service and check are deregistered after a crash

[brentley]

Hi guys,

First let me say, we are loving consul!

We are starting to use it to advertise services on our mesos cluster, and while testing, discovered that, if consul crashes (simulated with a 'kill -9'), when it is restarted, it rejoins the cluster, and deregisters the service and healthcheck.

it's worth noting that when the service was started, it was registered via the http api, not through a consul.d config file.

Also, it's worth noting that we are using: "rejoin_after_leave": true

Here's a gist of the relevant portion of the log:
https://gist.github.com/brentley/87be2f33ce80e7637e6d

Now, if the service happens to die, and mesos restarts it, naturally, it gets registered via the api and all is good again.

Is this expected behavior, or is there an option we can enable to have it continue with whatever state it was in when it crashed (assuming healthchecks pass, obviously)

Thanks!
Brent

[armon]

This is currently the expected behavior since the API driven checks/services are not persisted by the local agent. The plan is to eventually persist these out to the local -data-dir so the agent can recover from a crash.

[brentley]

Might this be planned for 0.5.0?

---

Brent

On Tue, Nov 11, 2014 at 5:36 PM, Armon Dadgar notifications@github.com
wrote:

This is currently the expected behavior since the API driven
checks/services are not persisted by the local agent. The plan is to
eventually persist these out to the local -data-dir so the agent can
recover from a crash.

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

[armon]

Most likely! It's a common request.

[brentley]

that would be excellent! you have my +1. =)