fix: recover faster after network outage (fixes #49)

lib/activity-state.js

@@ -26,4 +26,8 @@ export class ActivityState {
       Zinnia.activity.info('SPARK retrieval reporting resumed')
     }
   }
+  //expose the healthy state
+  isHealthy() {
+    return this.#healthy === true
+  }
 }

lib/constants.js

@@ -7,3 +7,4 @@ export const RPC_AUTH = 'KZLIUb9ejreYOm-mZFM3UNADE0ux6CrHjxnS2D2Qgb8='
 export const MINER_TO_PEERID_CONTRACT_ADDRESS =
   '0x14183aD016Ddc83D638425D6328009aa390339Ce' // Contract address on the Filecoin EVM
 export const MAX_REQUEST_DURATION_MS = 90_000
+export const OFFLINE_RETRY_DELAY_MS = 5_000 // 5 seconds

lib/spark.js

@@ -7,6 +7,7 @@ import {
   APPROX_ROUND_LENGTH_IN_MS,
   MAX_JITTER_BETWEEN_TASKS_IN_MS,
   MAX_REQUEST_DURATION_MS,
+  OFFLINE_RETRY_DELAY_MS,
 } from './constants.js'
 import { queryTheIndex } from './ipni-client.js'
 import { assertOkResponse } from './http-assertions.js'
@@ -273,7 +274,10 @@ export default class Spark {
         this.handleRunError(err)
       }
       const duration = Date.now() - started
+      const isHealthy = this.#activity.isHealthy()
+
       const delay = calculateDelayBeforeNextTask({
+        isHealthy,
         roundLengthInMs: APPROX_ROUND_LENGTH_IN_MS,
         maxJitterInMs: MAX_JITTER_BETWEEN_TASKS_IN_MS,
         maxTasksPerRound: this.#tasker.maxTasksPerRound,
@@ -282,8 +286,7 @@ export default class Spark {
 
       if (delay > 0) {
         console.log(
-          'Sleeping for %s seconds before starting the next task...',
-          Math.round(delay / 1000),
+          `${isHealthy ? 'Online' : 'Offline'} – sleeping for ${Math.round(delay / 1000)}s before next task.`,
         )
         await sleep(delay)
         console.log() // add an empty line to visually delimit logs from different tasks
@@ -309,11 +312,14 @@ export default class Spark {
  * @param {number} args.lastTaskDurationInMs
  */
 export function calculateDelayBeforeNextTask({
+  isHealthy = true,
   roundLengthInMs,
   maxJitterInMs = 0,
   maxTasksPerRound,
   lastTaskDurationInMs,
 }) {
+  if (!isHealthy) return OFFLINE_RETRY_DELAY_MS
+
   const baseDelay = roundLengthInMs / maxTasksPerRound
   const delay = baseDelay - lastTaskDurationInMs
   const base = Math.min(delay, 60_000)

test/spark.js

@@ -10,7 +10,7 @@ import {
   assertLessOrEqual,
   assertGreaterOrEqual,
 } from 'zinnia:assert'
-import { SPARK_VERSION } from '../lib/constants.js'
+import { SPARK_VERSION, OFFLINE_RETRY_DELAY_MS } from '../lib/constants.js'
 
 const KNOWN_CID = 'bafkreih25dih6ug3xtj73vswccw423b56ilrwmnos4cbwhrceudopdp5sq'
 
@@ -504,6 +504,22 @@ test('calculateDelayBeforeNextTask() introduces random jitter for zero tasks in
   )
 })
 
+test('calculateDelayBeforeNextTask() returns OFFLINE_RETRY_DELAY_MS when offline', () => {
+  const delay = calculateDelayBeforeNextTask({
+    isHealthy: false,
+    roundLengthInMs: 20 * 60_000,
+    maxJitterInMs: 10_000,
+    maxTasksPerRound: 1,
+    lastTaskDurationInMs: 1000,
+  })
+
+  assertEquals(
+    delay,
+    OFFLINE_RETRY_DELAY_MS,
+    `Expected delay to match OFFLINE_RETRY_DELAY_MS (${OFFLINE_RETRY_DELAY_MS}) when offline`,
+  )
+})
+
 test('fetchCAR triggers timeout after long retrieval', async () => {
   const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms))
   const fetch = async (_url, { signal }) => {