1

Trying to set up an ELK stack on a google compute engine server I found out that the "click to deploy" solution google is providing is outdated.

None the less I tried installing Logstash and Kibana on top of the ES only to find out that I could not connect to Kibana (although I had no problem connecting to the elasticsearch itself both from inside google compute and the outside world).

Having the ELK stack already working on an AWS Ububnu 14.04 machine, I thought I'll try to start for a fresh machine and install everything myself on google compute.

After installing ES I again found that I could not connect to ES from outside (on the default port 9200, that of course worked with the click-to-deploy).

It's worth noting here that the click-to-deploy servers where probably intended for internal use, because they had ephemeral IPs, while I defined all my machines with static IPs (testing around this subject was a dead end as well).

In all cases the result for my servers was the same - the error I got was ERR_CONNECTION_REFUSED.

This (surprisingly) led me to think I was missing some firewall definition.

  1. I tried opening a different port and listening on it on the server side with netcat -lp <some-port>
  2. I then tried to run from my remote machine the same request I would run on elastic, but on the new port: curl -X GET 'http://<static-ip>:<some-port>'
  3. The request went on with no problems
  4. I stopped elasticsearch and started netcat -lp 9200

  5. curl -X GET 'http://<static-ip>:9200' from the remove machine came through like a charm.

    Even after opening the relevant ports for all machines and repeating the entire process (with all the different combinations) I realized I am probably missing something else.

Compering to the best of my ability with the google instance was yet another dead end.

What am I missing here? is there another connectivity definition in ES or GCE?

daTokenizer
  • 21
  • 5
  • can you add a screenshot of your firewall rules and your routes setup in compute engine? Mh but since you tested with curl and netcat already. Have you checked with `netstat -tulpen` whether your service is listening on the correct interface? It might listen on an unrouted interface (lo?) – konqi Nov 18 '15 at 11:40

1 Answers1

1

Well, konqi was sort of right. The problem was ES connecting to the loopback device, but setting network.bind_host to _non_loopback_ is only half of the solution, I can now acsess ES only form the outside (of course).

Apparently, there is a GCE plugin to fix this, but only up to ES version 1.7, guess I'll have to wait for it to catch up. The rest can be hacked around.

Community
  • 1
daTokenizer
  • 21
  • 5