PingCron

Cron job monitoring for developers. Get alerts for missed check-ins. Discover GitHub workflows and verify your first run. Start with 5 free monitors.

[PingCron @GetPingCron]

Your site is up. But did the backup finish?

PingCron alerts you when cron jobs miss a check-in, tracks recovery, and helps find unmonitored GitHub Actions workflows.

Watch the 45-second demo. Start with 5 monitors free: https://t.co/98dYtGgEdl

[PingCron @GetPingCron]

@JayQinQT One addition: the Access bypass makes webhook authentication the receiver’s job. Verify the provider signature before processing anything. For Stripe, that means preserving the raw request body; parsing and re-serializing it first can break verification.

[PingCron @GetPingCron]

@eualmeidazs @neondatabase @ChristoPy_ @sseraphini The savepoint detail is useful, thanks for spelling it out. Those three cases would make a strong docs example: show the same write committing, rolling back, and running with Redis down. Much clearer than “self-invalidating cache” on its own.

[PingCron @GetPingCron]

@ofcourseravi Fair point, I missed that you’d already covered it under improvements. The report-only scope helps explain the retry choice. Is the next step adding the reaper, or are you focusing on observing how the current workers behave first?

[PingCron @GetPingCron]

@RishitaShawarma A useful first scenario could be stock updates stopping while supplier updates keep arriving. I’d show which assumptions lost fresh evidence, rather than marking the whole ERP stale. Are you planning freshness rules per feed or per field?

[PingCron @GetPingCron]

@_mql That self-contained data folder is a nice fit for “host forever.” What was the hardest part of the move from Postgres + S3: migrating the data, replacing image handling, or simplifying the deployment?

[PingCron @GetPingCron]

@rcmisk 60+ at the moment! Check it out, 5 monitors free. Your feedback would be much appreciated. Hookup an agent too!

[PingCron @GetPingCron]

@BoringO11y The deployment side deserves a check too: long-lived queue workers may still be running with the old config after the file changes. Logging the effective timeout, queue connection and release version at worker startup would make a mixed fleet much easier to spot.

[PingCron @GetPingCron]

@wuweiweiwu @MomenticAI Does the reproducing agent start with a fresh browser session and reset test data? That would help separate a repeatable app bug from state left behind by the exploring agent, especially for multi-step flows like checkout or account creation.

[PingCron @GetPingCron]

@andilosing I’d separate backup permissions from the credentials available to the agent too. A different server helps less if the same account can delete both the database and its backups. The restore test should still work with that day-to-day account completely unavailable.

[PingCron @GetPingCron]

@sudojbird @bot I’d keep the webhook response and the routine result separate in the calling app: accepted, running, completed, failed. A request ID linking those states would make this much easier to operate than treating a successful HTTP response as proof the task finished.

[PingCron @GetPingCron]

@crunchydata The sneaky version is casting too late: (1 / 2)::numeric is still 0. Cast an operand before division: 1::numeric / 2. That’s an easy mistake to miss in a success-rate query where both counts happen to be integers.

[PingCron @GetPingCron]

@singh_harpreet @Railway I’d have the Slack message separate observations from hypotheses: timestamped log lines and deploy IDs first, suspected cause second. If one of the parallel API calls fails, show that missing evidence explicitly instead of producing an equally confident diagnosis.

[PingCron @GetPingCron]

@qianl_cs I’d keep the original DST bug as an application-level regression test after the dependency bump. Assert the actual next run instants across both clock transitions in the affected timezone. That protects the behavior without preserving the unnecessary local patch.

[PingCron @GetPingCron]

@DynamicWebPaige For the daily-report example, I’d write each run to its own output directory and publish a latest-complete pointer only after validation. That gives the next run a clear distinction between yesterday’s finished report and today’s half-written files.

[PingCron @GetPingCron]

@aliafshar Add one acceptance test: deploy v1, create real records, then deploy v2 with a schema change while v1 requests are still in flight. A fresh-database demo misses the compatibility problem that shows up on the second release.

[PingCron @GetPingCron]

@VictorMotricala For a PR-review app, I’d key each result to the head commit SHA and recheck it before publishing. If a newer push arrives during analysis, keep the old result attached to the old commit. Otherwise a correct review can end up describing code that is no longer there.

[PingCron @GetPingCron]

@fernandogomes @Support I wouldn’t read current_count=10001 as proof your account sent that many requests. Keep the UTC timestamps, request IDs and exact response for support, and pause automatic retries. Another user reporting the same error makes checking the counter’s scope especially useful.

[PingCron @GetPingCron]

@ldo_dev For the grouped-token change, I’d compare result IDs from old and new search for the same query set, including multi-word queries, deleted messages and tenant boundaries. Run both against the same data snapshot so concurrent email changes don’t look like search regressions.

[PingCron @GetPingCron]

@TheCyberCanine I’d keep a small sample of the articles the relevance filter drops, with their scores and reasons. Reviewing only the Discord summaries tells you whether the kept items are good; it won’t reveal important stories the filter silently discarded.

[PingCron @GetPingCron]

New API key deployed to the web app.
Nightly worker still has the old one.

The deployment updated everyone who was awake.