Github user cestella commented on the issue:
https://github.com/apache/metron/pull/834
Ok, so a couple of things here. I wanted to have a vigorous discussion,
because this change has some pros and cons.
At the moment, as Nick alluded, we do template install upon index topology
start. The problem is that we cannot depend on ES starting prior to the
indexing topology due to us not having a hard dependency listed there in the
[role_command_order.json](https://github.com/cestella/incubator-metron/blob/e83390a39910903ccba313a7a1b00433bf347058/metron-deployment/packaging/ambari/metron-mpack/src/main/resources/addon-services/METRON/CURRENT/role_command_order.json).
The reason why, I believe, is because we don't assume that users are using ES
managed via ambari necessarily.
All this to say that there may be a failure to start the indexing topology
upon startup, which would necessitate a restart of the indexing topology to
retry template install. The downside to what we are doing currently, though,
is that a warning is hidden in logs that users probably wont' see. They will
see green lights and the UI may or may not work (probably not). This will get
worse in ES 5, because without those templates, tuples will fail.
So, my question to you guys, is there sufficient value in failing fast here
given the broader context. I think so, but I want everyone to realize that we
may end up in a situation where indexing fails to start becuase of an ordering
problem with ES.
I will not push this through without discussion, so I'd like at least 2
+1's from committers and a week of discussion before a commit happens.
Thoughts?
---
