?block_stream_session_range, "Starting block streaming");

        let mut block_stream = fetch_block_stream(
            self.api().clone(),
            *self.core_api_version(),
            client_ctx.decoders(),
            block_stream_session_range,
            self.task_group().clone(),
        );
        let client_ctx = self.context();

        while common_state.next_session < common_state.end_session {
            make_progress(
                &client_ctx,
                &mut common_state,
                &mut state,
                &mut block_stream,
            )
            .await?;

            let mut dbtx = db.begin_transaction().await;
            state.store_dbtx(&mut dbtx.to_ref_nc(), &common_state).await;
            dbtx.commit_tx().await;

            self.update_recovery_progress(RecoveryProgress {
                complete: (common_state.next_session - common_state.start_session)
                    .try_into()
                    .unwrap_or(u32::MAX),
                total: (common_state.end_session - common_state.start_session)
                    .try_into()
                    .unwrap_or(u32::MAX),
            });
        }

        state.pre_finalize().await?;

        let mut dbtx = db.begin_transaction().await;
        state.store_dbtx(&mut dbtx.to_ref_nc(), &common_state).await;
        dbtx.commit_tx().await;

        debug!(
            target: LOG_CLIENT_RECOVERY,
            ?state,
            "Finalizing restore"
        );

        db.autocommit(
            |dbtx, _| {
                let state = state.clone();
                {
                    Box::pin(async move {
                        state.delete_dbtx(dbtx).await;
                        state.finalize_dbtx(dbtx).await?;
                        Recovery::store_finalized(dbtx, true).await;

                        Ok::<_, anyhow::Error>(())
                    })
                }
            },
            None,
        )
        .await?;

        Ok(())
    }
}