Cross Data Center Replication about crate

[mfussenegger]

What you're trying to do is possible using the shard allocation awareness feature. It's not documented in our docs yet, but you can take a look at the elasticsearch documentation here http://www.elasticsearch.org/guide/en/elasticsearch/reference/current/modules-cluster.html#allocation-awareness

We've it on our roadmap to work on cross-dc stuff. So it should pop up in our own docs eventually.

Hope that helps.

Update:

Just as an addition in case you don't know Elasticsearch. Crate uses Elasticsearch as a library and is responsible for various tasks. That also involves some parts of the shard/replication logic which is why the settings that are documented on the site I linked should work if you apply them in the crate.yml file.

[michabbb]

hi there, for people who have never worked with elasticsearch its not that easy, today i setup my first crate server, only one node... seems to work, i still habe problems[https://github.com/crate/crate-admin/issues/86] but anyway... could someone please provide a simple info how the replication has to be setup between 2 diffrent locations ? its not the part we are too lazy.. hours went away to setup all this... so please have some mercy and help me/us here with a simple config example, that would be great! thanks a lot....

[seut]

michabb: do you just having problems that your nodes won't join the same cluster? than maybe you need to configure unicast instead of the multicast (default) on each node, this blog post should help: https://crate.io/blog/using-crate-in-multinode-setup/
otherwise please give us some more hints about your wanted setup.

[michabbb]

private: hi sebastian (übrigens, wir sprachen auf der IPC14, ich hatte die idee/vorschlag mit dem mysql-plugin, vielleicht erinnerst du dich ? ;) ich hab also gleich mal euer crate getestet, aber derzeit nur eine node... eine zweite hatte ich noch nicht probiert, weil ich nirgendwo was in der doku dazu gefunden hatte, daher: vielen dank für den blog-link, ich werde es testen ;-)

public: thanks for the blog link, i will try ;)

[ahto]

There is some wording in elasticsearch docs that say:

"Avoid clusters that span multiple data-centers, even if the data-centers are colocated in close proximity. Definitely avoid clusters that span large geographic distances."

Found at: http://www.elasticsearch.org/guide/en/elasticsearch/guide/current/hardware.html#_network

It is pretty clear wording, do you know why they say this?

[mfussenegger]

Yes the reason they say this is the increased latency:

Elasticsearch clusters assume that all nodes are equal…not that half the nodes are actually 150ms distant in another datacenter. Larger latencies tend to exacerbate problems in distributed systems and make debugging and resolution more difficult.

This of course also applies to Crate.

[michabbb]

btw... i have a 2 node cluster right now, but everybody trying this should take care.. the default port for node 2 node communication is 4300, not 4200 ... this simple fact cost me half an hour g

[dobe]

@michabbb
you are right, we should document the ports more prominent and clear. those ports are currently only covered in the configuration documentation found at https://crate.io/docs/stable/configuration.html?#ports

we will update the documentation

[dobe]

there are already users using the routing allocation https://crate.io/docs/stable/configuration.html?#routing-allocation to implement a multi zone deployment on AWS.

it is possible with the forced allocation awareness to ensure that queries only touch nodes in the same zone. and also to make sure that replicas are distributed accross such zones.

we will create a document explaining this explicitly and update this issue when done.

[chaudum]

@weswam @ahto @michabbb
we've added a best practice document about multi zone setup / cross datacenter replication to our documentation.
you can find it here https://crate.io/docs/en/latest/best_practice/multi_zone_setup.html

[michabbb]

thanks

[mfussenegger]

Guess this can be closed since it is now documented properly.

[sahil98541]

How to do this across 2 data centres for version 1.0.2?
I have a lab instance consisting 6 nodes across 2 zones as described in
https://crate.io/docs/crate/guide/en/latest/scaling/multi-zone-setup.html

When i start replication (number of replication =3 ) shard suppose 100 should start 400 shards .
But replication isnt working only 25% replicated data (i.e original copy with 100 shrads running)

Please advice