Juju bootstrap fails because mongodb is unreachable

[Chris Glass]

Juju version: 1.19.4-trusty-amd64

I hit an issue this morning where bootstrapping would fail, when juju seemingly fails to start/connect to the mongodb server:

http://pastebin.ubuntu.com/7742147/

Destroying an re-bootstrapping the environment did not display the same problem, despite the versions use being the exact same (ubuntu, maas, juju and charm versions are pinned).

"can't dial mongo to initiate replicaset: no reachable servers"

[Curtis Hovey (sinzui)]

This may relate to slow servers. I am seeing this in KVM testing testing too.

[Dean Henrichsmeyer (dean)]

FWIW, we're seeing this with servers that are not slow.

[Ian Booth (wallyworld)]

Hi, I notice the issue raised occurred when running Juju 1.19.4
Before 1.20 was released, some work was done to improve how Juju agents talk to Mongo. I'd be very interested to know if the issue is still observed when running Juju 1.20. I am hoping the issue was already been fixed.

[James Troup (elmo)]

I'm seeing this with juju 1.20 on Softlayer servers which are not slow.

https://pastebin.canonical.com/113156/

[Ian Booth (wallyworld)]

We expect this issue has been fixed with the changes made to address bug 1339240 but would love confirmation that is the case

[Andrew Wilkins (axwalk)]

The original issue is a little different to the one linked by elmo (note the different error message, "Closed explicitly" vs. "no reachable servers"). I believe the issue that elmo reports has been fixed (lp:1339240).

The original error is because Mongo has taken >30s to start listening on its socket, probably because it's setting up replica sets which can take a while. The timeout has been increased on trunk and in 1.20.1.

[Andrew Wilkins (axwalk)]

Marking fix committed as per above comments, please reopen if the issue persists.