Erratic behaviour on /${index}/_aliases after ROR installation

(Guillaume Thomas) #1


Since we’ve installed ROR on a cluster, we observed a strange behaviour when we fetch aliases.

In our setup, we have an alias foo and one and only one index behind (something like foo_f7a3549175bf47ca9558a055cbb2d2fb).

In one of our processes, we regularly check if there’s alias behind foo and create one if not. Since we’ve installed ROR, the call to /foo/_aliases has an erratic behaviour as it sometimes return nothing even there’s an index behind. I’m not able to reproduce it on demand and need to wait several calls before it happens.

Do you know any reason that could cause this?


(Simone Scarduzio) #2

@gtnx, is there anything particular being logged in ES when this happens?

By the way, the only index/alias logic we have in place in ROR takes place when the indices rule is triggered.

Not sure if this is feasible for you at the moment, but - as a test/workaround - If your alias poller/creator only has that job, you might want to create a block just for it such that does not trigger any indices rule.