Nothing specifically about our Hive setup although some of us at Forward have blogged bits and pieces about Hive + Hadoop and have a few Hadoop/Hive related libs on our GitHub account: https://github.com/forward.
I've blogged a few bits (http://www.oobaloo.co.uk/) as has one of my colleagues (http://blog.fingertap.org/post/1255463384/hive-thrift-client). Another colleague also presented a little about our setup during a Hadoop meetup last summer (http://skillsmatter.com/podcast/home/hadoop-in-context-1591). The numbers Andy mentioned will be a little out of date but it does include some screenshots of a few of the surrounding apps we built that connect to Hive and Hadoop (including a web based Hive query tool + work queue). I had a quick search through the mailing lists when we had connection problems but I think most of it was discussed/resolved during a chat I had with Shevek from Karmasphere at a London pub following a Hadoop meetup :) If you're interested, I've posted a gist (https://gist.github.com/953926) that contains our HAProxy config; clients connect to 10000 and are balanced between :10001 and :10005 on 2 servers (so actually 10 backend servers). Be happy to talk more about our experience- feel free to ping me an email off list if you'd like. On 3 May 2011, at 19:18, Matthew Rathbone wrote: > Hey Paul, > > I'd be very interested in reading about your hadoop/hive setup, do you have a > blog post or anything describing this setup, or some of the issues you've > have with hive? > > -- > Matthew Rathbone > Foursquare | Software Engineer | Server Engineering Team > matt...@foursquare.com | @rathboma | 4sq > > On Tuesday, May 3, 2011 at 2:15 PM, Paul Ingles wrote: > HiveServer does seem to support multiple connections but I think it still has > thread-safety problems (https://issues.apache.org/jira/browse/HIVE-80). >> >> We've (www.forward.co.uk) certainly had instability problems with the thrift >> server in the past and now run 5 or so instances behind the HAProxy >> load-balancer (http://haproxy.1wt.eu/). Since we did that it's been >> significantly better. >> >> I think the JDBC server still operates using thrift to connect to the >> HiveServer so I would expect it to have similar problems (but I may have got >> that wrong :) >> >> >> On 3 May 2011, at 18:59, Matthew Rathbone wrote: >> >>> Even if it is single threaded it certainly seems to support multiple >>> connections. >>> >>> We run 5 workers all connected at the same time executing a different query >>> each ( with a different connection per worker). >>> >>> Hope that helps >>> >>> Matthew >>> On Tuesday, May 3, 2011 at 1:40 PM, V.Senthil Kumar wrote: >>> Thanks Matthew. The wiki page http://wiki.apache.org/hadoop/Hive/HiveServer >>> says >>>> its single threaded. I have a queue of queries which gets added >>>> dynamically all >>>> the time. By the time I run 1 query using 1 JDBC connection, the queue >>>> gets >>>> added more queries and builds up a backlog. So, I was that's why I was >>>> wondering >>>> whether I can run two or more instances to avoid having a big backlog in >>>> queue. >>>> >>>> >>>> >>>> ----- Original Message ---- >>>> From: Matthew Rathbone <matt...@foursquare.com> >>>> To: user@hive.apache.org >>>> Sent: Tue, May 3, 2011 7:46:49 AM >>>> Subject: Re: HIVE Server multiple instances >>>> >>>> Why would you want to run two? I think it is multithreaded, so you can >>>> query it >>>> from two different connections >>>> >>>> -- >>>> Matthew Rathbone >>>> Foursquare | Software Engineer | Server Engineering Team >>>> matt...@foursquare.com | @rathboma | 4sq >>>> >>>> On Monday, May 2, 2011 at 6:41 PM, V.Senthil Kumar wrote: >>>> Hello, >>>>> >>>>> I have one instance of HIVE JDBC server running on port 10000. Can I run >>>>> another >>>>> >>>>> instance on different port ? Would it cause a concurrency issue on the >>>>> underlying data warehouse files ? Please clarify. >>>>> >>>>> Thanks, >>>>> V.Senthil Kumar >> >
