App lifecycle management: add support for periodic task definitions

[steveej]

I couldn't identify this problem in any other of the lifecycle open issues.

Polling still exists nowadays and it would be handy to have periodic app support in the spec.

I imagine this so that one can define what happens to an application on what exit codes.
At least the following information is necessary for a correct implementation of this feature:

• Interval
• Whether or not to reschedule on failure
• If so, how many retry attempts should be made
• Which states indicate failure

[jonboulle]

Can you give some use cases? Why should this be handled by the pod lifecycle rather than within the app itself?

This is tightly related to #276

[steveej]

This is for applications that don't support re-scheduling themselves because they are not daemons and don't come with one, and which you would typically put into your crontab.
Possibly and by far not complete list of examples:

• process/collect stats
• poll mails
• run garbage collection

[jonboulle]

OK, but so far it's unclear why this shouldn't be handled at a higher level
by an orchestration layer or init system.

Stefan Junker [email protected] schrieb am Di., 17. Nov. 2015
08:50:

This is for applications that don't support re-scheduling themselves
because they are not daemons and don't come with one, and which you would
typically put into your crontab.
Possibly and by far not complete list of examples:

• process/collect stats
• poll mails
• run garbage collection

—
Reply to this email directly or view it on GitHub
#537 (comment).

[steveej]

AFAIK an orchestration layer or an init system is not a specification for a pod. I think that the pod spec is where this belongs along with other metrics about my applications which are already there. Why would I want to split up this information?

[jonboulle]

The pod lifecycle can rapidly become arbitrarily complex, so thus far we've
tried to set a very high barrier for adding different inter-process
dependencies or features like this.

On Tue, Nov 17, 2015 at 9:27 AM, Stefan Junker [email protected]
wrote:

AFAIK an orchestration layer or an init system is not a specification for
a pod. I think that the pod spec is where this belongs along with other
metrics about my applications which are already there. Why would I want to
split up this information?

—
Reply to this email directly or view it on GitHub
#537 (comment).