At least once VS exactly once

[jarl-dk]

On https://github.com/mperham/sidekiq/wiki/Best-Practices#2-make-your-job-idempotent-and-transactional there is a description of a scenario where Sidekiq will perform the worker job more than once. The scenarios is if the Worker job fails in which case sidekiq will retry at a later time.

Does Sidekiq guarantee that Workers jobs that completes successfully runs exactly once. Or does Sidekiq not gurantee this either?

[mperham]

FYI the term "worker" is outdated, the modern term is "job".

Even a job which has completed can be re-run. Redis can go down between the point where your job finished but before Sidekiq has acknowledged it in Redis. Sidekiq makes no exactly-once guarantee at all.

[jarl-dk]

It would be lovely if this non-guarantee could be emphasized on the page. Because the current example (of failing job) seems to indicate that this is the (only) situation where a job may run more than once.

[nateberkopec]

I agree. I've added Mike's reply to that section 👍

[jarl-dk]

Follow up question:
When/if sidekiq jobs are performed more than once, will they have the same or different jid?
Please also provide this information in the documentation (https://github.com/sidekiq/sidekiq/wiki/Best-Practices#2-make-your-job-idempotent-and-transactional).

[mperham]

A different JID is a different job. We are talking about the same job instance, same JID, executing twice.

[jarl-dk]

Thanks. Consider emphasize this on the documentation page.