@aztec/validator-ha-signer 0.0.1-commit.e558bd1c → 0.0.1-commit.e57c76e
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +10 -2
- package/dest/db/index.d.ts +2 -1
- package/dest/db/index.d.ts.map +1 -1
- package/dest/db/index.js +1 -0
- package/dest/db/lmdb.d.ts +70 -0
- package/dest/db/lmdb.d.ts.map +1 -0
- package/dest/db/lmdb.js +223 -0
- package/dest/db/migrations/1_initial-schema.d.ts +4 -2
- package/dest/db/migrations/1_initial-schema.d.ts.map +1 -1
- package/dest/db/migrations/1_initial-schema.js +34 -4
- package/dest/db/migrations/2_add-checkpoint-number.d.ts +7 -0
- package/dest/db/migrations/2_add-checkpoint-number.d.ts.map +1 -0
- package/dest/db/migrations/2_add-checkpoint-number.js +17 -0
- package/dest/db/postgres.d.ts +4 -2
- package/dest/db/postgres.d.ts.map +1 -1
- package/dest/db/postgres.js +17 -17
- package/dest/db/schema.d.ts +10 -9
- package/dest/db/schema.d.ts.map +1 -1
- package/dest/db/schema.js +13 -7
- package/dest/db/types.d.ts +46 -21
- package/dest/db/types.d.ts.map +1 -1
- package/dest/db/types.js +31 -15
- package/dest/factory.d.ts +39 -4
- package/dest/factory.d.ts.map +1 -1
- package/dest/factory.js +81 -7
- package/dest/metrics.d.ts +51 -0
- package/dest/metrics.d.ts.map +1 -0
- package/dest/metrics.js +103 -0
- package/dest/slashing_protection_service.d.ts +12 -3
- package/dest/slashing_protection_service.d.ts.map +1 -1
- package/dest/slashing_protection_service.js +19 -8
- package/dest/types.d.ts +18 -70
- package/dest/types.d.ts.map +1 -1
- package/dest/types.js +4 -20
- package/dest/validator_ha_signer.d.ts +12 -4
- package/dest/validator_ha_signer.d.ts.map +1 -1
- package/dest/validator_ha_signer.js +19 -9
- package/package.json +11 -7
- package/src/db/index.ts +1 -0
- package/src/db/lmdb.ts +308 -0
- package/src/db/migrations/1_initial-schema.ts +35 -4
- package/src/db/migrations/2_add-checkpoint-number.ts +19 -0
- package/src/db/postgres.ts +17 -15
- package/src/db/schema.ts +13 -7
- package/src/db/types.ts +66 -19
- package/src/factory.ts +111 -6
- package/src/metrics.ts +138 -0
- package/src/slashing_protection_service.ts +30 -9
- package/src/types.ts +38 -104
- package/src/validator_ha_signer.ts +37 -13
- package/dest/config.d.ts +0 -101
- package/dest/config.d.ts.map +0 -1
- package/dest/config.js +0 -92
- package/src/config.ts +0 -149
package/src/db/types.ts
CHANGED
|
@@ -1,6 +1,7 @@
|
|
|
1
|
-
import
|
|
2
|
-
import
|
|
1
|
+
import { BlockNumber, CheckpointNumber, type IndexWithinCheckpoint, SlotNumber } from '@aztec/foundation/branded-types';
|
|
2
|
+
import { EthAddress } from '@aztec/foundation/eth-address';
|
|
3
3
|
import type { Signature } from '@aztec/foundation/eth-signature';
|
|
4
|
+
import { DutyType } from '@aztec/stdlib/ha-signing';
|
|
4
5
|
|
|
5
6
|
/**
|
|
6
7
|
* Row type from PostgreSQL query
|
|
@@ -10,6 +11,7 @@ export interface DutyRow {
|
|
|
10
11
|
validator_address: string;
|
|
11
12
|
slot: string;
|
|
12
13
|
block_number: string;
|
|
14
|
+
checkpoint_number: string;
|
|
13
15
|
block_index_within_checkpoint: number;
|
|
14
16
|
duty_type: DutyType;
|
|
15
17
|
status: DutyStatus;
|
|
@@ -23,24 +25,35 @@ export interface DutyRow {
|
|
|
23
25
|
}
|
|
24
26
|
|
|
25
27
|
/**
|
|
26
|
-
*
|
|
28
|
+
* Plain-primitive representation of a duty record suitable for serialization
|
|
29
|
+
* (e.g. msgpackr for LMDB). All domain types are stored as their string/number
|
|
30
|
+
* equivalents. Timestamps are Unix milliseconds.
|
|
27
31
|
*/
|
|
28
|
-
export interface
|
|
29
|
-
|
|
32
|
+
export interface StoredDutyRecord {
|
|
33
|
+
rollupAddress: string;
|
|
34
|
+
validatorAddress: string;
|
|
35
|
+
slot: string;
|
|
36
|
+
blockNumber: string;
|
|
37
|
+
checkpointNumber: string;
|
|
38
|
+
blockIndexWithinCheckpoint: number;
|
|
39
|
+
dutyType: DutyType;
|
|
40
|
+
status: DutyStatus;
|
|
41
|
+
messageHash: string;
|
|
42
|
+
signature?: string;
|
|
43
|
+
nodeId: string;
|
|
44
|
+
lockToken: string;
|
|
45
|
+
/** Unix timestamp in milliseconds when signing started */
|
|
46
|
+
startedAtMs: number;
|
|
47
|
+
/** Unix timestamp in milliseconds when signing completed */
|
|
48
|
+
completedAtMs?: number;
|
|
49
|
+
errorMessage?: string;
|
|
30
50
|
}
|
|
31
51
|
|
|
32
52
|
/**
|
|
33
|
-
*
|
|
53
|
+
* Row type from INSERT_OR_GET_DUTY query (includes is_new flag)
|
|
34
54
|
*/
|
|
35
|
-
export
|
|
36
|
-
|
|
37
|
-
CHECKPOINT_PROPOSAL = 'CHECKPOINT_PROPOSAL',
|
|
38
|
-
ATTESTATION = 'ATTESTATION',
|
|
39
|
-
ATTESTATIONS_AND_SIGNERS = 'ATTESTATIONS_AND_SIGNERS',
|
|
40
|
-
GOVERNANCE_VOTE = 'GOVERNANCE_VOTE',
|
|
41
|
-
SLASHING_VOTE = 'SLASHING_VOTE',
|
|
42
|
-
AUTH_REQUEST = 'AUTH_REQUEST',
|
|
43
|
-
TXS = 'TXS',
|
|
55
|
+
export interface InsertOrGetRow extends DutyRow {
|
|
56
|
+
is_new: boolean;
|
|
44
57
|
}
|
|
45
58
|
|
|
46
59
|
/**
|
|
@@ -51,8 +64,12 @@ export enum DutyStatus {
|
|
|
51
64
|
SIGNED = 'signed',
|
|
52
65
|
}
|
|
53
66
|
|
|
67
|
+
// Re-export DutyType from stdlib
|
|
68
|
+
export { DutyType };
|
|
69
|
+
|
|
54
70
|
/**
|
|
55
|
-
*
|
|
71
|
+
* Rich representation of a validator duty, with branded types and Date objects.
|
|
72
|
+
* This is the common output type returned by all SlashingProtectionDatabase implementations.
|
|
56
73
|
*/
|
|
57
74
|
export interface ValidatorDutyRecord {
|
|
58
75
|
/** Ethereum address of the rollup contract */
|
|
@@ -61,8 +78,10 @@ export interface ValidatorDutyRecord {
|
|
|
61
78
|
validatorAddress: EthAddress;
|
|
62
79
|
/** Slot number for this duty */
|
|
63
80
|
slot: SlotNumber;
|
|
64
|
-
/** Block number for this duty */
|
|
81
|
+
/** Block number for this duty (0 for non-block-proposal duties) */
|
|
65
82
|
blockNumber: BlockNumber;
|
|
83
|
+
/** Checkpoint number for this duty (0 for attestation and vote duties) */
|
|
84
|
+
checkpointNumber: CheckpointNumber;
|
|
66
85
|
/** Block index within checkpoint (0, 1, 2... for block proposals, -1 for other duty types) */
|
|
67
86
|
blockIndexWithinCheckpoint: number;
|
|
68
87
|
/** Type of duty being performed */
|
|
@@ -85,6 +104,32 @@ export interface ValidatorDutyRecord {
|
|
|
85
104
|
errorMessage?: string;
|
|
86
105
|
}
|
|
87
106
|
|
|
107
|
+
/**
|
|
108
|
+
* Convert a {@link StoredDutyRecord} (plain-primitive wire format) to a
|
|
109
|
+
* {@link ValidatorDutyRecord} (rich domain type).
|
|
110
|
+
*
|
|
111
|
+
* Shared by LMDB and any future non-Postgres backend implementations.
|
|
112
|
+
*/
|
|
113
|
+
export function recordFromFields(stored: StoredDutyRecord): ValidatorDutyRecord {
|
|
114
|
+
return {
|
|
115
|
+
rollupAddress: EthAddress.fromString(stored.rollupAddress),
|
|
116
|
+
validatorAddress: EthAddress.fromString(stored.validatorAddress),
|
|
117
|
+
slot: SlotNumber.fromString(stored.slot),
|
|
118
|
+
blockNumber: BlockNumber.fromString(stored.blockNumber),
|
|
119
|
+
checkpointNumber: CheckpointNumber.fromString(stored.checkpointNumber),
|
|
120
|
+
blockIndexWithinCheckpoint: stored.blockIndexWithinCheckpoint,
|
|
121
|
+
dutyType: stored.dutyType,
|
|
122
|
+
status: stored.status,
|
|
123
|
+
messageHash: stored.messageHash,
|
|
124
|
+
signature: stored.signature,
|
|
125
|
+
nodeId: stored.nodeId,
|
|
126
|
+
lockToken: stored.lockToken,
|
|
127
|
+
startedAt: new Date(stored.startedAtMs),
|
|
128
|
+
completedAt: stored.completedAtMs !== undefined ? new Date(stored.completedAtMs) : undefined,
|
|
129
|
+
errorMessage: stored.errorMessage,
|
|
130
|
+
};
|
|
131
|
+
}
|
|
132
|
+
|
|
88
133
|
/**
|
|
89
134
|
* Duty identifier for block proposals.
|
|
90
135
|
* blockIndexWithinCheckpoint is REQUIRED and must be >= 0.
|
|
@@ -163,8 +208,10 @@ export function getBlockIndexFromDutyIdentifier(duty: DutyIdentifier): number {
|
|
|
163
208
|
* Additional parameters for checking and recording a new duty
|
|
164
209
|
*/
|
|
165
210
|
interface CheckAndRecordExtra {
|
|
166
|
-
/** Block number for this duty */
|
|
167
|
-
blockNumber: BlockNumber
|
|
211
|
+
/** Block number for this duty (0 for non-block-proposal duties) */
|
|
212
|
+
blockNumber: BlockNumber;
|
|
213
|
+
/** Checkpoint number for this duty (0 for attestation and vote duties) */
|
|
214
|
+
checkpointNumber: CheckpointNumber;
|
|
168
215
|
/** The signing root (hash) for this duty */
|
|
169
216
|
messageHash: string;
|
|
170
217
|
/** Identifier for the node that acquired the lock */
|
package/src/factory.ts
CHANGED
|
@@ -1,11 +1,17 @@
|
|
|
1
1
|
/**
|
|
2
2
|
* Factory functions for creating validator HA signers
|
|
3
3
|
*/
|
|
4
|
+
import { DateProvider } from '@aztec/foundation/timer';
|
|
5
|
+
import { createStore } from '@aztec/kv-store/lmdb-v2';
|
|
6
|
+
import type { LocalSignerConfig, ValidatorHASignerConfig } from '@aztec/stdlib/ha-signing';
|
|
7
|
+
import { getTelemetryClient } from '@aztec/telemetry-client';
|
|
8
|
+
|
|
4
9
|
import { Pool } from 'pg';
|
|
5
10
|
|
|
6
|
-
import
|
|
11
|
+
import { LmdbSlashingProtectionDatabase, migrateLmdbSlashingProtectionDatabase } from './db/lmdb.js';
|
|
7
12
|
import { PostgresSlashingProtectionDatabase } from './db/postgres.js';
|
|
8
|
-
import
|
|
13
|
+
import { HASignerMetrics } from './metrics.js';
|
|
14
|
+
import type { CreateHASignerDeps, CreateLocalSignerWithProtectionDeps, SlashingProtectionDatabase } from './types.js';
|
|
9
15
|
import { ValidatorHASigner } from './validator_ha_signer.js';
|
|
10
16
|
|
|
11
17
|
/**
|
|
@@ -23,7 +29,6 @@ import { ValidatorHASigner } from './validator_ha_signer.js';
|
|
|
23
29
|
* ```typescript
|
|
24
30
|
* const { signer, db } = await createHASigner({
|
|
25
31
|
* databaseUrl: process.env.DATABASE_URL,
|
|
26
|
-
* haSigningEnabled: true,
|
|
27
32
|
* nodeId: 'validator-node-1',
|
|
28
33
|
* pollingIntervalMs: 100,
|
|
29
34
|
* signingTimeoutMs: 3000,
|
|
@@ -52,14 +57,19 @@ export async function createHASigner(
|
|
|
52
57
|
const { databaseUrl, poolMaxCount, poolMinCount, poolIdleTimeoutMs, poolConnectionTimeoutMs, ...signerConfig } =
|
|
53
58
|
config;
|
|
54
59
|
|
|
55
|
-
|
|
60
|
+
const databaseUrlValue = databaseUrl?.getValue();
|
|
61
|
+
if (!databaseUrlValue) {
|
|
56
62
|
throw new Error('databaseUrl is required for createHASigner');
|
|
57
63
|
}
|
|
64
|
+
|
|
65
|
+
const telemetryClient = deps?.telemetryClient ?? getTelemetryClient();
|
|
66
|
+
const dateProvider = deps?.dateProvider ?? new DateProvider();
|
|
67
|
+
|
|
58
68
|
// Create connection pool (or use provided pool)
|
|
59
69
|
let pool: Pool;
|
|
60
70
|
if (!deps?.pool) {
|
|
61
71
|
pool = new Pool({
|
|
62
|
-
connectionString:
|
|
72
|
+
connectionString: databaseUrlValue,
|
|
63
73
|
max: poolMaxCount ?? 10,
|
|
64
74
|
min: poolMinCount ?? 0,
|
|
65
75
|
idleTimeoutMillis: poolIdleTimeoutMs ?? 10_000,
|
|
@@ -75,8 +85,103 @@ export async function createHASigner(
|
|
|
75
85
|
// Verify database schema is initialized and version matches
|
|
76
86
|
await db.initialize();
|
|
77
87
|
|
|
88
|
+
// Create metrics
|
|
89
|
+
const metrics = new HASignerMetrics(telemetryClient, signerConfig.nodeId);
|
|
90
|
+
|
|
78
91
|
// Create signer
|
|
79
|
-
const signer = new ValidatorHASigner(db, {
|
|
92
|
+
const signer = new ValidatorHASigner(db, signerConfig, { metrics, dateProvider });
|
|
80
93
|
|
|
81
94
|
return { signer, db };
|
|
82
95
|
}
|
|
96
|
+
|
|
97
|
+
/**
|
|
98
|
+
* Create a local (single-node) signing protection signer backed by LMDB.
|
|
99
|
+
*
|
|
100
|
+
* This provides double-signing protection for nodes that are NOT running in a
|
|
101
|
+
* high-availability (multi-node) setup. It prevents a proposer from sending two
|
|
102
|
+
* proposals for the same slot if the node crashes and restarts mid-proposal.
|
|
103
|
+
*
|
|
104
|
+
* When `config.dataDirectory` is set, the protection database is persisted to disk
|
|
105
|
+
* and survives crashes/restarts. When unset, an ephemeral in-memory store is
|
|
106
|
+
* used which protects within a single run but not across restarts.
|
|
107
|
+
*
|
|
108
|
+
* @param config - Local signer config
|
|
109
|
+
* @param deps - Optional dependencies (telemetry, date provider).
|
|
110
|
+
* @returns An object containing the signer and database instances.
|
|
111
|
+
*/
|
|
112
|
+
export async function createLocalSignerWithProtection(
|
|
113
|
+
config: LocalSignerConfig,
|
|
114
|
+
deps?: CreateLocalSignerWithProtectionDeps,
|
|
115
|
+
): Promise<{
|
|
116
|
+
signer: ValidatorHASigner;
|
|
117
|
+
db: SlashingProtectionDatabase;
|
|
118
|
+
}> {
|
|
119
|
+
const telemetryClient = deps?.telemetryClient ?? getTelemetryClient();
|
|
120
|
+
const dateProvider = deps?.dateProvider ?? new DateProvider();
|
|
121
|
+
|
|
122
|
+
const kvStore = await createStore(
|
|
123
|
+
'signing-protection',
|
|
124
|
+
LmdbSlashingProtectionDatabase.SCHEMA_VERSION,
|
|
125
|
+
{
|
|
126
|
+
dataDirectory: config.dataDirectory,
|
|
127
|
+
dataStoreMapSizeKb: config.signingProtectionMapSizeKb ?? config.dataStoreMapSizeKb,
|
|
128
|
+
rollupAddress: config.rollupAddress,
|
|
129
|
+
},
|
|
130
|
+
undefined,
|
|
131
|
+
{
|
|
132
|
+
onUpgrade: (dataDirectory, currentVersion, latestVersion) =>
|
|
133
|
+
migrateLmdbSlashingProtectionDatabase(
|
|
134
|
+
dataDirectory,
|
|
135
|
+
currentVersion,
|
|
136
|
+
latestVersion,
|
|
137
|
+
config.signingProtectionMapSizeKb ?? config.dataStoreMapSizeKb,
|
|
138
|
+
),
|
|
139
|
+
schemaVersionMismatchPolicy: 'throw',
|
|
140
|
+
},
|
|
141
|
+
);
|
|
142
|
+
|
|
143
|
+
const db = new LmdbSlashingProtectionDatabase(kvStore, dateProvider);
|
|
144
|
+
|
|
145
|
+
const signerConfig = {
|
|
146
|
+
...config,
|
|
147
|
+
nodeId: config.nodeId || 'local',
|
|
148
|
+
};
|
|
149
|
+
|
|
150
|
+
const metrics = new HASignerMetrics(telemetryClient, signerConfig.nodeId, 'LocalSigningProtectionMetrics');
|
|
151
|
+
|
|
152
|
+
const signer = new ValidatorHASigner(db, signerConfig, { metrics, dateProvider });
|
|
153
|
+
|
|
154
|
+
return { signer, db };
|
|
155
|
+
}
|
|
156
|
+
|
|
157
|
+
/**
|
|
158
|
+
* Create an in-memory LMDB-backed SlashingProtectionDatabase that can be shared across
|
|
159
|
+
* multiple validator nodes in the same process. Used for testing HA setups.
|
|
160
|
+
*/
|
|
161
|
+
export async function createSharedSlashingProtectionDb(
|
|
162
|
+
dateProvider: DateProvider = new DateProvider(),
|
|
163
|
+
): Promise<SlashingProtectionDatabase> {
|
|
164
|
+
const kvStore = await createStore('shared-signing-protection', LmdbSlashingProtectionDatabase.SCHEMA_VERSION, {
|
|
165
|
+
dataStoreMapSizeKb: 1024 * 1024,
|
|
166
|
+
});
|
|
167
|
+
return new LmdbSlashingProtectionDatabase(kvStore, dateProvider);
|
|
168
|
+
}
|
|
169
|
+
|
|
170
|
+
/**
|
|
171
|
+
* Create a ValidatorHASigner backed by a pre-existing SlashingProtectionDatabase.
|
|
172
|
+
* Used for testing HA setups where multiple nodes share the same protection database.
|
|
173
|
+
*/
|
|
174
|
+
export function createSignerFromSharedDb(
|
|
175
|
+
db: SlashingProtectionDatabase,
|
|
176
|
+
config: Pick<
|
|
177
|
+
ValidatorHASignerConfig,
|
|
178
|
+
'nodeId' | 'pollingIntervalMs' | 'signingTimeoutMs' | 'maxStuckDutiesAgeMs' | 'rollupAddress'
|
|
179
|
+
>,
|
|
180
|
+
deps?: CreateLocalSignerWithProtectionDeps,
|
|
181
|
+
): { signer: ValidatorHASigner; db: SlashingProtectionDatabase } {
|
|
182
|
+
const telemetryClient = deps?.telemetryClient ?? getTelemetryClient();
|
|
183
|
+
const dateProvider = deps?.dateProvider ?? new DateProvider();
|
|
184
|
+
const metrics = new HASignerMetrics(telemetryClient, config.nodeId, 'SharedSigningProtectionMetrics');
|
|
185
|
+
const signer = new ValidatorHASigner(db, config, { metrics, dateProvider });
|
|
186
|
+
return { signer, db };
|
|
187
|
+
}
|
package/src/metrics.ts
ADDED
|
@@ -0,0 +1,138 @@
|
|
|
1
|
+
import {
|
|
2
|
+
Attributes,
|
|
3
|
+
type Histogram,
|
|
4
|
+
Metrics,
|
|
5
|
+
type TelemetryClient,
|
|
6
|
+
type UpDownCounter,
|
|
7
|
+
createUpDownCounterWithDefault,
|
|
8
|
+
} from '@aztec/telemetry-client';
|
|
9
|
+
|
|
10
|
+
export type HACleanupType = 'stuck' | 'old' | 'outdated_rollup';
|
|
11
|
+
|
|
12
|
+
/**
|
|
13
|
+
* Metrics for HA signer tracking signing operations, lock acquisition, and cleanup.
|
|
14
|
+
*/
|
|
15
|
+
export class HASignerMetrics {
|
|
16
|
+
// Signing lifecycle metrics
|
|
17
|
+
private signingDuration: Histogram;
|
|
18
|
+
private signingSuccessCount: UpDownCounter;
|
|
19
|
+
private dutyAlreadySignedCount: UpDownCounter;
|
|
20
|
+
private slashingProtectionCount: UpDownCounter;
|
|
21
|
+
private signingErrorCount: UpDownCounter;
|
|
22
|
+
|
|
23
|
+
// Lock acquisition metrics
|
|
24
|
+
private lockAcquiredCount: UpDownCounter;
|
|
25
|
+
|
|
26
|
+
// Cleanup metrics
|
|
27
|
+
private cleanupStuckDutiesCount: UpDownCounter;
|
|
28
|
+
private cleanupOldDutiesCount: UpDownCounter;
|
|
29
|
+
private cleanupOutdatedRollupDutiesCount: UpDownCounter;
|
|
30
|
+
|
|
31
|
+
constructor(
|
|
32
|
+
client: TelemetryClient,
|
|
33
|
+
private nodeId: string,
|
|
34
|
+
name = 'HASignerMetrics',
|
|
35
|
+
) {
|
|
36
|
+
const meter = client.getMeter(name);
|
|
37
|
+
|
|
38
|
+
// Signing lifecycle
|
|
39
|
+
this.signingDuration = meter.createHistogram(Metrics.HA_SIGNER_SIGNING_DURATION);
|
|
40
|
+
this.signingSuccessCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_SIGNING_SUCCESS_COUNT);
|
|
41
|
+
this.dutyAlreadySignedCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_DUTY_ALREADY_SIGNED_COUNT);
|
|
42
|
+
this.slashingProtectionCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_SLASHING_PROTECTION_COUNT);
|
|
43
|
+
this.signingErrorCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_SIGNING_ERROR_COUNT);
|
|
44
|
+
|
|
45
|
+
// Lock acquisition
|
|
46
|
+
this.lockAcquiredCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_LOCK_ACQUIRED_COUNT);
|
|
47
|
+
|
|
48
|
+
// Cleanup
|
|
49
|
+
this.cleanupStuckDutiesCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_CLEANUP_STUCK_DUTIES_COUNT);
|
|
50
|
+
this.cleanupOldDutiesCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_CLEANUP_OLD_DUTIES_COUNT);
|
|
51
|
+
this.cleanupOutdatedRollupDutiesCount = createUpDownCounterWithDefault(
|
|
52
|
+
meter,
|
|
53
|
+
Metrics.HA_SIGNER_CLEANUP_OUTDATED_ROLLUP_DUTIES_COUNT,
|
|
54
|
+
);
|
|
55
|
+
}
|
|
56
|
+
|
|
57
|
+
/**
|
|
58
|
+
* Record a successful signing operation.
|
|
59
|
+
* @param dutyType - The type of duty signed
|
|
60
|
+
* @param durationMs - Duration from start of signWithProtection to completion
|
|
61
|
+
*/
|
|
62
|
+
public recordSigningSuccess(dutyType: string, durationMs: number): void {
|
|
63
|
+
const attributes = {
|
|
64
|
+
[Attributes.HA_DUTY_TYPE]: dutyType,
|
|
65
|
+
[Attributes.HA_NODE_ID]: this.nodeId,
|
|
66
|
+
};
|
|
67
|
+
this.signingSuccessCount.add(1, attributes);
|
|
68
|
+
this.signingDuration.record(durationMs, attributes);
|
|
69
|
+
}
|
|
70
|
+
|
|
71
|
+
/**
|
|
72
|
+
* Record a DutyAlreadySignedError (expected in HA; another node signed first).
|
|
73
|
+
* @param dutyType - The type of duty
|
|
74
|
+
*/
|
|
75
|
+
public recordDutyAlreadySigned(dutyType: string): void {
|
|
76
|
+
const attributes = {
|
|
77
|
+
[Attributes.HA_DUTY_TYPE]: dutyType,
|
|
78
|
+
[Attributes.HA_NODE_ID]: this.nodeId,
|
|
79
|
+
};
|
|
80
|
+
this.dutyAlreadySignedCount.add(1, attributes);
|
|
81
|
+
}
|
|
82
|
+
|
|
83
|
+
/**
|
|
84
|
+
* Record a SlashingProtectionError (attempted to sign different data for same duty).
|
|
85
|
+
* @param dutyType - The type of duty
|
|
86
|
+
*/
|
|
87
|
+
public recordSlashingProtection(dutyType: string): void {
|
|
88
|
+
const attributes = {
|
|
89
|
+
[Attributes.HA_DUTY_TYPE]: dutyType,
|
|
90
|
+
[Attributes.HA_NODE_ID]: this.nodeId,
|
|
91
|
+
};
|
|
92
|
+
this.slashingProtectionCount.add(1, attributes);
|
|
93
|
+
}
|
|
94
|
+
|
|
95
|
+
/**
|
|
96
|
+
* Record a signing function failure (lock will be deleted for retry).
|
|
97
|
+
* @param dutyType - The type of duty
|
|
98
|
+
*/
|
|
99
|
+
public recordSigningError(dutyType: string): void {
|
|
100
|
+
const attributes = {
|
|
101
|
+
[Attributes.HA_DUTY_TYPE]: dutyType,
|
|
102
|
+
[Attributes.HA_NODE_ID]: this.nodeId,
|
|
103
|
+
};
|
|
104
|
+
this.signingErrorCount.add(1, attributes);
|
|
105
|
+
}
|
|
106
|
+
|
|
107
|
+
/**
|
|
108
|
+
* Record lock acquisition.
|
|
109
|
+
* @param acquired - Whether a new lock was acquired (true) or existing record found (false)
|
|
110
|
+
*/
|
|
111
|
+
public recordLockAcquire(acquired: boolean): void {
|
|
112
|
+
if (acquired) {
|
|
113
|
+
const attributes = {
|
|
114
|
+
[Attributes.HA_NODE_ID]: this.nodeId,
|
|
115
|
+
};
|
|
116
|
+
this.lockAcquiredCount.add(1, attributes);
|
|
117
|
+
}
|
|
118
|
+
}
|
|
119
|
+
|
|
120
|
+
/**
|
|
121
|
+
* Record cleanup metrics.
|
|
122
|
+
* @param type - Type of cleanup
|
|
123
|
+
* @param count - Number of duties cleaned up
|
|
124
|
+
*/
|
|
125
|
+
public recordCleanup(type: HACleanupType, count: number): void {
|
|
126
|
+
const attributes = {
|
|
127
|
+
[Attributes.HA_NODE_ID]: this.nodeId,
|
|
128
|
+
};
|
|
129
|
+
|
|
130
|
+
if (type === 'stuck') {
|
|
131
|
+
this.cleanupStuckDutiesCount.add(count, attributes);
|
|
132
|
+
} else if (type === 'old') {
|
|
133
|
+
this.cleanupOldDutiesCount.add(count, attributes);
|
|
134
|
+
} else if (type === 'outdated_rollup') {
|
|
135
|
+
this.cleanupOutdatedRollupDutiesCount.add(count, attributes);
|
|
136
|
+
}
|
|
137
|
+
}
|
|
138
|
+
}
|
|
@@ -7,6 +7,8 @@
|
|
|
7
7
|
import { type Logger, createLogger } from '@aztec/foundation/log';
|
|
8
8
|
import { RunningPromise } from '@aztec/foundation/promise';
|
|
9
9
|
import { sleep } from '@aztec/foundation/sleep';
|
|
10
|
+
import type { DateProvider } from '@aztec/foundation/timer';
|
|
11
|
+
import type { BaseSignerConfig } from '@aztec/stdlib/ha-signing';
|
|
10
12
|
|
|
11
13
|
import {
|
|
12
14
|
type CheckAndRecordParams,
|
|
@@ -16,7 +18,13 @@ import {
|
|
|
16
18
|
getBlockIndexFromDutyIdentifier,
|
|
17
19
|
} from './db/types.js';
|
|
18
20
|
import { DutyAlreadySignedError, SlashingProtectionError } from './errors.js';
|
|
19
|
-
import type {
|
|
21
|
+
import type { HASignerMetrics } from './metrics.js';
|
|
22
|
+
import type { SlashingProtectionDatabase } from './types.js';
|
|
23
|
+
|
|
24
|
+
export interface SlashingProtectionServiceDeps {
|
|
25
|
+
metrics: HASignerMetrics;
|
|
26
|
+
dateProvider: DateProvider;
|
|
27
|
+
}
|
|
20
28
|
|
|
21
29
|
/**
|
|
22
30
|
* Slashing Protection Service
|
|
@@ -39,12 +47,16 @@ export class SlashingProtectionService {
|
|
|
39
47
|
private readonly signingTimeoutMs: number;
|
|
40
48
|
private readonly maxStuckDutiesAgeMs: number;
|
|
41
49
|
|
|
50
|
+
private readonly metrics: HASignerMetrics;
|
|
51
|
+
private readonly dateProvider: DateProvider;
|
|
52
|
+
|
|
42
53
|
private cleanupRunningPromise: RunningPromise;
|
|
43
54
|
private lastOldDutiesCleanupAtMs?: number;
|
|
44
55
|
|
|
45
56
|
constructor(
|
|
46
57
|
private readonly db: SlashingProtectionDatabase,
|
|
47
|
-
private readonly config:
|
|
58
|
+
private readonly config: BaseSignerConfig,
|
|
59
|
+
deps: SlashingProtectionServiceDeps,
|
|
48
60
|
) {
|
|
49
61
|
this.log = createLogger('slashing-protection');
|
|
50
62
|
this.pollingIntervalMs = config.pollingIntervalMs;
|
|
@@ -53,6 +65,8 @@ export class SlashingProtectionService {
|
|
|
53
65
|
this.maxStuckDutiesAgeMs = config.maxStuckDutiesAgeMs ?? 144_000;
|
|
54
66
|
|
|
55
67
|
this.cleanupRunningPromise = new RunningPromise(this.cleanup.bind(this), this.log, this.maxStuckDutiesAgeMs);
|
|
68
|
+
this.metrics = deps.metrics;
|
|
69
|
+
this.dateProvider = deps.dateProvider;
|
|
56
70
|
}
|
|
57
71
|
|
|
58
72
|
/**
|
|
@@ -72,7 +86,7 @@ export class SlashingProtectionService {
|
|
|
72
86
|
*/
|
|
73
87
|
async checkAndRecord(params: CheckAndRecordParams): Promise<string> {
|
|
74
88
|
const { validatorAddress, slot, dutyType, messageHash, nodeId } = params;
|
|
75
|
-
const startTime =
|
|
89
|
+
const startTime = this.dateProvider.now();
|
|
76
90
|
|
|
77
91
|
this.log.debug(`Checking duty: ${dutyType} for slot ${slot}`, {
|
|
78
92
|
validatorAddress: validatorAddress.toString(),
|
|
@@ -85,10 +99,11 @@ export class SlashingProtectionService {
|
|
|
85
99
|
|
|
86
100
|
if (isNew) {
|
|
87
101
|
// We successfully acquired the lock
|
|
88
|
-
this.log.
|
|
102
|
+
this.log.verbose(`Acquired lock for duty ${dutyType} at slot ${slot}`, {
|
|
89
103
|
validatorAddress: validatorAddress.toString(),
|
|
90
104
|
nodeId,
|
|
91
105
|
});
|
|
106
|
+
this.metrics.recordLockAcquire(true);
|
|
92
107
|
return record.lockToken;
|
|
93
108
|
}
|
|
94
109
|
|
|
@@ -103,6 +118,7 @@ export class SlashingProtectionService {
|
|
|
103
118
|
existingNodeId: record.nodeId,
|
|
104
119
|
attemptingNodeId: nodeId,
|
|
105
120
|
});
|
|
121
|
+
this.metrics.recordSlashingProtection(dutyType);
|
|
106
122
|
throw new SlashingProtectionError(
|
|
107
123
|
slot,
|
|
108
124
|
dutyType,
|
|
@@ -112,15 +128,17 @@ export class SlashingProtectionService {
|
|
|
112
128
|
record.nodeId,
|
|
113
129
|
);
|
|
114
130
|
}
|
|
131
|
+
this.metrics.recordDutyAlreadySigned(dutyType);
|
|
115
132
|
throw new DutyAlreadySignedError(slot, dutyType, record.blockIndexWithinCheckpoint, record.nodeId);
|
|
116
133
|
} else if (record.status === DutyStatus.SIGNING) {
|
|
117
134
|
// Another node is currently signing - check for timeout
|
|
118
|
-
if (
|
|
135
|
+
if (this.dateProvider.now() - startTime > this.signingTimeoutMs) {
|
|
119
136
|
this.log.warn(`Timeout waiting for signing to complete for duty ${dutyType} at slot ${slot}`, {
|
|
120
137
|
validatorAddress: validatorAddress.toString(),
|
|
121
138
|
timeoutMs: this.signingTimeoutMs,
|
|
122
139
|
signingNodeId: record.nodeId,
|
|
123
140
|
});
|
|
141
|
+
this.metrics.recordDutyAlreadySigned(dutyType);
|
|
124
142
|
throw new DutyAlreadySignedError(slot, dutyType, record.blockIndexWithinCheckpoint, 'unknown (timeout)');
|
|
125
143
|
}
|
|
126
144
|
|
|
@@ -159,7 +177,7 @@ export class SlashingProtectionService {
|
|
|
159
177
|
);
|
|
160
178
|
|
|
161
179
|
if (success) {
|
|
162
|
-
this.log.
|
|
180
|
+
this.log.verbose(`Recorded successful signing for duty ${dutyType} at slot ${slot}`, {
|
|
163
181
|
validatorAddress: validatorAddress.toString(),
|
|
164
182
|
nodeId,
|
|
165
183
|
});
|
|
@@ -223,11 +241,12 @@ export class SlashingProtectionService {
|
|
|
223
241
|
*/
|
|
224
242
|
async start() {
|
|
225
243
|
// One-time cleanup at startup: remove duties from previous rollup versions
|
|
226
|
-
const numOutdatedRollupDuties = await this.db.cleanupOutdatedRollupDuties(this.config.
|
|
244
|
+
const numOutdatedRollupDuties = await this.db.cleanupOutdatedRollupDuties(this.config.rollupAddress);
|
|
227
245
|
if (numOutdatedRollupDuties > 0) {
|
|
228
246
|
this.log.info(`Cleaned up ${numOutdatedRollupDuties} duties with outdated rollup address at startup`, {
|
|
229
|
-
currentRollupAddress: this.config.
|
|
247
|
+
currentRollupAddress: this.config.rollupAddress.toString(),
|
|
230
248
|
});
|
|
249
|
+
this.metrics.recordCleanup('outdated_rollup', numOutdatedRollupDuties);
|
|
231
250
|
}
|
|
232
251
|
|
|
233
252
|
this.cleanupRunningPromise.start();
|
|
@@ -263,13 +282,14 @@ export class SlashingProtectionService {
|
|
|
263
282
|
nodeId: this.config.nodeId,
|
|
264
283
|
maxStuckDutiesAgeMs: this.maxStuckDutiesAgeMs,
|
|
265
284
|
});
|
|
285
|
+
this.metrics.recordCleanup('stuck', numStuckDuties);
|
|
266
286
|
}
|
|
267
287
|
|
|
268
288
|
// 2. Clean up old signed duties if configured
|
|
269
289
|
// we shouldn't run this as often as stuck duty cleanup.
|
|
270
290
|
if (this.config.cleanupOldDutiesAfterHours !== undefined) {
|
|
271
291
|
const maxAgeMs = this.config.cleanupOldDutiesAfterHours * 60 * 60 * 1000;
|
|
272
|
-
const nowMs =
|
|
292
|
+
const nowMs = this.dateProvider.now();
|
|
273
293
|
const shouldRun =
|
|
274
294
|
this.lastOldDutiesCleanupAtMs === undefined || nowMs - this.lastOldDutiesCleanupAtMs >= maxAgeMs;
|
|
275
295
|
if (shouldRun) {
|
|
@@ -280,6 +300,7 @@ export class SlashingProtectionService {
|
|
|
280
300
|
cleanupOldDutiesAfterHours: this.config.cleanupOldDutiesAfterHours,
|
|
281
301
|
maxAgeMs,
|
|
282
302
|
});
|
|
303
|
+
this.metrics.recordCleanup('old', numOldDuties);
|
|
283
304
|
}
|
|
284
305
|
}
|
|
285
306
|
}
|