Fwd: Job Scheduler, Task Scheduler and Fair Scheduler
Any updates!! -- Forwarded message -- From: kartheek muthyala kartheek0...@gmail.com Date: Fri, Sep 16, 2011 at 8:38 PM Subject: Job Scheduler, Task Scheduler and Fair Scheduler To: common-user@hadoop.apache.org Hi all, Can any one explain me the responsibilities of each scheduler?. I am interested in the flow of commands that goes between these scheduler. And if any one have any info regarding how the job scheduler schedules a job based on the data locality?. As of I know, there is some heartbeat mechanism that goes from task scheduler to job scheduler and in response job scheduler does something here to find out the node where the data is more closely located and schedules the task in that node. Is there an elaborate way of explanation around this area?. Any help will be greatly appreciated. Thanks and Regards, Kartheek.
Re: Job Scheduler, Task Scheduler and Fair Scheduler
On Sep 16, 2011, at 11:26 PM, kartheek muthyala wrote: Any updates!! A bit of patience will help. It also helps to do some homework and ask specific questions. I don't know if you have looked at any of the code, but there are 3 schedulers: JobQueueTaskScheduler (aka default scheduler or fifo scheduler) Capacity Scheduler (CS) Fair Scheduler (FS). TaskScheduler is just an interface for all schedulers (default, CS, FS). Then there is JobInProgress which handles scheduling for map tasks of an individual job based on data locality (JobInProgress.obtainNew*MapTask). Other than that each of the schedulers (default, CS, FS) use different criteria for picking a certain job to offer a 'slot' on a given TT when it's available. All this has changed radically and completely with MRv2 which is now in branch-0.23 and trunk to allow MR and non-MR apps on same Hadoop cluster: http://wiki.apache.org/hadoop/NextGenMapReduce Arun -- Forwarded message -- From: kartheek muthyala kartheek0...@gmail.com Date: Fri, Sep 16, 2011 at 8:38 PM Subject: Job Scheduler, Task Scheduler and Fair Scheduler To: common-user@hadoop.apache.org Hi all, Can any one explain me the responsibilities of each scheduler?. I am interested in the flow of commands that goes between these scheduler. And if any one have any info regarding how the job scheduler schedules a job based on the data locality?. As of I know, there is some heartbeat mechanism that goes from task scheduler to job scheduler and in response job scheduler does something here to find out the node where the data is more closely located and schedules the task in that node. Is there an elaborate way of explanation around this area?. Any help will be greatly appreciated. Thanks and Regards, Kartheek.
Ganglia for hadoop monitoring
Hi all, First of all, ganglia integration with hadoop is an awesome feature. Kudos to the hadoop devs. Unfortunately its not working out for me. I am unable to see hadoop specific metrics in my ganglia frontend. My configurations are as follows: gmetad.coinf : data_source hadoop test host-name of gmetad ( I also tried data_source hadoop test list of machine:port separated by spaces. None of them worked. gmond conf: cluster { name = hadoop test owner = jS latlong = url = abc.com } udp_send_channel { host = host-name of gmetad port = 8649 ttl = 1 } udp_recv_channel { port = 8649 } I restarted my cluster several times and also gmonds and gmetad. However I am unable to see hadoop metrics in my page. Infact nothing with name hadoop test turns up Am I missing something? I have a couple of doutbs here. 1) Is it compulsory that hadoop NN should run the gmetad daemon and the front end ? Every tutorial on the net assumes it that way. However I am running gmetad on a separate node and I included NN in the node list (running a gmond on it). 2) How does hadoop communicate its metrics with the gmond Any help is highly appreciated. Thanks , JS
Any way to recover CORRUPT/MISSING blocks? (was: HELP NEEDED: What to do after crash and fsck says that .2% Blocks missing. Namenode in safemode)
Just want to follow up to first thank QwertyM aka Harsh Chouraria for helping me out on the IRC channel. Well beyond the call of duty! Its people like Harsh that make the HBase/Hadoop community what it is and one of the joys of working with this technology. And then one follow on question on how to recover from CORRUPT blocks. The main thing I learnt other than being careful not to install packages on all the regionservers/slaves at one time that may cause Out of Memory Errors and crash all your java processes.. is that if: Your namenode is stuck in safe mode, and even though the namenode log says that Safe mode will be turned off automatically. If there is enough wrong with your HDFS system like too many under-replicated blocks. It seems that it has to be out of safe mode to correct the problem... I hallucinated that the datanodes by doing verifications were doing the work to get the namenode out of safe mode. And probably would have waited another few hours if Harsh hadn't helped me out and told me what probably everyone but me knew: hadoop dfsadmin -safemode leave CURRENT QUESTION ON CORRUPT BLOCKS: -- After that the namenode did get all the under-replicated blocks replicated, but I ended up with about 200 blocks that fsck considered CORRUPT and/or MISSING. It looked like tables were being compacted when the outage occurred. Otherwise I don't know why a lot of the bad blocks are in old tables, not data being written at the time of the crash. The hdfs filesystem dates also showed them as being old. I am not sure what is the best thing to do now to be able to recover the CORRUPT/MISSING blocks and to get fsck to say all is healthy. Is the best thing to just do: hadoop fsck -move which will move what is left of the corrupt blocks into hdfs /lost+found? Is there any way to recover those blocks? I may be able to get them from the backup/export of all our tables we did recently and I believe I can regenerate the rest. But it would be nice to know if there is a way to recover them if there was no other way. Thanks in advance. Rob On Sep 16, 2011, at 12:50 AM, Robert J Berger wrote: Just had an HDFS/HBase instance where all the slave/regionservers processes crashed, but the namenode stayed up. I did proper shutdown of the namenode After bringing Hadoop back up the namenode is stuck in safe mode. Fsck shows 235 corrupt/missing blocks out of 117280 Blocks. All the slaves are doing DataBlockScanner: Verification succeeded. As far as I can tell there are no errors in the datanodes. Can I expect it to self-heal? Or do I need to do something to help it along? Anyway to tell how long it will take to recover if I do have to just wait? Other than the verification messages on the datanodes, the namenode fsck numbers are not changing and the namenode log continues to say: The ratio of reported blocks 0.9980 has not reached the threshold 0.9990. Safe mode will be turned off automatically. The ratio has not changed for over an hour now. If you happen to know the answer, please get back to me right away by email or on #hadoop IRC as I'm trying to figure it out now... Thanks! __ Robert J Berger - CTO Runa Inc. +1 408-838-8896 http://blog.ibd.com __ Robert J Berger - CTO Runa Inc. +1 408-838-8896 http://blog.ibd.com
Re: Job Scheduler, Task Scheduler and Fair Scheduler
Hey Arun, Thanks for the information. And sorry for my previous mail regarding updates!! I just wanted to emphasize the importance of the query. I couldn't get enough time to go through the code that's why i approached you guys, as you are expertise in this area. Thanks Regards, Kartheek. On Sat, Sep 17, 2011 at 12:09 PM, Arun C Murthy a...@hortonworks.com wrote: On Sep 16, 2011, at 11:26 PM, kartheek muthyala wrote: Any updates!! A bit of patience will help. It also helps to do some homework and ask specific questions. I don't know if you have looked at any of the code, but there are 3 schedulers: JobQueueTaskScheduler (aka default scheduler or fifo scheduler) Capacity Scheduler (CS) Fair Scheduler (FS). TaskScheduler is just an interface for all schedulers (default, CS, FS). Then there is JobInProgress which handles scheduling for map tasks of an individual job based on data locality (JobInProgress.obtainNew*MapTask). Other than that each of the schedulers (default, CS, FS) use different criteria for picking a certain job to offer a 'slot' on a given TT when it's available. All this has changed radically and completely with MRv2 which is now in branch-0.23 and trunk to allow MR and non-MR apps on same Hadoop cluster: http://wiki.apache.org/hadoop/NextGenMapReduce Arun -- Forwarded message -- From: kartheek muthyala kartheek0...@gmail.com Date: Fri, Sep 16, 2011 at 8:38 PM Subject: Job Scheduler, Task Scheduler and Fair Scheduler To: common-user@hadoop.apache.org Hi all, Can any one explain me the responsibilities of each scheduler?. I am interested in the flow of commands that goes between these scheduler. And if any one have any info regarding how the job scheduler schedules a job based on the data locality?. As of I know, there is some heartbeat mechanism that goes from task scheduler to job scheduler and in response job scheduler does something here to find out the node where the data is more closely located and schedules the task in that node. Is there an elaborate way of explanation around this area?. Any help will be greatly appreciated. Thanks and Regards, Kartheek.
Re: risks of using Hadoop
Hi George, You can use it noramally as well. Append interfaces will be exposed. For Hbase, append support is required very much. Regards, Uma - Original Message - From: George Kousiouris gkous...@mail.ntua.gr Date: Saturday, September 17, 2011 12:29 pm Subject: Re: risks of using Hadoop To: common-user@hadoop.apache.org Cc: Uma Maheswara Rao G 72686 mahesw...@huawei.com Hi, When you say that 0.20.205 will support appends, you mean for general purpose writes on the HDFS? or only Hbase? Thanks, George On 9/17/2011 7:08 AM, Uma Maheswara Rao G 72686 wrote: 6. If you plan to use Hbase, it requires append support. 20Append has the support for append. 0.20.205 release also will have append support but not yet released. Choose your correct version to avoid sudden surprises. Regards, Uma - Original Message - From: Kobina Kwarkokobina.kwa...@gmail.com Date: Saturday, September 17, 2011 3:42 am Subject: Re: risks of using Hadoop To: common-user@hadoop.apache.org We are planning to use Hadoop in my organisation for quality of servicesanalysis out of CDR records from mobile operators. We are thinking of having a small cluster of may be 10 - 15 nodes and I'm preparing the proposal. my office requires that i provide some risk analysis in the proposal. thank you. On 16 September 2011 20:34, Uma Maheswara Rao G 72686 mahesw...@huawei.comwrote: Hello, First of all where you are planning to use Hadoop? Regards, Uma - Original Message - From: Kobina Kwarkokobina.kwa...@gmail.com Date: Saturday, September 17, 2011 0:41 am Subject: risks of using Hadoop To: common-usercommon-user@hadoop.apache.org Hello, Please can someone point some of the risks we may incur if we decide to implement Hadoop? BR, Isaac. -- --- George Kousiouris Electrical and Computer Engineer Division of Communications, Electronics and Information Engineering School of Electrical and Computer Engineering Tel: +30 210 772 2546 Mobile: +30 6939354121 Fax: +30 210 772 2569 Email: gkous...@mail.ntua.gr Site: http://users.ntua.gr/gkousiou/ National Technical University of Athens 9 Heroon Polytechniou str., 157 73 Zografou, Athens, Greece
Re: risks of using Hadoop
To clarify, *append* is not supported and is known to be buggy. *sync* support is what HBase needs and what 0.20.205 will support. Before 205 is released, you can also find these features in CDH3 or by building your own release from SVN. -Todd On Sat, Sep 17, 2011 at 4:59 AM, Uma Maheswara Rao G 72686 mahesw...@huawei.com wrote: Hi George, You can use it noramally as well. Append interfaces will be exposed. For Hbase, append support is required very much. Regards, Uma - Original Message - From: George Kousiouris gkous...@mail.ntua.gr Date: Saturday, September 17, 2011 12:29 pm Subject: Re: risks of using Hadoop To: common-user@hadoop.apache.org Cc: Uma Maheswara Rao G 72686 mahesw...@huawei.com Hi, When you say that 0.20.205 will support appends, you mean for general purpose writes on the HDFS? or only Hbase? Thanks, George On 9/17/2011 7:08 AM, Uma Maheswara Rao G 72686 wrote: 6. If you plan to use Hbase, it requires append support. 20Append has the support for append. 0.20.205 release also will have append support but not yet released. Choose your correct version to avoid sudden surprises. Regards, Uma - Original Message - From: Kobina Kwarkokobina.kwa...@gmail.com Date: Saturday, September 17, 2011 3:42 am Subject: Re: risks of using Hadoop To: common-user@hadoop.apache.org We are planning to use Hadoop in my organisation for quality of servicesanalysis out of CDR records from mobile operators. We are thinking of having a small cluster of may be 10 - 15 nodes and I'm preparing the proposal. my office requires that i provide some risk analysis in the proposal. thank you. On 16 September 2011 20:34, Uma Maheswara Rao G 72686 mahesw...@huawei.comwrote: Hello, First of all where you are planning to use Hadoop? Regards, Uma - Original Message - From: Kobina Kwarkokobina.kwa...@gmail.com Date: Saturday, September 17, 2011 0:41 am Subject: risks of using Hadoop To: common-usercommon-user@hadoop.apache.org Hello, Please can someone point some of the risks we may incur if we decide to implement Hadoop? BR, Isaac. -- --- George Kousiouris Electrical and Computer Engineer Division of Communications, Electronics and Information Engineering School of Electrical and Computer Engineering Tel: +30 210 772 2546 Mobile: +30 6939354121 Fax: +30 210 772 2569 Email: gkous...@mail.ntua.gr Site: http://users.ntua.gr/gkousiou/ National Technical University of Athens 9 Heroon Polytechniou str., 157 73 Zografou, Athens, Greece -- Todd Lipcon Software Engineer, Cloudera
Re: risks of using Hadoop
Yes, I was mentioning before append beacuse branch name itself is 20Append. Sync is the main api name to sync editlogs. @George mainly need to consider the Hbase usage. sync is supported. append api has some open issues. for example https://issues.apache.org/jira/browse/HDFS-1228 Apologies for the confusions if any. Thanks a lot for more clarification! Thanks Uma - Original Message - From: Todd Lipcon t...@cloudera.com Date: Sunday, September 18, 2011 1:35 am Subject: Re: risks of using Hadoop To: common-user@hadoop.apache.org To clarify, *append* is not supported and is known to be buggy. *sync* support is what HBase needs and what 0.20.205 will support. Before 205 is released, you can also find these features in CDH3 or by building your own release from SVN. -Todd On Sat, Sep 17, 2011 at 4:59 AM, Uma Maheswara Rao G 72686 mahesw...@huawei.com wrote: Hi George, You can use it noramally as well. Append interfaces will be exposed. For Hbase, append support is required very much. Regards, Uma - Original Message - From: George Kousiouris gkous...@mail.ntua.gr Date: Saturday, September 17, 2011 12:29 pm Subject: Re: risks of using Hadoop To: common-user@hadoop.apache.org Cc: Uma Maheswara Rao G 72686 mahesw...@huawei.com Hi, When you say that 0.20.205 will support appends, you mean for general purpose writes on the HDFS? or only Hbase? Thanks, George On 9/17/2011 7:08 AM, Uma Maheswara Rao G 72686 wrote: 6. If you plan to use Hbase, it requires append support. 20Append has the support for append. 0.20.205 release also will have append support but not yet released. Choose your correct version to avoid sudden surprises. Regards, Uma - Original Message - From: Kobina Kwarkokobina.kwa...@gmail.com Date: Saturday, September 17, 2011 3:42 am Subject: Re: risks of using Hadoop To: common-user@hadoop.apache.org We are planning to use Hadoop in my organisation for quality of servicesanalysis out of CDR records from mobile operators. We are thinking of having a small cluster of may be 10 - 15 nodes and I'm preparing the proposal. my office requires that i provide some risk analysis in the proposal. thank you. On 16 September 2011 20:34, Uma Maheswara Rao G 72686 mahesw...@huawei.comwrote: Hello, First of all where you are planning to use Hadoop? Regards, Uma - Original Message - From: Kobina Kwarkokobina.kwa...@gmail.com Date: Saturday, September 17, 2011 0:41 am Subject: risks of using Hadoop To: common-usercommon-user@hadoop.apache.org Hello, Please can someone point some of the risks we may incur if we decide to implement Hadoop? BR, Isaac. -- --- George Kousiouris Electrical and Computer Engineer Division of Communications, Electronics and Information Engineering School of Electrical and Computer Engineering Tel: +30 210 772 2546 Mobile: +30 6939354121 Fax: +30 210 772 2569 Email: gkous...@mail.ntua.gr Site: http://users.ntua.gr/gkousiou/ National Technical University of Athens 9 Heroon Polytechniou str., 157 73 Zografou, Athens, Greece -- Todd Lipcon Software Engineer, Cloudera
Re: risks of using Hadoop
On Sep 16, 2011, at 11:08 PM, Uma Maheswara Rao G 72686 wrote: Hi Kobina, Some experiences which may helpful for you with respective to DFS. 1. Selecting the correct version. I will recommend to use 0.20X version. This is pretty stable version and all other organizations prefers it. Well tested as well. Dont go for 21 version.This version is not a stable version.This is risk. 2. You should perform thorough test with your customer operations. (of-course you will do this :-)) 3. 0.20x version has the problem of SPOF. If NameNode goes down you will loose the data.One way of recovering is by using the secondaryNameNode.You can recover the data till last checkpoint.But here manual intervention is required. In latest trunk SPOF will be addressed bu HDFS-1623. 4. 0.20x NameNodes can not scale. Federation changes included in latest versions. ( i think in 22). this may not be the problem for your cluster. But please consider this aspect as well. With respect to (3) and (4) - these are often completely overblown for many Hadoop use cases. If you use Hadoop as originally designed (large scale batch data processing), these likely don't matter. If you're looking at some of the newer use cases (low latency stuff or time-critical processing), or if you architect your solution poorly (lots of small files), these issues become relevant. Another case where I see folks get frustrated is using Hadoop as a plain old batch system; for non-data workflows, it doesn't measure up against specialized systems. You really want to make sure that Hadoop is the best tool for your job. Brian
Re: risks of using Hadoop
I disagree Brian - data loss and system down time (both potentially non-trival) should not be taken lightly. Use cases and thus availability requirements do vary, but I would not encourage anyone to shrug them off as overblown, especially as Hadoop become more production oriented in utilization. --- Sent from my Blackberry so please excuse typing and spelling errors. - Original Message - From: Brian Bockelman [bbock...@cse.unl.edu] Sent: 09/17/2011 05:11 PM EST To: common-user@hadoop.apache.org Subject: Re: risks of using Hadoop On Sep 16, 2011, at 11:08 PM, Uma Maheswara Rao G 72686 wrote: Hi Kobina, Some experiences which may helpful for you with respective to DFS. 1. Selecting the correct version. I will recommend to use 0.20X version. This is pretty stable version and all other organizations prefers it. Well tested as well. Dont go for 21 version.This version is not a stable version.This is risk. 2. You should perform thorough test with your customer operations. (of-course you will do this :-)) 3. 0.20x version has the problem of SPOF. If NameNode goes down you will loose the data.One way of recovering is by using the secondaryNameNode.You can recover the data till last checkpoint.But here manual intervention is required. In latest trunk SPOF will be addressed bu HDFS-1623. 4. 0.20x NameNodes can not scale. Federation changes included in latest versions. ( i think in 22). this may not be the problem for your cluster. But please consider this aspect as well. With respect to (3) and (4) - these are often completely overblown for many Hadoop use cases. If you use Hadoop as originally designed (large scale batch data processing), these likely don't matter. If you're looking at some of the newer use cases (low latency stuff or time-critical processing), or if you architect your solution poorly (lots of small files), these issues become relevant. Another case where I see folks get frustrated is using Hadoop as a plain old batch system; for non-data workflows, it doesn't measure up against specialized systems. You really want to make sure that Hadoop is the best tool for your job. Brian
Re: risks of using Hadoop
Data loss in a batch-oriented environment is different than data loss in an online/production environment. It's a trade-off, and I personally think many folks don't weigh the costs well. As you mention - Hadoop is becoming more production oriented in utilization. *In those cases*, you definitely don't want to shrug off data loss / downtime. However, there's many people who simply don't need this. If I'm told that I can buy a 10% larger cluster by accepting up to 15 minutes of data loss, I'd do it in a heartbeat where I work. Brian On Sep 17, 2011, at 6:38 PM, Tom Deutsch wrote: I disagree Brian - data loss and system down time (both potentially non-trival) should not be taken lightly. Use cases and thus availability requirements do vary, but I would not encourage anyone to shrug them off as overblown, especially as Hadoop become more production oriented in utilization. --- Sent from my Blackberry so please excuse typing and spelling errors. - Original Message - From: Brian Bockelman [bbock...@cse.unl.edu] Sent: 09/17/2011 05:11 PM EST To: common-user@hadoop.apache.org Subject: Re: risks of using Hadoop On Sep 16, 2011, at 11:08 PM, Uma Maheswara Rao G 72686 wrote: Hi Kobina, Some experiences which may helpful for you with respective to DFS. 1. Selecting the correct version. I will recommend to use 0.20X version. This is pretty stable version and all other organizations prefers it. Well tested as well. Dont go for 21 version.This version is not a stable version.This is risk. 2. You should perform thorough test with your customer operations. (of-course you will do this :-)) 3. 0.20x version has the problem of SPOF. If NameNode goes down you will loose the data.One way of recovering is by using the secondaryNameNode.You can recover the data till last checkpoint.But here manual intervention is required. In latest trunk SPOF will be addressed bu HDFS-1623. 4. 0.20x NameNodes can not scale. Federation changes included in latest versions. ( i think in 22). this may not be the problem for your cluster. But please consider this aspect as well. With respect to (3) and (4) - these are often completely overblown for many Hadoop use cases. If you use Hadoop as originally designed (large scale batch data processing), these likely don't matter. If you're looking at some of the newer use cases (low latency stuff or time-critical processing), or if you architect your solution poorly (lots of small files), these issues become relevant. Another case where I see folks get frustrated is using Hadoop as a plain old batch system; for non-data workflows, it doesn't measure up against specialized systems. You really want to make sure that Hadoop is the best tool for your job. Brian
Re: risks of using Hadoop
Not trying to give you a hard time Brian - we just have different users/customers/expectations on us. --- Sent from my Blackberry so please excuse typing and spelling errors. - Original Message - From: Brian Bockelman [bbock...@cse.unl.edu] Sent: 09/17/2011 08:10 PM EST To: common-user@hadoop.apache.org Subject: Re: risks of using Hadoop Data loss in a batch-oriented environment is different than data loss in an online/production environment. It's a trade-off, and I personally think many folks don't weigh the costs well. As you mention - Hadoop is becoming more production oriented in utilization. *In those cases*, you definitely don't want to shrug off data loss / downtime. However, there's many people who simply don't need this. If I'm told that I can buy a 10% larger cluster by accepting up to 15 minutes of data loss, I'd do it in a heartbeat where I work. Brian On Sep 17, 2011, at 6:38 PM, Tom Deutsch wrote: I disagree Brian - data loss and system down time (both potentially non-trival) should not be taken lightly. Use cases and thus availability requirements do vary, but I would not encourage anyone to shrug them off as overblown, especially as Hadoop become more production oriented in utilization. --- Sent from my Blackberry so please excuse typing and spelling errors. - Original Message - From: Brian Bockelman [bbock...@cse.unl.edu] Sent: 09/17/2011 05:11 PM EST To: common-user@hadoop.apache.org Subject: Re: risks of using Hadoop On Sep 16, 2011, at 11:08 PM, Uma Maheswara Rao G 72686 wrote: Hi Kobina, Some experiences which may helpful for you with respective to DFS. 1. Selecting the correct version. I will recommend to use 0.20X version. This is pretty stable version and all other organizations prefers it. Well tested as well. Dont go for 21 version.This version is not a stable version.This is risk. 2. You should perform thorough test with your customer operations. (of-course you will do this :-)) 3. 0.20x version has the problem of SPOF. If NameNode goes down you will loose the data.One way of recovering is by using the secondaryNameNode.You can recover the data till last checkpoint.But here manual intervention is required. In latest trunk SPOF will be addressed bu HDFS-1623. 4. 0.20x NameNodes can not scale. Federation changes included in latest versions. ( i think in 22). this may not be the problem for your cluster. But please consider this aspect as well. With respect to (3) and (4) - these are often completely overblown for many Hadoop use cases. If you use Hadoop as originally designed (large scale batch data processing), these likely don't matter. If you're looking at some of the newer use cases (low latency stuff or time-critical processing), or if you architect your solution poorly (lots of small files), these issues become relevant. Another case where I see folks get frustrated is using Hadoop as a plain old batch system; for non-data workflows, it doesn't measure up against specialized systems. You really want to make sure that Hadoop is the best tool for your job. Brian
Re: risks of using Hadoop
:) I think we can agree to that point. Hopefully a plethora of viewpoints is good for the community! (And when we run into something that needs higher availability, I'll drop by and say hi!) On Sep 17, 2011, at 8:32 PM, Tom Deutsch wrote: Not trying to give you a hard time Brian - we just have different users/customers/expectations on us. --- Sent from my Blackberry so please excuse typing and spelling errors. - Original Message - From: Brian Bockelman [bbock...@cse.unl.edu] Sent: 09/17/2011 08:10 PM EST To: common-user@hadoop.apache.org Subject: Re: risks of using Hadoop Data loss in a batch-oriented environment is different than data loss in an online/production environment. It's a trade-off, and I personally think many folks don't weigh the costs well. As you mention - Hadoop is becoming more production oriented in utilization. *In those cases*, you definitely don't want to shrug off data loss / downtime. However, there's many people who simply don't need this. If I'm told that I can buy a 10% larger cluster by accepting up to 15 minutes of data loss, I'd do it in a heartbeat where I work. Brian On Sep 17, 2011, at 6:38 PM, Tom Deutsch wrote: I disagree Brian - data loss and system down time (both potentially non-trival) should not be taken lightly. Use cases and thus availability requirements do vary, but I would not encourage anyone to shrug them off as overblown, especially as Hadoop become more production oriented in utilization. --- Sent from my Blackberry so please excuse typing and spelling errors. - Original Message - From: Brian Bockelman [bbock...@cse.unl.edu] Sent: 09/17/2011 05:11 PM EST To: common-user@hadoop.apache.org Subject: Re: risks of using Hadoop On Sep 16, 2011, at 11:08 PM, Uma Maheswara Rao G 72686 wrote: Hi Kobina, Some experiences which may helpful for you with respective to DFS. 1. Selecting the correct version. I will recommend to use 0.20X version. This is pretty stable version and all other organizations prefers it. Well tested as well. Dont go for 21 version.This version is not a stable version.This is risk. 2. You should perform thorough test with your customer operations. (of-course you will do this :-)) 3. 0.20x version has the problem of SPOF. If NameNode goes down you will loose the data.One way of recovering is by using the secondaryNameNode.You can recover the data till last checkpoint.But here manual intervention is required. In latest trunk SPOF will be addressed bu HDFS-1623. 4. 0.20x NameNodes can not scale. Federation changes included in latest versions. ( i think in 22). this may not be the problem for your cluster. But please consider this aspect as well. With respect to (3) and (4) - these are often completely overblown for many Hadoop use cases. If you use Hadoop as originally designed (large scale batch data processing), these likely don't matter. If you're looking at some of the newer use cases (low latency stuff or time-critical processing), or if you architect your solution poorly (lots of small files), these issues become relevant. Another case where I see folks get frustrated is using Hadoop as a plain old batch system; for non-data workflows, it doesn't measure up against specialized systems. You really want to make sure that Hadoop is the best tool for your job. Brian
RE: risks of using Hadoop
Gee Tom, No disrespect, but I don't believe you have any personal practical experience in designing and building out clusters or putting them to the test. Now to the points that Brian raised.. 1) SPOF... it sounds great on paper. Some FUD to scare someone away from Hadoop. But in reality... you can mitigate your risks by setting up raid on your NN/HM node. You can also NFS mount a copy to your SN (or whatever they're calling it these days...) Or you can go to MapR which has redesigned HDFS which removes this problem. But with your Apache Hadoop or Cloudera's release, losing your NN is rare. Yes it can happen, but not your greatest risk. (Not by a long shot) 2) Data Loss. You can mitigate this as well. Do I need to go through all of the options and DR/BCP planning? Sure there's always a chance that you have some Luser who does something brain dead. This is true of all databases and systems. (I know I can probably recount some of IBM's Informix and DB2 having data loss issues. But that's a topic for another time. ;-) I can't speak for Brian, but I don't think he's trivializing it. In fact I think he's doing a fine job of level setting expectations. And if you talk to Ted Dunning of MapR, I'm sure he'll point out that their current release does address points 3 and 4 again making their risks moot. (At least if you're using MapR) -Mike Subject: Re: risks of using Hadoop From: tdeut...@us.ibm.com Date: Sat, 17 Sep 2011 17:38:27 -0600 To: common-user@hadoop.apache.org I disagree Brian - data loss and system down time (both potentially non-trival) should not be taken lightly. Use cases and thus availability requirements do vary, but I would not encourage anyone to shrug them off as overblown, especially as Hadoop become more production oriented in utilization. --- Sent from my Blackberry so please excuse typing and spelling errors. - Original Message - From: Brian Bockelman [bbock...@cse.unl.edu] Sent: 09/17/2011 05:11 PM EST To: common-user@hadoop.apache.org Subject: Re: risks of using Hadoop On Sep 16, 2011, at 11:08 PM, Uma Maheswara Rao G 72686 wrote: Hi Kobina, Some experiences which may helpful for you with respective to DFS. 1. Selecting the correct version. I will recommend to use 0.20X version. This is pretty stable version and all other organizations prefers it. Well tested as well. Dont go for 21 version.This version is not a stable version.This is risk. 2. You should perform thorough test with your customer operations. (of-course you will do this :-)) 3. 0.20x version has the problem of SPOF. If NameNode goes down you will loose the data.One way of recovering is by using the secondaryNameNode.You can recover the data till last checkpoint.But here manual intervention is required. In latest trunk SPOF will be addressed bu HDFS-1623. 4. 0.20x NameNodes can not scale. Federation changes included in latest versions. ( i think in 22). this may not be the problem for your cluster. But please consider this aspect as well. With respect to (3) and (4) - these are often completely overblown for many Hadoop use cases. If you use Hadoop as originally designed (large scale batch data processing), these likely don't matter. If you're looking at some of the newer use cases (low latency stuff or time-critical processing), or if you architect your solution poorly (lots of small files), these issues become relevant. Another case where I see folks get frustrated is using Hadoop as a plain old batch system; for non-data workflows, it doesn't measure up against specialized systems. You really want to make sure that Hadoop is the best tool for your job. Brian