Piavlo | 16 May 12:14 2012
Picon

is it possible to run cassandra process in client mode as smart proxy


  Hi,

I'm interested in using some smart proxy cassandra process that could 
act as coordinator node and be aware of cluster state.
And run this smart proxy cassandra process on each client side host  
where the application(php) with short lived cassandra connections runs.
Besides being aware of cluster state if it could act as coordinator node 
it would save unneeded network trips.
And maybe even have an option to take care of hinted handoffs.
IMHO the best candidate for this is the cassandra itself (like it's done 
in elasticsearch 
http://www.elasticsearch.org/guide/reference/modules/node.html)
I also see there was a work done in this direction at 
https://issues.apache.org/jira/browse/CASSANDRA-535
So maybe this is something that is already usable?

Or maybe there is some third party project that could be used as smart 
cassandra proxy?

Thanks
Alex

R. Verlangen | 16 May 12:24 2012
Picon

Re: is it possible to run cassandra process in client mode as smart proxy

Hi there,


I'm using HAProxy for PHP projects to take care of this. It improved connection pooling enormous on the client side: with preserving failover capabilities. Maybe that is something for you to use in combination with PHP.

Good luck!

2012/5/16 Piavlo <lolitushka <at> gmail.com>

 Hi,

I'm interested in using some smart proxy cassandra process that could act as coordinator node and be aware of cluster state.
And run this smart proxy cassandra process on each client side host  where the application(php) with short lived cassandra connections runs.
Besides being aware of cluster state if it could act as coordinator node it would save unneeded network trips.
And maybe even have an option to take care of hinted handoffs.
IMHO the best candidate for this is the cassandra itself (like it's done in elasticsearch http://www.elasticsearch.org/guide/reference/modules/node.html)
I also see there was a work done in this direction at https://issues.apache.org/jira/browse/CASSANDRA-535
So maybe this is something that is already usable?

Or maybe there is some third party project that could be used as smart cassandra proxy?

Thanks
Alex



--
With kind regards,

Robin Verlangen

Piavlo | 16 May 13:01 2012
Picon

Re: is it possible to run cassandra process in client mode as smart proxy

On 05/16/2012 01:24 PM, R. Verlangen wrote:
Hi there,

I'm using HAProxy for PHP projects to take care of this. It improved connection pooling enormous on the client side: with preserving failover capabilities. Maybe that is something for you to use in combination with PHP.
I already use it exactly like this :)
But i don't think it's a good solution. And it's totally unaware of thrift/cassandra protocol, it's was pretty well discussed here
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Cassandra-amp-HAProxy-td5473144.html
I even see the plain tcp healchecks failing from time to time for no reason.
I'm planning to make it a bit smarter with localhost http level healthchecks - where it would make a cassandra write to CF in
 keyspace wich has replication of 1, and will write to a key that maps to the specific cassandra node being checked by healthcheck (of course  the keys need to be recaclulated each time the cluster is rebalanced).
But IMHO it's very very ugly hack, and not as reliable as real smart proxy which is way more superior and efficient (especially if it could do the reads/writes coordination itself).

Haproxy also has an issue that then once of the backend ip changes (which happens often in clouds) it's has to be restarted to resolve the the correct hostname,
though it looks like Willy is finally seriously considering to implement more dynamic for hostnames  lookups (which was not the case about a year ago then I asked  for such feature)
the problem was is discussed here recently - http://marc.info/?l=haproxy&m=133559164408814&w=1

haproxy has some more issues - i don't remember off top of my head.

Smart proxy would simply not have all those issues, as it's aware of the ring state and the protocol and if smaprt proxy was  the cassandra itself then it would have all the needed features tested and reliable at no effort.

Thanks
Alex

Good luck!

2012/5/16 Piavlo <lolitushka <at> gmail.com>

 Hi,

I'm interested in using some smart proxy cassandra process that could act as coordinator node and be aware of cluster state.
And run this smart proxy cassandra process on each client side host  where the application(php) with short lived cassandra connections runs.
Besides being aware of cluster state if it could act as coordinator node it would save unneeded network trips.
And maybe even have an option to take care of hinted handoffs.
IMHO the best candidate for this is the cassandra itself (like it's done in elasticsearch http://www.elasticsearch.org/guide/reference/modules/node.html)
I also see there was a work done in this direction at https://issues.apache.org/jira/browse/CASSANDRA-535
So maybe this is something that is already usable?

Or maybe there is some third party project that could be used as smart cassandra proxy?

Thanks
Alex



--
With kind regards,

Robin Verlangen


R. Verlangen | 16 May 13:09 2012
Picon

Re: is it possible to run cassandra process in client mode as smart proxy

Yes, I'm aware of those issues however in our use case they don't cause any problems. 


But ... If there's something better out there I'm really curious: so I'll keep up with this thread.

2012/5/16 Piavlo <lolitushka <at> gmail.com>
On 05/16/2012 01:24 PM, R. Verlangen wrote:
Hi there,

I'm using HAProxy for PHP projects to take care of this. It improved connection pooling enormous on the client side: with preserving failover capabilities. Maybe that is something for you to use in combination with PHP.
I already use it exactly like this :)
But i don't think it's a good solution. And it's totally unaware of thrift/cassandra protocol, it's was pretty well discussed here
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Cassandra-amp-HAProxy-td5473144.html
I even see the plain tcp healchecks failing from time to time for no reason.
I'm planning to make it a bit smarter with localhost http level healthchecks - where it would make a cassandra write to CF in
 keyspace wich has replication of 1, and will write to a key that maps to the specific cassandra node being checked by healthcheck (of course  the keys need to be recaclulated each time the cluster is rebalanced).
But IMHO it's very very ugly hack, and not as reliable as real smart proxy which is way more superior and efficient (especially if it could do the reads/writes coordination itself).

Haproxy also has an issue that then once of the backend ip changes (which happens often in clouds) it's has to be restarted to resolve the the correct hostname,
though it looks like Willy is finally seriously considering to implement more dynamic for hostnames  lookups (which was not the case about a year ago then I asked  for such feature)
the problem was is discussed here recently - http://marc.info/?l=haproxy&m=133559164408814&w=1

haproxy has some more issues - i don't remember off top of my head.

Smart proxy would simply not have all those issues, as it's aware of the ring state and the protocol and if smaprt proxy was  the cassandra itself then it would have all the needed features tested and reliable at no effort.

Thanks
Alex


Good luck!

2012/5/16 Piavlo <lolitushka <at> gmail.com>

 Hi,

I'm interested in using some smart proxy cassandra process that could act as coordinator node and be aware of cluster state.
And run this smart proxy cassandra process on each client side host  where the application(php) with short lived cassandra connections runs.
Besides being aware of cluster state if it could act as coordinator node it would save unneeded network trips.
And maybe even have an option to take care of hinted handoffs.
IMHO the best candidate for this is the cassandra itself (like it's done in elasticsearch http://www.elasticsearch.org/guide/reference/modules/node.html)
I also see there was a work done in this direction at https://issues.apache.org/jira/browse/CASSANDRA-535
So maybe this is something that is already usable?

Or maybe there is some third party project that could be used as smart cassandra proxy?

Thanks
Alex



--
With kind regards,

Robin Verlangen





--
With kind regards,

Robin Verlangen


Gmane