Run service update at a fixed time

[etzelc]

I am running some nightly-build servers in docker swarm which should be updated once during every night. So i first set SLEEP_TIME to 24h and launched the service at night. :)
Since the service updates take some time the next update window (sleep 24h) is slowly shifting into the morning. Additionally, if i change something in the service configuration and shepherd gets redeployed, i would have to do this again at night.
So, I came up with the idea to implement a configurable fixed time for the service update.

The first version is straight forward for my mentioned nightly use case. An environment variable SLEEP_UNTIL is introduced, which uses a 24-hour time format with the option to specify additional days: HH[:MM[:SS]] [add days]. This allows 24h, 48h, 72h, ... schedules that run at a fixed time. Examples: 23 --> every day at 23:00:00; 13:30:10 2 --> every third day at 13:30:10
This is the corresponding pull request: #86

I have also used watchtower in the past (which does not support swarm by the way containrrr/watchtower#43 ) and liked the idea of being able to specify cron schedules. After I finished the SLEEP_UNTIL version, I had the ambition to make it more general by using a cron schedule definition. Parsing a cron schedule in Bash was a little bit to much for me, but during my research I found out that the builtin printf of the Korn Shell ksh93 can calculate the time to the next run for a given crontab. I have not found a package of ksh93 in the alpine repos, so ksh93 is compiled during docker build. Using an environment variable SLEEP_SCHEDULE, a cron schedule can be defined to control the service updates.
The pull request for the SLEEP_SCHEDULE version: #87

I have made pull requests for both versions to have an open discussion, but my favorite is the second one using the crontab. 😀

[djmaze]

@etzelc Sorry for the wait and thanks for your effort.

It is not documented (yet), but my preferred approach is not adding any cronjob functionality here but use swarm-cronjob as a separate service instead. There a discussion about this here. Essentially, you need to set RUN_ONCE_AND_EXIT and then adjust your service as described in the swarm-cronjob documentation.

As this has now been discussed several times, I am feeling pressed to finally add this to the documentation.

If you have objections against using swarm-cronjob, please go forward. Personally, I am using swarm-cronjob all the time in my clusters, and being very happy with it.

[djmaze]

See #89.

[etzelc]

Sorry for the late response, but my notifications have been off.
Swarm-cronjob is a very good idea and doing exactly what I want. Thanks for updating the documentation. 👍🏻