Re: Master /Slave Set up on AWS - 3.6.1
Hi Sujatha, If I understand correctly, you will have only 1 slave (and 1 master), so that's not really a HA architecture. You could manually turn master into slave, but that's going to mean some down time... Otis -- Solr ElasticSearch Support http://sematext.com/ On Tue, Mar 5, 2013 at 3:05 AM, Sujatha Arun suja.a...@gmail.com wrote: Hi, We are planning to set up *2* *High-Memory Quadruple Extra Large Instance *as master and slave for our multicore solr setup which has more than 200 cores spread between a couple of webapps on a single JVM on *AWS* All indexing [via a queue will go to master ] . One Slave Server will replicate all the core level indexes from the master , slave Configurations are defined in the solr.xml at the webapp level with a different poll interval for each webapp. We are planning to LB the search requests by fronting the master and slave with an *AWS ELB *. The master configuration will not enable the slave properties as master is not replicating from any other machine. The master and slave have similar hardware configurations [*High-Memory Quadruple Extra Large Instance] .*This is mainly for HA if the slave goes down. * * Any issue with the above set up ,please advice. Regards, Sujatha * *
Re: Master /Slave Set up on AWS - 3.6.1
Hi Otis, Since currently we are planning for only one slave due to cost considerations, can we have an ELB fronting the master and slave for HA. 1. All index requests will go to the master . 2. Slave replicates from master . 3. Search request can go either to master /slave via ELB. is that resonable HA for search ? Regards Sujatha On Tue, Mar 5, 2013 at 5:12 PM, Otis Gospodnetic otis.gospodne...@gmail.com wrote: Hi Sujatha, If I understand correctly, you will have only 1 slave (and 1 master), so that's not really a HA architecture. You could manually turn master into slave, but that's going to mean some down time... Otis -- Solr ElasticSearch Support http://sematext.com/ On Tue, Mar 5, 2013 at 3:05 AM, Sujatha Arun suja.a...@gmail.com wrote: Hi, We are planning to set up *2* *High-Memory Quadruple Extra Large Instance *as master and slave for our multicore solr setup which has more than 200 cores spread between a couple of webapps on a single JVM on *AWS* All indexing [via a queue will go to master ] . One Slave Server will replicate all the core level indexes from the master , slave Configurations are defined in the solr.xml at the webapp level with a different poll interval for each webapp. We are planning to LB the search requests by fronting the master and slave with an *AWS ELB *. The master configuration will not enable the slave properties as master is not replicating from any other machine. The master and slave have similar hardware configurations [*High-Memory Quadruple Extra Large Instance] .*This is mainly for HA if the slave goes down. * * Any issue with the above set up ,please advice. Regards, Sujatha * *
Re: Master /Slave Set up on AWS - 3.6.1
Is there anything wrong with set up? On Tue, Mar 5, 2013 at 5:43 PM, Sujatha Arun suja.a...@gmail.com wrote: Hi Otis, Since currently we are planning for only one slave due to cost considerations, can we have an ELB fronting the master and slave for HA. 1. All index requests will go to the master . 2. Slave replicates from master . 3. Search request can go either to master /slave via ELB. is that resonable HA for search ? Regards Sujatha On Tue, Mar 5, 2013 at 5:12 PM, Otis Gospodnetic otis.gospodne...@gmail.com wrote: Hi Sujatha, If I understand correctly, you will have only 1 slave (and 1 master), so that's not really a HA architecture. You could manually turn master into slave, but that's going to mean some down time... Otis -- Solr ElasticSearch Support http://sematext.com/ On Tue, Mar 5, 2013 at 3:05 AM, Sujatha Arun suja.a...@gmail.com wrote: Hi, We are planning to set up *2* *High-Memory Quadruple Extra Large Instance *as master and slave for our multicore solr setup which has more than 200 cores spread between a couple of webapps on a single JVM on *AWS* All indexing [via a queue will go to master ] . One Slave Server will replicate all the core level indexes from the master , slave Configurations are defined in the solr.xml at the webapp level with a different poll interval for each webapp. We are planning to LB the search requests by fronting the master and slave with an *AWS ELB *. The master configuration will not enable the slave properties as master is not replicating from any other machine. The master and slave have similar hardware configurations [*High-Memory Quadruple Extra Large Instance] .*This is mainly for HA if the slave goes down. * * Any issue with the above set up ,please advice. Regards, Sujatha * *
Re: Master /Slave Set up on AWS - 3.6.1
Hi, That my be fine. I'd use the sticky session setting in ELB to avoid having the same user's query hit both master and slave, say while paging through results, and risking seeing inconsistent results. THis will also help with cache utilization. This said, this is not a recommended setup. Note that AWS *just* lowered the prices for Reserved Instances today. Otis -- Solr ElasticSearch Support http://sematext.com/ On Tue, Mar 5, 2013 at 7:13 AM, Sujatha Arun suja.a...@gmail.com wrote: Hi Otis, Since currently we are planning for only one slave due to cost considerations, can we have an ELB fronting the master and slave for HA. 1. All index requests will go to the master . 2. Slave replicates from master . 3. Search request can go either to master /slave via ELB. is that resonable HA for search ? Regards Sujatha On Tue, Mar 5, 2013 at 5:12 PM, Otis Gospodnetic otis.gospodne...@gmail.com wrote: Hi Sujatha, If I understand correctly, you will have only 1 slave (and 1 master), so that's not really a HA architecture. You could manually turn master into slave, but that's going to mean some down time... Otis -- Solr ElasticSearch Support http://sematext.com/ On Tue, Mar 5, 2013 at 3:05 AM, Sujatha Arun suja.a...@gmail.com wrote: Hi, We are planning to set up *2* *High-Memory Quadruple Extra Large Instance *as master and slave for our multicore solr setup which has more than 200 cores spread between a couple of webapps on a single JVM on *AWS* All indexing [via a queue will go to master ] . One Slave Server will replicate all the core level indexes from the master , slave Configurations are defined in the solr.xml at the webapp level with a different poll interval for each webapp. We are planning to LB the search requests by fronting the master and slave with an *AWS ELB *. The master configuration will not enable the slave properties as master is not replicating from any other machine. The master and slave have similar hardware configurations [*High-Memory Quadruple Extra Large Instance] .*This is mainly for HA if the slave goes down. * * Any issue with the above set up ,please advice. Regards, Sujatha * *
Re: Master /Slave Set up on AWS - 3.6.1
If your index is on EBS, you'll see big iowait percentages when merges happen. I'm not sure what that's going to do to your master's ability to service requests. You should test. Alternatively, you might figure out the size of machine you need to index vs. the size of machine you need to service queries. They're very likely not the same, in which case, that may afford you the ability to have 2 slaves and 1 master in a similar budget. Also, once you've settled on an infrastructure, you should investigate buying reserved instances for a year. It will greatly reduce your costs. Michael Della Bitta Appinions 18 East 41st Street, 2nd Floor New York, NY 10017-6271 www.appinions.com Where Influence Isn’t a Game On Tue, Mar 5, 2013 at 8:59 AM, Sujatha Arun suja.a...@gmail.com wrote: Is there anything wrong with set up? On Tue, Mar 5, 2013 at 5:43 PM, Sujatha Arun suja.a...@gmail.com wrote: Hi Otis, Since currently we are planning for only one slave due to cost considerations, can we have an ELB fronting the master and slave for HA. 1. All index requests will go to the master . 2. Slave replicates from master . 3. Search request can go either to master /slave via ELB. is that resonable HA for search ? Regards Sujatha On Tue, Mar 5, 2013 at 5:12 PM, Otis Gospodnetic otis.gospodne...@gmail.com wrote: Hi Sujatha, If I understand correctly, you will have only 1 slave (and 1 master), so that's not really a HA architecture. You could manually turn master into slave, but that's going to mean some down time... Otis -- Solr ElasticSearch Support http://sematext.com/ On Tue, Mar 5, 2013 at 3:05 AM, Sujatha Arun suja.a...@gmail.com wrote: Hi, We are planning to set up *2* *High-Memory Quadruple Extra Large Instance *as master and slave for our multicore solr setup which has more than 200 cores spread between a couple of webapps on a single JVM on *AWS* All indexing [via a queue will go to master ] . One Slave Server will replicate all the core level indexes from the master , slave Configurations are defined in the solr.xml at the webapp level with a different poll interval for each webapp. We are planning to LB the search requests by fronting the master and slave with an *AWS ELB *. The master configuration will not enable the slave properties as master is not replicating from any other machine. The master and slave have similar hardware configurations [*High-Memory Quadruple Extra Large Instance] .*This is mainly for HA if the slave goes down. * * Any issue with the above set up ,please advice. Regards, Sujatha * *
Re: Master /Slave Set up on AWS - 3.6.1
Hi Otis,Michael, Thanks for your input and suggestions . Yes, we were considering the sticky session for pagination and we are not planning for having index on EBS I would like to understand why its not the recommended approach, can you please explain? Till now we were having a single server for both Indexing and search ,though this was on dedicated server and not on cloud. Indexing would happen sequentially via a queue due to which commits would happen for only one or at most 2 cores simultaneously. With Master /Slave approach I see that when a slave replicates form master based on the poll time which I have defined in solr.xml file depending on the number of cores/webapp ,the replication and hence commit is going to happen simultaneously for many cores increasing load on slave server. By having 2 slaves and one master - whenever we create a core,which happens quite frequently we need to create this on 3 servers instead of two , which has to done manually by running a script on each server.We have requirement for adding cores per each customer. I do understand that hardware requirements for master can be quite different [Lower memory /higher CPU/Cache setting in config /autowarming etc ] from slave.But given that we will be Indexing sequentially and having the same configuration in terms of memory and CPU/cache for both master and slave,would this be a reasonable approach? ThanksRegards, Sujatha On Tue, Mar 5, 2013 at 9:10 PM, Michael Della Bitta michael.della.bi...@appinions.com wrote: If your index is on EBS, you'll see big iowait percentages when merges happen. I'm not sure what that's going to do to your master's ability to service requests. You should test. Alternatively, you might figure out the size of machine you need to index vs. the size of machine you need to service queries. They're very likely not the same, in which case, that may afford you the ability to have 2 slaves and 1 master in a similar budget. Also, once you've settled on an infrastructure, you should investigate buying reserved instances for a year. It will greatly reduce your costs. Michael Della Bitta Appinions 18 East 41st Street, 2nd Floor New York, NY 10017-6271 www.appinions.com Where Influence Isn’t a Game On Tue, Mar 5, 2013 at 8:59 AM, Sujatha Arun suja.a...@gmail.com wrote: Is there anything wrong with set up? On Tue, Mar 5, 2013 at 5:43 PM, Sujatha Arun suja.a...@gmail.com wrote: Hi Otis, Since currently we are planning for only one slave due to cost considerations, can we have an ELB fronting the master and slave for HA. 1. All index requests will go to the master . 2. Slave replicates from master . 3. Search request can go either to master /slave via ELB. is that resonable HA for search ? Regards Sujatha On Tue, Mar 5, 2013 at 5:12 PM, Otis Gospodnetic otis.gospodne...@gmail.com wrote: Hi Sujatha, If I understand correctly, you will have only 1 slave (and 1 master), so that's not really a HA architecture. You could manually turn master into slave, but that's going to mean some down time... Otis -- Solr ElasticSearch Support http://sematext.com/ On Tue, Mar 5, 2013 at 3:05 AM, Sujatha Arun suja.a...@gmail.com wrote: Hi, We are planning to set up *2* *High-Memory Quadruple Extra Large Instance *as master and slave for our multicore solr setup which has more than 200 cores spread between a couple of webapps on a single JVM on *AWS* All indexing [via a queue will go to master ] . One Slave Server will replicate all the core level indexes from the master , slave Configurations are defined in the solr.xml at the webapp level with a different poll interval for each webapp. We are planning to LB the search requests by fronting the master and slave with an *AWS ELB *. The master configuration will not enable the slave properties as master is not replicating from any other machine. The master and slave have similar hardware configurations [*High-Memory Quadruple Extra Large Instance] .*This is mainly for HA if the slave goes down. * * Any issue with the above set up ,please advice. Regards, Sujatha * *
Re: Master /Slave Set up on AWS - 3.6.1
Hello, This is not recommended because people typically don't want the load from indexing affect queries/user experience. If your numbers are low, then this may not be a big deal. If you already need to create a core on 2 machines, creating it on 3 doesn't seem a big deal. There is a slight conflict in what you wrote, which is that cores will be created frequently, which implies you will quickly have lots of cores vs. desire to have minimal hardware. If it were up to me, I'd consider a weaker/cheaper master, and more slaves. Otis -- Solr ElasticSearch Support http://sematext.com/ On Tue, Mar 5, 2013 at 12:46 PM, Sujatha Arun suja.a...@gmail.com wrote: Hi Otis,Michael, Thanks for your input and suggestions . Yes, we were considering the sticky session for pagination and we are not planning for having index on EBS I would like to understand why its not the recommended approach, can you please explain? Till now we were having a single server for both Indexing and search ,though this was on dedicated server and not on cloud. Indexing would happen sequentially via a queue due to which commits would happen for only one or at most 2 cores simultaneously. With Master /Slave approach I see that when a slave replicates form master based on the poll time which I have defined in solr.xml file depending on the number of cores/webapp ,the replication and hence commit is going to happen simultaneously for many cores increasing load on slave server. By having 2 slaves and one master - whenever we create a core,which happens quite frequently we need to create this on 3 servers instead of two , which has to done manually by running a script on each server.We have requirement for adding cores per each customer. I do understand that hardware requirements for master can be quite different [Lower memory /higher CPU/Cache setting in config /autowarming etc ] from slave.But given that we will be Indexing sequentially and having the same configuration in terms of memory and CPU/cache for both master and slave,would this be a reasonable approach? ThanksRegards, Sujatha On Tue, Mar 5, 2013 at 9:10 PM, Michael Della Bitta michael.della.bi...@appinions.com wrote: If your index is on EBS, you'll see big iowait percentages when merges happen. I'm not sure what that's going to do to your master's ability to service requests. You should test. Alternatively, you might figure out the size of machine you need to index vs. the size of machine you need to service queries. They're very likely not the same, in which case, that may afford you the ability to have 2 slaves and 1 master in a similar budget. Also, once you've settled on an infrastructure, you should investigate buying reserved instances for a year. It will greatly reduce your costs. Michael Della Bitta Appinions 18 East 41st Street, 2nd Floor New York, NY 10017-6271 www.appinions.com Where Influence Isn’t a Game On Tue, Mar 5, 2013 at 8:59 AM, Sujatha Arun suja.a...@gmail.com wrote: Is there anything wrong with set up? On Tue, Mar 5, 2013 at 5:43 PM, Sujatha Arun suja.a...@gmail.com wrote: Hi Otis, Since currently we are planning for only one slave due to cost considerations, can we have an ELB fronting the master and slave for HA. 1. All index requests will go to the master . 2. Slave replicates from master . 3. Search request can go either to master /slave via ELB. is that resonable HA for search ? Regards Sujatha On Tue, Mar 5, 2013 at 5:12 PM, Otis Gospodnetic otis.gospodne...@gmail.com wrote: Hi Sujatha, If I understand correctly, you will have only 1 slave (and 1 master), so that's not really a HA architecture. You could manually turn master into slave, but that's going to mean some down time... Otis -- Solr ElasticSearch Support http://sematext.com/ On Tue, Mar 5, 2013 at 3:05 AM, Sujatha Arun suja.a...@gmail.com wrote: Hi, We are planning to set up *2* *High-Memory Quadruple Extra Large Instance *as master and slave for our multicore solr setup which has more than 200 cores spread between a couple of webapps on a single JVM on *AWS* All indexing [via a queue will go to master ] . One Slave Server will replicate all the core level indexes from the master , slave Configurations are defined in the solr.xml at the webapp level with a different poll interval for each webapp. We are planning to LB the search requests by fronting the master and slave with an *AWS ELB *. The master configuration will not enable the slave properties as master is not replicating from any other machine. The master and slave have similar hardware configurations [*High-Memory Quadruple Extra Large
Re: Master /Slave Set up on AWS - 3.6.1
Thanks Otis . Yes true but considering that the Indexing is via a queue, there would actually be minimal load on the machine. And we are planning to replicate this set up via adding more machines when the server reaches about 80% capacity for adding more cores. Regards, Sujatha On Wed, Mar 6, 2013 at 9:26 AM, Otis Gospodnetic otis.gospodne...@gmail.com wrote: Hello, This is not recommended because people typically don't want the load from indexing affect queries/user experience. If your numbers are low, then this may not be a big deal. If you already need to create a core on 2 machines, creating it on 3 doesn't seem a big deal. There is a slight conflict in what you wrote, which is that cores will be created frequently, which implies you will quickly have lots of cores vs. desire to have minimal hardware. If it were up to me, I'd consider a weaker/cheaper master, and more slaves. Otis -- Solr ElasticSearch Support http://sematext.com/ On Tue, Mar 5, 2013 at 12:46 PM, Sujatha Arun suja.a...@gmail.com wrote: Hi Otis,Michael, Thanks for your input and suggestions . Yes, we were considering the sticky session for pagination and we are not planning for having index on EBS I would like to understand why its not the recommended approach, can you please explain? Till now we were having a single server for both Indexing and search ,though this was on dedicated server and not on cloud. Indexing would happen sequentially via a queue due to which commits would happen for only one or at most 2 cores simultaneously. With Master /Slave approach I see that when a slave replicates form master based on the poll time which I have defined in solr.xml file depending on the number of cores/webapp ,the replication and hence commit is going to happen simultaneously for many cores increasing load on slave server. By having 2 slaves and one master - whenever we create a core,which happens quite frequently we need to create this on 3 servers instead of two , which has to done manually by running a script on each server.We have requirement for adding cores per each customer. I do understand that hardware requirements for master can be quite different [Lower memory /higher CPU/Cache setting in config /autowarming etc ] from slave.But given that we will be Indexing sequentially and having the same configuration in terms of memory and CPU/cache for both master and slave,would this be a reasonable approach? ThanksRegards, Sujatha On Tue, Mar 5, 2013 at 9:10 PM, Michael Della Bitta michael.della.bi...@appinions.com wrote: If your index is on EBS, you'll see big iowait percentages when merges happen. I'm not sure what that's going to do to your master's ability to service requests. You should test. Alternatively, you might figure out the size of machine you need to index vs. the size of machine you need to service queries. They're very likely not the same, in which case, that may afford you the ability to have 2 slaves and 1 master in a similar budget. Also, once you've settled on an infrastructure, you should investigate buying reserved instances for a year. It will greatly reduce your costs. Michael Della Bitta Appinions 18 East 41st Street, 2nd Floor New York, NY 10017-6271 www.appinions.com Where Influence Isn’t a Game On Tue, Mar 5, 2013 at 8:59 AM, Sujatha Arun suja.a...@gmail.com wrote: Is there anything wrong with set up? On Tue, Mar 5, 2013 at 5:43 PM, Sujatha Arun suja.a...@gmail.com wrote: Hi Otis, Since currently we are planning for only one slave due to cost considerations, can we have an ELB fronting the master and slave for HA. 1. All index requests will go to the master . 2. Slave replicates from master . 3. Search request can go either to master /slave via ELB. is that resonable HA for search ? Regards Sujatha On Tue, Mar 5, 2013 at 5:12 PM, Otis Gospodnetic otis.gospodne...@gmail.com wrote: Hi Sujatha, If I understand correctly, you will have only 1 slave (and 1 master), so that's not really a HA architecture. You could manually turn master into slave, but that's going to mean some down time... Otis -- Solr ElasticSearch Support http://sematext.com/ On Tue, Mar 5, 2013 at 3:05 AM, Sujatha Arun suja.a...@gmail.com wrote: Hi, We are planning to set up *2* *High-Memory Quadruple Extra Large Instance *as master and slave for our multicore solr setup which has more than 200 cores spread between a couple of webapps on a single JVM on *AWS* All indexing [via a queue will go to master ] . One Slave Server will replicate