Re: Cluster design distribution and JBOD vs RAID

2014-04-23 Thread Bert Corderman
ly absolutely necessary. :) > > The high level Kafka consumer is good at "at least once" processing. > Exactly once is a harder nut to crack. Exactly once processing may require > some custom code around the low-level Kafka consumer client. > > - Bob > > > > -Ori

Re: Cluster design distribution and JBOD vs RAID

2014-04-21 Thread Andrew Otto
I >> suggest >>> you try testing it. >>> >>> >>> - Bob >>> >>> >>> -----Original Message- >>> From: Andrew Otto [mailto:ao...@wikimedia.org] >>> Sent: Friday, April 18, 2014 8:36 AM >>> To: users@

Re: Cluster design distribution and JBOD vs RAID

2014-04-19 Thread Jay Kreps
are missing or not. I > suggest > > you try testing it. > > > > > > - Bob > > > > > > -Original Message- > > From: Andrew Otto [mailto:ao...@wikimedia.org] > > Sent: Friday, April 18, 2014 8:36 AM > > To: users@kafka.apache.org >

RE: Cluster design distribution and JBOD vs RAID

2014-04-18 Thread Maxime Brugidou
topics/partitions are missing or not. I suggest > you try testing it. > > > - Bob > > > -Original Message- > From: Andrew Otto [mailto:ao...@wikimedia.org] > Sent: Friday, April 18, 2014 8:36 AM > To: users@kafka.apache.org > Subject: Re: Cluster design

RE: Cluster design distribution and JBOD vs RAID

2014-04-18 Thread Bello, Bob
AM To: users@kafka.apache.org Subject: Re: Cluster design distribution and JBOD vs RAID > BOB> We are using RAID10. It was a requirement from our Unix guys. The > rationale for this was we didn't want to lose just a disk and to have to > rebuild/re-replicate 20TB of data. We haven&#

Re: Cluster design distribution and JBOD vs RAID

2014-04-18 Thread Jay Kreps
mers are flush per > message processed. It's the flush per message that causes the high-volume. > > > > Push back on DEVs and software architecture if they want to flush per > message. Do it where it's only absolutely necessary. :) > > > > The high level Kafka con

Re: Cluster design distribution and JBOD vs RAID

2014-04-18 Thread Andrew Otto
consumers are flush per message processed. It's the flush > per message that causes the high-volume. > > Push back on DEVs and software architecture if they want to flush per > message. Do it where it's only absolutely necessary. :) > > The high level Kafka consumer is good at "at le

RE: Cluster design distribution and JBOD vs RAID

2014-04-17 Thread Bello, Bob
crack. Exactly once processing may require some custom code around the low-level Kafka consumer client. - Bob -Original Message- From: bertc...@gmail.com [mailto:bertc...@gmail.com] On Behalf Of Bert Corderman Sent: Thursday, April 17, 2014 7:21 AM To: users@kafka.apache.org Subjec

Re: Cluster design distribution and JBOD vs RAID

2014-04-17 Thread Bert Corderman
Hey Bob, thanks for your detailed response. I have added comments inline. On Wed, Apr 16, 2014 at 7:41 PM, Bello, Bob wrote: > Perhaps as you consider the size of your cluster, a few questions about > the kind of messaging you are looking at? I can use an example of what we > do in our produc

RE: Cluster design distribution and JBOD vs RAID

2014-04-16 Thread Bello, Bob
Perhaps as you consider the size of your cluster, a few questions about the kind of messaging you are looking at? I can use an example of what we do in our production environment while not going into specifics. These are just observations from an OPS perspective. (sorry for the wall of text.) *