fix(cron): prevent daily jobs from skipping days (48h jump) #17852 (#17903)

Merged via /review-pr -> /prepare-pr -> /merge-pr.

Prepared head SHA: 1ffe6a45af
Co-authored-by: pierreeurope <248892285+pierreeurope@users.noreply.github.com>
Co-authored-by: sebslight <19554889+sebslight@users.noreply.github.com>
Reviewed-by: @sebslight

src/cron/service/jobs.ts

@@ -102,6 +102,35 @@ export function computeJobNextRunAtMs(job: CronJob, nowMs: number): number | und
 /** Maximum consecutive schedule errors before auto-disabling a job. */
 const MAX_SCHEDULE_ERRORS = 3;
 
+function recordScheduleComputeError(params: {
+  state: CronServiceState;
+  job: CronJob;
+  err: unknown;
+}): boolean {
+  const { state, job, err } = params;
+  const errorCount = (job.state.scheduleErrorCount ?? 0) + 1;
+  const errText = String(err);
+
+  job.state.scheduleErrorCount = errorCount;
+  job.state.nextRunAtMs = undefined;
+  job.state.lastError = `schedule error: ${errText}`;
+
+  if (errorCount >= MAX_SCHEDULE_ERRORS) {
+    job.enabled = false;
+    state.deps.log.error(
+      { jobId: job.id, name: job.name, errorCount, err: errText },
+      "cron: auto-disabled job after repeated schedule errors",
+    );
+  } else {
+    state.deps.log.warn(
+      { jobId: job.id, name: job.name, errorCount, err: errText },
+      "cron: failed to compute next run for job (skipping)",
+    );
+  }
+
+  return true;
+}
+
 function normalizeJobTickState(params: { state: CronServiceState; job: CronJob; nowMs: number }): {
   changed: boolean;
   skip: boolean;
@@ -184,23 +213,8 @@ export function recomputeNextRuns(state: CronServiceState): boolean {
           changed = true;
         }
       } catch (err) {
-        const errorCount = (job.state.scheduleErrorCount ?? 0) + 1;
-        job.state.scheduleErrorCount = errorCount;
-        job.state.nextRunAtMs = undefined;
-        job.state.lastError = `schedule error: ${String(err)}`;
-        changed = true;
-
-        if (errorCount >= MAX_SCHEDULE_ERRORS) {
-          job.enabled = false;
-          state.deps.log.error(
-            { jobId: job.id, name: job.name, errorCount, err: String(err) },
-            "cron: auto-disabled job after repeated schedule errors",
-          );
-        } else {
-          state.deps.log.warn(
-            { jobId: job.id, name: job.name, errorCount, err: String(err) },
-            "cron: failed to compute next run for job (skipping)",
-          );
+        if (recordScheduleComputeError({ state, job, err })) {
+          changed = true;
         }
       }
     }
@@ -222,10 +236,21 @@ export function recomputeNextRunsForMaintenance(state: CronServiceState): boolea
     // If a job was past-due but not found by findDueJobs, recomputing would
     // cause it to be silently skipped.
     if (job.state.nextRunAtMs === undefined) {
-      const newNext = computeJobNextRunAtMs(job, now);
-      if (newNext !== undefined) {
-        job.state.nextRunAtMs = newNext;
-        changed = true;
+      try {
+        const newNext = computeJobNextRunAtMs(job, now);
+        if (job.state.nextRunAtMs !== newNext) {
+          job.state.nextRunAtMs = newNext;
+          changed = true;
+        }
+        // Clear schedule error count on successful computation.
+        if (job.state.scheduleErrorCount) {
+          job.state.scheduleErrorCount = undefined;
+          changed = true;
+        }
+      } catch (err) {
+        if (recordScheduleComputeError({ state, job, err })) {
+          changed = true;
+        }
       }
     }
     return changed;

src/cron/service/timer.ts

@@ -7,7 +7,6 @@ import { sweepCronRunSessions } from "../session-reaper.js";
 import {
   computeJobNextRunAtMs,
   nextWakeAtMs,
-  recomputeNextRuns,
   recomputeNextRunsForMaintenance,
   resolveJobPayloadTextForMain,
 } from "./jobs.js";
@@ -283,7 +282,12 @@ export async function onTimer(state: CronServiceState) {
           }
         }
 
-        recomputeNextRuns(state);
+        // Use maintenance-only recompute to avoid advancing past-due
+        // nextRunAtMs values that became due between findDueJobs and this
+        // locked block.  The full recomputeNextRuns would silently skip
+        // those jobs (advancing nextRunAtMs without execution), causing
+        // daily cron schedules to jump 48 h instead of 24 h (#17852).
+        recomputeNextRunsForMaintenance(state);
         await persist(state);
       });
     }