You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Shu Zhang <sz...@mediosystems.com> on 2010/12/27 22:17:13 UTC
read repair across datacenters?
Hi, I'm pretty new to cassandra and read a couple of contradictory things on this topic. Does read repair get triggered across datacenters if you query with a consistency level of local_quorum?
Thanks,
Shu
Re: read repair across datacenters?
Posted by Jonathan Ellis <jb...@gmail.com>.
That is part of 982.
On Tue, Dec 28, 2010 at 12:46 PM, Shu Zhang <sz...@mediosystems.com> wrote:
> I see, thank you. So that feature is coming. Any idea if the same will be true for local_quorum? That is, if a read repair will happen in the background for all nodes including nodes in other datacenters?
> ________________________________________
> From: Jonathan Ellis [jbellis@gmail.com]
> Sent: Monday, December 27, 2010 6:59 PM
> To: user
> Subject: Re: read repair across datacenters?
>
> https://issues.apache.org/jira/browse/CASSANDRA-982
>
> On Mon, Dec 27, 2010 at 5:55 PM, Shu Zhang <sz...@mediosystems.com> wrote:
>> Brandon, for a read with quorum CL, a response is returned to the client after half (rounded up) the replica nodes responds. Do you mean that those half are RR'ed before the response is returned to the client and the other half RR'ed asynchronously? Or is it that the other half of replicas are not RR'ed?
>>
>> ________________________________________
>> From: Brandon Williams [driftx@gmail.com]
>> Sent: Monday, December 27, 2010 3:00 PM
>> To: user@cassandra.apache.org
>> Subject: Re: read repair across datacenters?
>>
>> On Mon, Dec 27, 2010 at 4:44 PM, Narendra Sharma <na...@gmail.com>> wrote:
>> The comment in the cassandra.yaml says:
>> "specifies the probability with which read repairs should be invoked on non-quorum reads"
>>
>> Does this mean RR chance is applicable only for non-quorum reads?
>>
>> Yes, because on quorum or greater the RR is performed before the response is returned.
>>
>> Another question on same topic:
>> Will RR use one of the node in the other datacenter as coordinator instead of sending the data to each replica (assuming they have old data)? This is to reduce the amount of network traffic between datacenters.
>>
>> https://issues.apache.org/jira/browse/CASSANDRA-1530
>>
>> <https://issues.apache.org/jira/browse/CASSANDRA-1530>-Brandon
>>
>
>
>
> --
> Jonathan Ellis
> Project Chair, Apache Cassandra
> co-founder of Riptano, the source for professional Cassandra support
> http://riptano.com
>
--
Jonathan Ellis
Project Chair, Apache Cassandra
co-founder of Riptano, the source for professional Cassandra support
http://riptano.com
RE: read repair across datacenters?
Posted by Shu Zhang <sz...@mediosystems.com>.
I see, thank you. So that feature is coming. Any idea if the same will be true for local_quorum? That is, if a read repair will happen in the background for all nodes including nodes in other datacenters?
________________________________________
From: Jonathan Ellis [jbellis@gmail.com]
Sent: Monday, December 27, 2010 6:59 PM
To: user
Subject: Re: read repair across datacenters?
https://issues.apache.org/jira/browse/CASSANDRA-982
On Mon, Dec 27, 2010 at 5:55 PM, Shu Zhang <sz...@mediosystems.com> wrote:
> Brandon, for a read with quorum CL, a response is returned to the client after half (rounded up) the replica nodes responds. Do you mean that those half are RR'ed before the response is returned to the client and the other half RR'ed asynchronously? Or is it that the other half of replicas are not RR'ed?
>
> ________________________________________
> From: Brandon Williams [driftx@gmail.com]
> Sent: Monday, December 27, 2010 3:00 PM
> To: user@cassandra.apache.org
> Subject: Re: read repair across datacenters?
>
> On Mon, Dec 27, 2010 at 4:44 PM, Narendra Sharma <na...@gmail.com>> wrote:
> The comment in the cassandra.yaml says:
> "specifies the probability with which read repairs should be invoked on non-quorum reads"
>
> Does this mean RR chance is applicable only for non-quorum reads?
>
> Yes, because on quorum or greater the RR is performed before the response is returned.
>
> Another question on same topic:
> Will RR use one of the node in the other datacenter as coordinator instead of sending the data to each replica (assuming they have old data)? This is to reduce the amount of network traffic between datacenters.
>
> https://issues.apache.org/jira/browse/CASSANDRA-1530
>
> <https://issues.apache.org/jira/browse/CASSANDRA-1530>-Brandon
>
--
Jonathan Ellis
Project Chair, Apache Cassandra
co-founder of Riptano, the source for professional Cassandra support
http://riptano.com
Re: read repair across datacenters?
Posted by Jonathan Ellis <jb...@gmail.com>.
https://issues.apache.org/jira/browse/CASSANDRA-982
On Mon, Dec 27, 2010 at 5:55 PM, Shu Zhang <sz...@mediosystems.com> wrote:
> Brandon, for a read with quorum CL, a response is returned to the client after half (rounded up) the replica nodes responds. Do you mean that those half are RR'ed before the response is returned to the client and the other half RR'ed asynchronously? Or is it that the other half of replicas are not RR'ed?
>
> ________________________________________
> From: Brandon Williams [driftx@gmail.com]
> Sent: Monday, December 27, 2010 3:00 PM
> To: user@cassandra.apache.org
> Subject: Re: read repair across datacenters?
>
> On Mon, Dec 27, 2010 at 4:44 PM, Narendra Sharma <na...@gmail.com>> wrote:
> The comment in the cassandra.yaml says:
> "specifies the probability with which read repairs should be invoked on non-quorum reads"
>
> Does this mean RR chance is applicable only for non-quorum reads?
>
> Yes, because on quorum or greater the RR is performed before the response is returned.
>
> Another question on same topic:
> Will RR use one of the node in the other datacenter as coordinator instead of sending the data to each replica (assuming they have old data)? This is to reduce the amount of network traffic between datacenters.
>
> https://issues.apache.org/jira/browse/CASSANDRA-1530
>
> <https://issues.apache.org/jira/browse/CASSANDRA-1530>-Brandon
>
--
Jonathan Ellis
Project Chair, Apache Cassandra
co-founder of Riptano, the source for professional Cassandra support
http://riptano.com
RE: read repair across datacenters?
Posted by Shu Zhang <sz...@mediosystems.com>.
Brandon, for a read with quorum CL, a response is returned to the client after half (rounded up) the replica nodes responds. Do you mean that those half are RR'ed before the response is returned to the client and the other half RR'ed asynchronously? Or is it that the other half of replicas are not RR'ed?
________________________________________
From: Brandon Williams [driftx@gmail.com]
Sent: Monday, December 27, 2010 3:00 PM
To: user@cassandra.apache.org
Subject: Re: read repair across datacenters?
On Mon, Dec 27, 2010 at 4:44 PM, Narendra Sharma <na...@gmail.com>> wrote:
The comment in the cassandra.yaml says:
"specifies the probability with which read repairs should be invoked on non-quorum reads"
Does this mean RR chance is applicable only for non-quorum reads?
Yes, because on quorum or greater the RR is performed before the response is returned.
Another question on same topic:
Will RR use one of the node in the other datacenter as coordinator instead of sending the data to each replica (assuming they have old data)? This is to reduce the amount of network traffic between datacenters.
https://issues.apache.org/jira/browse/CASSANDRA-1530
<https://issues.apache.org/jira/browse/CASSANDRA-1530>-Brandon
Re: read repair across datacenters?
Posted by Brandon Williams <dr...@gmail.com>.
On Mon, Dec 27, 2010 at 4:44 PM, Narendra Sharma
<na...@gmail.com>wrote:
> The comment in the cassandra.yaml says:
> "specifies the probability with which read repairs should be invoked on *
> non-quorum* reads"
>
> Does this mean RR chance is applicable only for non-quorum reads?
>
Yes, because on quorum or greater the RR is performed before the response is
returned.
Another question on same topic:
> Will RR use one of the node in the other datacenter as coordinator instead
> of sending the data to each replica (assuming they have old data)? This is
> to reduce the amount of network traffic between datacenters.
>
https://issues.apache.org/jira/browse/CASSANDRA-1530
<https://issues.apache.org/jira/browse/CASSANDRA-1530>-Brandon
Re: read repair across datacenters?
Posted by Narendra Sharma <na...@gmail.com>.
The comment in the cassandra.yaml says:
"specifies the probability with which read repairs should be invoked on *
non-quorum* reads"
Does this mean RR chance is applicable only for non-quorum reads?
Another question on same topic:
Will RR use one of the node in the other datacenter as coordinator instead
of sending the data to each replica (assuming they have old data)? This is
to reduce the amount of network traffic between datacenters.
-Naren
On Mon, Dec 27, 2010 at 1:26 PM, Brandon Williams <dr...@gmail.com> wrote:
> On Mon, Dec 27, 2010 at 3:17 PM, Shu Zhang <sz...@mediosystems.com>wrote:
>
>> Hi, I'm pretty new to cassandra and read a couple of contradictory things
>> on this topic. Does read repair get triggered across datacenters if you
>> query with a consistency level of local_quorum?
>>
>
> If the RR chance is 100% (default), it's triggered regardless of CL, which
> only controls how synchronously the repair is performed.
>
> -Brandon
>
Re: read repair across datacenters?
Posted by Brandon Williams <dr...@gmail.com>.
On Mon, Dec 27, 2010 at 3:17 PM, Shu Zhang <sz...@mediosystems.com> wrote:
> Hi, I'm pretty new to cassandra and read a couple of contradictory things
> on this topic. Does read repair get triggered across datacenters if you
> query with a consistency level of local_quorum?
>
If the RR chance is 100% (default), it's triggered regardless of CL, which
only controls how synchronously the repair is performed.
-Brandon