Actually, we just reply with 200 OK to users just after receiving the request (i.e before sending the message to kafka).
We have implemented a native storm topology which uses the core Kafka spout (https://github.com/apache/storm/tree/master/external/storm-kafka) 2014-09-19 12:24 GMT+02:00 Ayush Vatsyayan <ayushv...@gmail.com>: > Thanks Florian. I've one more query - how are you sending the reply back > to user. Are you using DRPC in trident topology or using the DRPC in native > storm. > > The native one is deprecated, so I'm trying to use the trident topology > one. > > On Fri, Sep 19, 2014 at 3:08 PM, Florian Hussonnois <fhussonn...@gmail.com > > wrote: > >> Hi, >> >> I currently working on a project which looks like yours. >> >> We are using vert.x for implementing web services. Each request is >> formatted as json and sent to a kafka topic. After that, we have a storm >> topology (using a KafkaSpout) which performs request parameters extraction >> and data enrichment on each message before writing them into hbase. >> >> In a development environment (one Vertx instance and storm cluster >> running on the same host ) we actually handle more than 10K req/sec. >> >> Hope this helps! >> >> 2014-09-19 8:43 GMT+02:00 Ayush Vatsyayan <ayushv...@gmail.com>: >> >>> can anyone provide any leads or insight into it? >>> >>> I also looked into the DRPC but, it's deprecated for the native storm. >>> While in trident it's a bit complex and I cannot find anything that >>> describes the indepth of what all trident is providing. >>> >>> On Tue, Sep 16, 2014 at 12:43 PM, Ayush Vatsyayan <ayushv...@gmail.com> >>> wrote: >>> >>>> We are trying to build a webservices application that can support *10k >>>> TPS*. I'm trying to do some POC's on strom, but I'm a bit concerned if >>>> using storm is the right fit here. Here is the scenario: >>>> >>>> Client will send a webservice request, which we will receive it (using >>>> apache CXF) and push it into JMS (probably kafka or RabbitMQ). From JMS >>>> storm spout will receive it and sent it to the bolt. In bolt we will be >>>> performing the validation that involves db calls, and once done we will >>>> persist the data in no-sql db. >>>> >>>> I understand the advantages of using storm, but my concern is that we >>>> are not performing some complex bolt chaining and might be using one or two >>>> bolts. I'm confused whether storm fits well in this case? >>>> >>>> P.S. we are planning to deploy webservices on the application server in >>>> cluster setup to support 10k TPS. Not sure if cluster setup is good >>>> approach, but I'll look into it later. >>>> >>> >>> >> >> >> -- >> Florian HUSSONNOIS >> Tel +33 6 26 92 82 23 >> > > -- Florian HUSSONNOIS Tel +33 6 26 92 82 23
