RE: how to use RegionCoprocessorEnvironment getSharedData() to share data among coprocessors?
Thank you Ted,
I am using 0.98.11. I do read the first example, but I don't find the second
one before, I will try to read it, but it seems too complex for me :-)
I read from
http://hadoop-hbase.blogspot.com/2012/10/coprocessor-access-to-hbase-internals.html
That : This shared data is per coprocessor class and per regionserver.
So it means to me that hbase cannot share between two different coprocessors
like an endpoint and an observer by using this sharedData, since there are two
different classes in my case. I use Zookeeper to share data for now, but I was
told not to depend on ZooKeeper too much, no idea why. Is there any other good
way I can use to share data among different coprocessors?
Thanks,
Ming
-Original Message-
From: Ted Yu [mailto:yuzhih...@gmail.com]
Sent: Wednesday, April 15, 2015 8:25 PM
To: user@hbase.apache.org
Subject: Re: how to use RegionCoprocessorEnvironment getSharedData() to share
data among coprocessors?
Which hbase release are you using ?
Please take a look at the following tests for example of using getSharedData() :
hbase-examples/src/main/java/org/apache/hadoop/hbase/coprocessor/example/ZooKeeperScanPolicyObserver.java
hbase-server/src/test/java/org/apache/hadoop/hbase/coprocessor/TestCoprocessorInterface.java
Cheers
On Tue, Apr 14, 2015 at 10:35 PM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com
wrote:
Hi, all,
I am trying to learn how to share data between two coprocessors. I
have one Observer coprocessor and one Endpoint coprocessor. In the
observer, it overload the prePut/preDelete to maintain a counter. And
I want the Endpoint coprocessor to read that counter and return to
client caller. So I want to use the getSharedData() method in
RegionCoprocessorEnvironment, but I cannot make it work. Could anybody help
me here?
In the Observer Coprocessor :
During start(), create the shared object counter:
-
public void start(CoprocessorEnvironment envi) throws IOException {
Env.getSharedData().put(counter, new Long(0) ); //create the
counter
-
In the Endpoint coprocessor:
During start(), try to read the shared counter , but failed.
--
public void start(CoprocessorEnvironment envi) throws IOException {
LOG.info(The size of sharedData map is: +
envi.getSharedData().size() ); //try to get the counter
--
Here it print 0, if I use evni.getSharedData().containsKey(counter),
it will return false.
When creating table, I call addCoprocessor() method to add Observer
first, then Endpoint coprocessor. I confirmed that by checking the
hbase log file message. I only have one region for that table during
the run. I confirmed by hbase shell status 'detailed' command.
There is not much example I can find about how to use getSharedData(),
could someone help me here? What is missing in my simple code? Thanks
very much in advance!
Thanks,
Ming
RE: how to use RegionCoprocessorEnvironment getSharedData() to share data among coprocessors?
Hi, Ted,
I am not sure. I am a C programmer, only know a very little java. It seems to
me that the two coprocessors have to 'extends' from two different base classes
for Endpoint and Observer. Java seems not be able to do this.
I am just trying to study coprocessor, there is no solid production requirement
to do this sharing thing. Coprocessor was introduced from 0.92, and no one need
this before, so I am thinking this is not something important.
Thanks always for your help! You are very kind to answer all my questions in
this maillist :-)
I will try to read the getSharedData code to further understand and feedback
here if I can have some more findings or any other good method to shared data
in a simple way.
Thanks,
Ming
-Original Message-
From: Ted Yu [mailto:yuzhih...@gmail.com]
Sent: Thursday, April 16, 2015 5:16 AM
To: user@hbase.apache.org
Subject: Re: how to use RegionCoprocessorEnvironment getSharedData() to share
data among coprocessors?
Can you implement the Observer coprocessor and the Endpoint coprocessor in one
class ?
Cheers
On Wed, Apr 15, 2015 at 9:00 AM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com
wrote:
Thank you Ted,
I am using 0.98.11. I do read the first example, but I don't find the
second one before, I will try to read it, but it seems too complex for
me
:-)
I read from
http://hadoop-hbase.blogspot.com/2012/10/coprocessor-access-to-hbase-i
nternals.html That : This shared data is per coprocessor class and
per regionserver.
So it means to me that hbase cannot share between two different
coprocessors like an endpoint and an observer by using this
sharedData, since there are two different classes in my case. I use
Zookeeper to share data for now, but I was told not to depend on
ZooKeeper too much, no idea why. Is there any other good way I can use
to share data among different coprocessors?
Thanks,
Ming
-Original Message-
From: Ted Yu [mailto:yuzhih...@gmail.com]
Sent: Wednesday, April 15, 2015 8:25 PM
To: user@hbase.apache.org
Subject: Re: how to use RegionCoprocessorEnvironment getSharedData()
to share data among coprocessors?
Which hbase release are you using ?
Please take a look at the following tests for example of using
getSharedData() :
hbase-examples/src/main/java/org/apache/hadoop/hbase/coprocessor/examp
le/ZooKeeperScanPolicyObserver.java
hbase-server/src/test/java/org/apache/hadoop/hbase/coprocessor/TestCop
rocessorInterface.java
Cheers
On Tue, Apr 14, 2015 at 10:35 PM, Liu, Ming (HPIT-GADSC)
ming.l...@hp.com
wrote:
Hi, all,
I am trying to learn how to share data between two coprocessors. I
have one Observer coprocessor and one Endpoint coprocessor. In the
observer, it overload the prePut/preDelete to maintain a counter.
And I want the Endpoint coprocessor to read that counter and return
to client caller. So I want to use the getSharedData() method in
RegionCoprocessorEnvironment, but I cannot make it work. Could
anybody
help me here?
In the Observer Coprocessor :
During start(), create the shared object counter:
-
public void start(CoprocessorEnvironment envi) throws IOException {
Env.getSharedData().put(counter, new Long(0) ); //create the
counter
-
In the Endpoint coprocessor:
During start(), try to read the shared counter , but failed.
--
public void start(CoprocessorEnvironment envi) throws IOException {
LOG.info(The size of sharedData map is: +
envi.getSharedData().size() ); //try to get the counter
--
Here it print 0, if I use
evni.getSharedData().containsKey(counter),
it will return false.
When creating table, I call addCoprocessor() method to add Observer
first, then Endpoint coprocessor. I confirmed that by checking the
hbase log file message. I only have one region for that table during
the run. I confirmed by hbase shell status 'detailed' command.
There is not much example I can find about how to use
getSharedData(), could someone help me here? What is missing in my
simple code? Thanks very much in advance!
Thanks,
Ming
how to use RegionCoprocessorEnvironment getSharedData() to share data among coprocessors?
Hi, all,
I am trying to learn how to share data between two coprocessors. I have one
Observer coprocessor and one Endpoint coprocessor. In the observer, it overload
the prePut/preDelete to maintain a counter. And I want the Endpoint coprocessor
to read that counter and return to client caller. So I want to use the
getSharedData() method in RegionCoprocessorEnvironment, but I cannot make it
work. Could anybody help me here?
In the Observer Coprocessor :
During start(), create the shared object counter:
-
public void start(CoprocessorEnvironment envi) throws IOException {
Env.getSharedData().put(counter, new Long(0) ); //create the counter
-
In the Endpoint coprocessor:
During start(), try to read the shared counter , but failed.
--
public void start(CoprocessorEnvironment envi) throws IOException {
LOG.info(The size of sharedData map is: + envi.getSharedData().size()
); //try to get the counter
--
Here it print 0, if I use evni.getSharedData().containsKey(counter), it will
return false.
When creating table, I call addCoprocessor() method to add Observer first, then
Endpoint coprocessor. I confirmed that by checking the hbase log file message.
I only have one region for that table during the run. I confirmed by hbase
shell status 'detailed' command.
There is not much example I can find about how to use getSharedData(), could
someone help me here? What is missing in my simple code? Thanks very much in
advance!
Thanks,
Ming
RE: HTable or HConnectionManager, how a client connect to HBase?
Thanks, Enis, Your reply is very clear, I finally understand it now. Best Regards, Ming -Original Message- From: Enis Söztutar [mailto:enis@gmail.com] Sent: Thursday, February 19, 2015 10:41 AM To: hbase-user Subject: Re: HTable or HConnectionManager, how a client connect to HBase? It is a bit more complex than that. It is actually a hash of some subset of the configuration properties. See HConnectionKey class if you want to learn more. But the important thing is that with the new style, you do not need to worry anything about these since there is no implicit connection sharing. Everything is explicit now. Enis On Tue, Feb 17, 2015 at 11:50 PM, Serega Sheypak serega.shey...@gmail.com wrote: Hi, Enis Söztutar You've wrote: You are right that the constructor new HTable(Configuration, ..) will share the underlying connection if same configuration object is used. What do it mean the same? is equality checked using reference (java == ) or using equals(Object other) method? 2015-02-18 7:34 GMT+03:00 Enis Söztutar enis@gmail.com: Hi, You are right that the constructor new HTable(Configuration, ..) will share the underlying connection if same configuration object is used. Connection is a heavy weight object, that holds the zookeeper connection, rpc client, socket connections to multiple region servers, master, and the thread pool, etc. You definitely do not want to create multiple connections per process unless you know what you are doing. The model is changed, and the old way of HTable(Configuration, ..) is deprecated because, we want to make the Connection lifecycle management explicit. In the new model, an opened Connection is closed by the user again, and light weight Table instances are obtained from the Connection. Having HTable's share their connections implicitly makes reasoning about it too hard. The new model should be pretty easy to follow. Enis On Sat, Feb 14, 2015 at 6:45 AM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com wrote: Hi, I am using HBase 0.98.6. I learned from this maillist before, that the recommended method to 'connect' to HBase from client is to use HConnectionManager like this: HConnection con=HConnectionManager.createConnection(configuration); HTableInterfacetable = con.getTable(hbase_table1); Instead of HTableInterface table = new HTable(configuration, hbase_table1); I don't quite understand the reason. I was thinking that each time I initialize a HTable instance, it needs to create a new HConnection. And that is expensive. But using the first method, multiple HTable instances can share the same HConnection. That is quite reasonable to me. However, I was reading from some articles on internet that , even if I use the 'new HTable(conf, tbl)' method, if the 'conf' object is the same one, all the HTable instances will still share the same HConnection. I was recently read yet another article and said when using 'new HTable(conf, tbl)', one don't need to use the exactly same 'conf' object (same one in memory). if two 'conf' objects, two different objects are all the same, I mean all attributes of these two are same (for example, created from the same hbase-site.xml and never change) then HTable objects can still share the same HConnection. I also try to read the HTable src code, it is very hard, but it seems to me the last statement is correct: 'HTable will share HConnection, if configuration is all the same'. Sorry for so verbose. My question: If two 'configuration' objects are same, then two HTable object instantiated with them respectively can still share the same HConnection or not? Directly using the 'new HTable()' method. If the answer is 'yes', then why I still need the HConnectionManager to create a shared connection? I am talking about 0.98.6. I googled for days, and even try to read HBase src code, but still get really confused. I try to do some tests also, but since I am too newbie, I don't know how to verify the difference, I really don't know what a HConnection do under the hood. I counted the ZooKeeper client requests, and I found some difference. If this ZooKeeper requests difference is a correct metrics, it means to me that two HTable do not share HConnetion even using same 'configuration' in the constructor. So it confused me more and more Please someone kindly help me for this newbie question and thanks in advance. Thanks, Ming
RE: managing HConnection
Hi, Thank you Serega for the helpful reply and thanks Jneidi for asking this. I have similar confusion. So Serega, when does your application finally close the HConnection? Or the connection is NEVER closed as long as your application is running? Is it OK to NOT close the HConnection and the application exit directly? My application is a long-running service, accept user request and do CRUD to hbase. So I would like to use your model here. But, is it reasonable to keep that HConnection open a very long time, for example months? Is there any potential problem I need to take care? Also as David Chen asked, if all threads share same HConnection, it may has limitation to support high throughput, so a pool of Connections maybe better? Thanks, Ming -Original Message- From: Serega Sheypak [mailto:serega.shey...@gmail.com] Sent: Wednesday, February 04, 2015 1:02 AM To: user Subject: Re: managing HConnection Hi, guys from group helped me a lot. I did solve pretty the same problem (CRUD web-app) 1. Use single instance of HConnection per application. 2. Instantiate it once. 3. create HTable instance for each CRUD operation and safely close it (try-catch-finally). Use the same HConnection to create any HTable for CRUD operation. 4. DO NOT close HConnection after CRUD operation I have logic controllers which get HConnection injection in HttpServlet.init method. So I have 5 HConnection instances per application created during servlet initialization 2015-02-03 18:12 GMT+03:00 Ted Yu yuzhih...@gmail.com: Please see '61.1. Cluster Connections' under http://hbase.apache.org/book.html#architecture.client Cheers On Tue, Feb 3, 2015 at 6:47 AM, sleimanjneidi jneidi.slei...@gmail.com wrote: Hi all, I am using hbase-0.98.1-cdh5.1.4 client and I am a bit confused by the documentation of HConnection. The document says the following: HConnection instances can be shared. Sharing is usually what you want because rather than each HConnection instance having to do its own discovery of regions out on the cluster, instead, all clients get to share the one cache of locations. HConnectionManager does the sharing for you if you go by it getting connections. Sharing makes cleanup of HConnections awkward. . So now I have a simple question: Can I share the same HConnection instance in my entire application? And write some magic code to know when to close or never close at all? Or I have to create an instance and close it every time I do a CRUD operation ? Many thanks
HTable or HConnectionManager, how a client connect to HBase?
Hi, I am using HBase 0.98.6. I learned from this maillist before, that the recommended method to 'connect' to HBase from client is to use HConnectionManager like this: HConnection con=HConnectionManager.createConnection(configuration); HTableInterfacetable = con.getTable(hbase_table1); Instead of HTableInterface table = new HTable(configuration, hbase_table1); I don't quite understand the reason. I was thinking that each time I initialize a HTable instance, it needs to create a new HConnection. And that is expensive. But using the first method, multiple HTable instances can share the same HConnection. That is quite reasonable to me. However, I was reading from some articles on internet that , even if I use the 'new HTable(conf, tbl)' method, if the 'conf' object is the same one, all the HTable instances will still share the same HConnection. I was recently read yet another article and said when using 'new HTable(conf, tbl)', one don't need to use the exactly same 'conf' object (same one in memory). if two 'conf' objects, two different objects are all the same, I mean all attributes of these two are same (for example, created from the same hbase-site.xml and never change) then HTable objects can still share the same HConnection. I also try to read the HTable src code, it is very hard, but it seems to me the last statement is correct: 'HTable will share HConnection, if configuration is all the same'. Sorry for so verbose. My question: If two 'configuration' objects are same, then two HTable object instantiated with them respectively can still share the same HConnection or not? Directly using the 'new HTable()' method. If the answer is 'yes', then why I still need the HConnectionManager to create a shared connection? I am talking about 0.98.6. I googled for days, and even try to read HBase src code, but still get really confused. I try to do some tests also, but since I am too newbie, I don't know how to verify the difference, I really don't know what a HConnection do under the hood. I counted the ZooKeeper client requests, and I found some difference. If this ZooKeeper requests difference is a correct metrics, it means to me that two HTable do not share HConnetion even using same 'configuration' in the constructor. So it confused me more and more Please someone kindly help me for this newbie question and thanks in advance. Thanks, Ming
RE: Given a Put object, is there any way to change the timestamp of it?
Thanks Ted!
This is exactly what I need.
This will be a memory copy, but it solves my problem. Hope HBase can provide a
setTimeStamp() method in future release.
Best Regards,
Ming
-Original Message-
From: Ted Yu [mailto:yuzhih...@gmail.com]
Sent: Wednesday, January 21, 2015 11:30 AM
To: user@hbase.apache.org
Subject: Re: Given a Put object, is there any way to change the timestamp of it?
bq. I have no way to get ALL its cells out Mutation has the following method:
public NavigableMapbyte [], ListCell getFamilyCellMap() {
FYI
On Tue, Jan 20, 2015 at 5:43 PM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com
wrote:
Hello, there,
I am developing a coprocessor under HBase 0.98.6. The client send a
Put object to the coprocessor in Protobuf, when the coprocessor
receive the message , it invokes ProtobufUtil.toPut to convert it to a
Put object. Do various checking and then put it into HBase table.
Now, I get a requirement to change the timestamp of that Put object,
but I found no way to do this.
I was first try to generate a new Put object with a new timestamp, and
try to copy the old one into this new object. But I found given a Put
object, I have no way to get ALL its cells out if I don't know the
column family and column qualifier name in advance. In my case, those
CF/Column names are random as user defined. So I stuck here. Could
anyone have idea how to workaround this?
The Mutation class has getTimestamp() method but no setTimestamp(). I
wish there is a setTimestamp() for it. Is there any reason it is not
provided? I hope in future release Mutation can expose a
setTimestamp() method, is it possible? If so, my job will get much easier...
Thanks,
Ming
Given a Put object, is there any way to change the timestamp of it?
Hello, there, I am developing a coprocessor under HBase 0.98.6. The client send a Put object to the coprocessor in Protobuf, when the coprocessor receive the message , it invokes ProtobufUtil.toPut to convert it to a Put object. Do various checking and then put it into HBase table. Now, I get a requirement to change the timestamp of that Put object, but I found no way to do this. I was first try to generate a new Put object with a new timestamp, and try to copy the old one into this new object. But I found given a Put object, I have no way to get ALL its cells out if I don't know the column family and column qualifier name in advance. In my case, those CF/Column names are random as user defined. So I stuck here. Could anyone have idea how to workaround this? The Mutation class has getTimestamp() method but no setTimestamp(). I wish there is a setTimestamp() for it. Is there any reason it is not provided? I hope in future release Mutation can expose a setTimestamp() method, is it possible? If so, my job will get much easier... Thanks, Ming
Durability of in-memory column family
Hi, all, I want to use a column family to save some runtime data, which is small so I set that CF in memory to increase the performance, and the user data is still save in a normal CF. My question is: will the data in the in-memory column family get lost, if there is a failure of regionServer? Or in other words, is the data in an in-memory CF as safe as in an ordinary CF? No difference? I could do test myself, but it needs some time, so I would like to be lazy and ask for help here :) If someone happened to know the answer, thanks in advance! Thanks, Ming
how to tell there is a OOM in regionserver
Hi, all, Recently, one of our HBase 0.98.5 instance meet with issues: when run some specific workload, all region servers will suddenly shut down at same time, but master is still running. When I check the log, in master log, I can see messages like 2014-12-01 08:28:11,072 DEBUG [main-EventThread] master.ServerManager: Added=n008.cluster,60020,1417413986550 to dead servers, submitted shutdown handler to be executed meta=false And on n008, regionserver log file, there is no ERROR message, the last log entry looks very like a ZooKeeper startup message. The log just stopped with that last ZooKeeper startup message, and the Region Server process was gone when we check with 'jps'. We then increased the heap size of regionserver, and it work fine. RegionServer no longer disappear. So we doubt there was a Out Of Memory issue, so the region server processes are killed. But my questions are: 1. What log message will indicate there is a OOM? Since the region server is 'kill -9', so I think there is no message can tell this. 2. If there is no typical log message about OOM, then how can an admin make sure there is a region server OOM happened? We just guess, but can not make sure. We hope there is a method to tell OOM occured for sure. 3. Does the Zookeeper message appears every time with RegionServer OOM (if it is a OOM). Or it is just a random event just in our system? So in sum, I want to know what is the typical clue that people can make sure there is a OOM issue in HBase region server? Thank you, Ming
RE: how to tell there is a OOM in regionserver
Thank you both! Yes, I can see there is the '.out' file with clear proof of process was 'killed'. So we can prove this issue now! And it is also true that we must rely on JVM itself for proof that the kill operation is due to OOM. Thank you both, this is a very good learning. Thanks, Ming -Original Message- From: Bharath Vissapragada [mailto:bhara...@cloudera.com] Sent: Tuesday, December 02, 2014 2:00 PM To: hbase-user Subject: Re: how to tell there is a OOM in regionserver I agree with Otis' response. Adding a few more details, there is a .out file in the logs/ directory, that is the stdout for each of these daemons and incase of an OOM crash, it prints something like this # java.lang.OutOfMemoryError: Java heap space # -XX:OnOutOfMemoryError=kill -9 %p # Executing /bin/sh -c kill -9 pid... On Tue, Dec 2, 2014 at 11:06 AM, Otis Gospodnetic otis.gospodne...@gmail.com wrote: Hi Ming, 1) There typically is an OOM message from the JVM itself 2) I would monitor the server instead of relying on log messages mentioning OOMs. For example, in SPM http://sematext.com/spm/ we have hearbeat alerts that tell us when we stop hearing from RegionServers and other types of servers. It also helps when servers simply die for reasons other than OOM. 3) You could (should?) monitor individual memory pools and possibly set alerts or anomaly detection on those. If you have that, if there was an OOM, you will typically see one of the memory pools approach 100% utilization. I personally really like this report in SPM because it gives a bit more insight than just heap size/utilization. So I'd point the admin to this sort of monitoring report. 4) High GC counts/time, or jump in those metrics, and then typically also jump in CPU usage is what often precedes OOMs. Otis -- Monitoring * Alerting * Anomaly Detection * Centralized Log Management Solr Elasticsearch Support * http://sematext.com/ On Tue, Dec 2, 2014 at 12:22 AM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com wrote: Hi, all, Recently, one of our HBase 0.98.5 instance meet with issues: when run some specific workload, all region servers will suddenly shut down at same time, but master is still running. When I check the log, in master log, I can see messages like 2014-12-01 08:28:11,072 DEBUG [main-EventThread] master.ServerManager: Added=n008.cluster,60020,1417413986550 to dead servers, submitted shutdown handler to be executed meta=false And on n008, regionserver log file, there is no ERROR message, the last log entry looks very like a ZooKeeper startup message. The log just stopped with that last ZooKeeper startup message, and the Region Server process was gone when we check with 'jps'. We then increased the heap size of regionserver, and it work fine. RegionServer no longer disappear. So we doubt there was a Out Of Memory issue, so the region server processes are killed. But my questions are: 1. What log message will indicate there is a OOM? Since the region server is 'kill -9', so I think there is no message can tell this. 2. If there is no typical log message about OOM, then how can an admin make sure there is a region server OOM happened? We just guess, but can not make sure. We hope there is a method to tell OOM occured for sure. 3. Does the Zookeeper message appears every time with RegionServer OOM (if it is a OOM). Or it is just a random event just in our system? So in sum, I want to know what is the typical clue that people can make sure there is a OOM issue in HBase region server? Thank you, Ming -- Bharath Vissapragada http://www.cloudera.com
RE: YCSB load failed because hbase region too busy
Hi, Louis, Sorry I cannot help here, I am just very curious about your YCSB test results. It is not easy to find latest YCSB testing result on internet. In my own hbase env, my testing result is 'update is always better than read' and 'Scan is slightly better than update'. I tried many times with various hbase configuration and tested on two different hardware and get same result, although the absolute number is different, but 'random single write is always better than random single read' and '100-length scan is better than single write', very stable result. I modified the workload to add a pure update workload, it will be very appreciated that you can do that test too (By setting readproportion to 0 and updateproportion to 1), I also change the CoreWorkload.java doTransactionScan() to always do a 100-len scan instead of a random len scan, so I can easily get how many rows scanned and compare to pure write and pure read result. If it is not proper to show the absolute number of your test result, could you at least tell me if in your test 'read is better than write' or 'write is better than read', and by how much? I asked a few times in this mailing list, and I think people explained to me that it is possible that write is better than read in HBase, but I still want to know if this is common or just in my env. And I thought you may meet the 'stuck' issue mentioned in http://hbase.apache.org/book.html in section 9.7.7.7.1.1 , but I am not sure. Happy to know how you solve the issue later. And as Ram and Qiang,Tian mentioned, you can only 'alleviate' the issue by increasing the knob but if you give hbase too much pressure, it will not work well sooner or later. Everyone has its own limitation :-) Thanks, Ming -Original Message- From: louis hust [mailto:louis.h...@gmail.com] Sent: Tuesday, November 25, 2014 9:44 PM To: user@hbase.apache.org Subject: Re: YCSB load failed because hbase region too busy hi ram, thanks for help, i just do a test for bucket cache, in product env, we will follow your suggestion Sent from my iPhone On 2014年11月25日, at 20:36, ramkrishna vasudevan ramkrishna.s.vasude...@gmail.com wrote: Your write ingest is too high. You have to control that by first adding more nodes and ensuring that you have a more distributed load. And also try with the changing the hbase.hstore.blockingStoreFiles. Even changing the above value if your write ingest is so high such that if it can reach this configured value again you can see blocking writes. Regards RAm On Tue, Nov 25, 2014 at 2:20 PM, Qiang Tian tian...@gmail.com wrote: in your log: 2014-11-25 13:31:35,048 WARN [MemStoreFlusher.13] regionserver.MemStoreFlusher: Region usertable2,user8289,1416889268210.7e8fd83bb34b155bd0385aa63124a875. has too many store files; delaying flush up to 9ms please see my original reply...you can try increasing hbase.hstore.blockingStoreFiles, also you have only 1 RS and you split to 100 regionsyou can try 2 RS with 20 regions. On Tue, Nov 25, 2014 at 3:42 PM, louis.hust louis.h...@gmail.com wrote: yes, the stack trace like below: 2014-11-25 13:35:40:946 4260 sec: 232700856 operations; 28173.18 current ops/sec; [INSERT AverageLatency(us)=637.59] 2014-11-25 13:35:50:946 4270 sec: 232700856 operations; 0 current ops/sec; 14/11/25 13:35:59 INFO client.AsyncProcess: #14, table=usertable2, attempt=10/35 failed 109 ops, last exception: org.apache.hadoop.hbase.RegionTooBusyException: org.apache.hadoop.hbase.RegionTooBusyException: Above memstore limit, regionName=usertable2,user8289,1416889268210.7e8fd83bb34b155bd0385aa6 3124a875., server=l-hbase10.dba.cn1.qunar.com,60020,1416889404151, memstoreSize=536886800, blockingMemStoreSize=536870912 at org.apache.hadoop.hbase.regionserver.HRegion.checkResources(HRegion.j ava:2822) at org.apache.hadoop.hbase.regionserver.HRegion.batchMutate(HRegion.java :2234) at org.apache.hadoop.hbase.regionserver.HRegion.batchMutate(HRegion.java :2201) at org.apache.hadoop.hbase.regionserver.HRegion.batchMutate(HRegion.java :2205) at org.apache.hadoop.hbase.regionserver.HRegionServer.doBatchOp(HRegionS erver.java:4253) at org.apache.hadoop.hbase.regionserver.HRegionServer.doNonAtomicRegionM utation(HRegionServer.java:3469) at org.apache.hadoop.hbase.regionserver.HRegionServer.multi(HRegionServe r.java:3359) at org.apache.hadoop.hbase.protobuf.generated.ClientProtos$ClientService $2.callBlockingMethod(ClientProtos.java:29503) at org.apache.hadoop.hbase.ipc.RpcServer.call(RpcServer.java:2012) at org.apache.hadoop.hbase.ipc.CallRunner.run(CallRunner.java:98) at org.apache.hadoop.hbase.ipc.SimpleRpcScheduler.consumerLoop(SimpleRpc Scheduler.java:160) at org.apache.hadoop.hbase.ipc.SimpleRpcScheduler.access$000(SimpleRpcSc heduler.java:38) at
RE: What is proper way to make a hbase connection? using HTable (conf,tbl) or createConnection? Zookeeper session run out.
Thank you Bharath,
This is a very helpful reply! I will share the connection between two threads.
Simply put, HTable is not safe for multi-thread, is this true? In
multi-threads, one must use HConnectionManager.
Thanks,
Ming
-Original Message-
From: Bharath Vissapragada [mailto:bhara...@cloudera.com]
Sent: Monday, November 24, 2014 4:52 PM
To: hbase-user
Subject: Re: What is proper way to make a hbase connection? using HTable
(conf,tbl) or createConnection? Zookeeper session run out.
Hi Ming,
HConnection connection = HConnectionManager.createConnection(conf);
HTableInterface table = connection.getTable(mytable); table.get(...); /
table.put(...);
Is the correct way to use. However
HConnectionManager.createConnection(conf) gives you a shared HConnection
which you can reuse across multiple threads that get their own
conn.getTable() and do their puts/gets.
Thanks,
Bharath
On Mon, Nov 24, 2014 at 2:03 PM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com
wrote:
Hello,
I am using HBase 0.98.5. In example hbase client programs, some use
createConnection() and some use HTable() directly, I found they behave
different, I wrote two simple tests program using these two different
methods, each program will start two threads and do simple put. and I
found:
One program will start only 1 zookeeper sessions shared in two threads
while another will start 2 zookeeper sessions. So I don't know why the
program using createConnection will do more zookeeper requests than
simply use HTable. Is it possible to use createConnection in two
threads but share the same zookeeper session?
Here is details:
Demo1 will make two zookeeper sessions, seems two HBase connections;
but
Demo2 will only make one zookeeper session. My real program is using
createConnection in multiple threads as in demo1, since I have a very
small zookeeper , it only allows 60 concurrent sessions, so my program
always fail when there are hundreds of threads started. But I saw if
using HTable directly, it will only consume 1 zookeeper session. But
it will change a lot in my current program, so I wish there is a way
to use createConnection and behave same as using HTable, is it possible?
Source code:
Demo1.java
class ClientThread extends Thread
{
public static Configuration configuration;
static {
configuration = HBaseConfiguration.create();
}
public void run()
{
try {
System.out.println(start insert data ..);
HConnection
con=HConnectionManager.createConnection(configuration);
HTable table = (HTable)con.getTable(hbase_table1);
Put put = new Put(1.getBytes());
put.add(c1.getBytes(), null, baidu.getBytes());
put.add(c2.getBytes(), null, http://www.baidu.com1
.getBytes());
try {
table.put(put);
} catch (IOException e) {
e.printStackTrace();
}
System.out.println(end insert data ..);
}
catch (Exception e) {
}
}
}
public class demo1 {
public static void main(String[] args) throws Exception {
Thread t1=new ClientThread();
Thread t2=new ClientThread();
t1.start();
t2.start();
}
}
Demo2.java
class ClientThread1 extends Thread
{
public static Configuration configuration;
static {
configuration = HBaseConfiguration.create();
}
public void run()
{
System.out.println(start insert data ..);
try {
HTableInterface table = new
HTable(configuration, hbase_table1);
Put put = new Put(1.getBytes());
put.add(c1.getBytes(), null, baidu.getBytes());
put.add(c2.getBytes(), null,
http://www.baidu.com1.getBytes());
table.put(put);
} catch (Exception e) {
e.printStackTrace();
}
System.out.println(end insert data ..);
}
}
public class demo2 {
public static void main(String[] args) throws Exception {
Thread t1=new ClientThread1();
Thread t2=new ClientThread1();
t1.start();
t2.start();
}
}
This should be a very basic question, sorry, I really did some search
but cannot find any good explaination. Any help will be very appreciated.
Thanks,
Ming
--
Bharath Vissapragada
http://www.cloudera.com
RE: how to explain read/write performance change after modifying the hfile.block.cache.size?
Thank you Nick, I will increase the heap size. The workstation is a development workstation. People use it to code and build software, and do unit testing. And I use 'who' and 'ps' to make sure it is exclusive to me when I did the test, but it is not accurate. You are right, the benchmark is not accepted in that env. I just use this to 'develop' a benchmark tool, since I wrote a new YCSB driver for your system on HBase, but the following test is using native HBase driver and test on Hbase. For the real benchmark, we will run it on a separated cluster later. But I hope the data at least make sensor even on a shared env. The heap configuration is something I really need to check , thank you. Best Regards, Ming -Original Message- From: Nick Dimiduk [mailto:ndimi...@gmail.com] Sent: Saturday, November 22, 2014 5:57 AM To: user@hbase.apache.org Cc: lars hofhansl Subject: Re: how to explain read/write performance change after modifying the hfile.block.cache.size? 400mb blockcache? Ouch. What's your hbase-env.sh? Have you configured a heap size? My guess is you're using the un configured default of 1G. Should be at least 8G, and maybe more like 30G with this kind of host. How many users are sharing it and with what kinds of tasks? If there's no IO isolation between processes, I suspect your benchmarks will be worthless on this shared environment. -n On Friday, November 21, 2014, Liu, Ming (HPIT-GADSC) ming.l...@hp.com wrote: Thank you Lars, There must be something wrong with my testing yesterday. I cannot reproduce the issue anymore. Now, changing the cache.size from 0 to 0.4 will not slow down the write perf dramatically, but still will slow down write (Yesterday I saw a 7x slowdown, today it is about 1.3x slowdown which is acceptable for me). As Ted pointed out, it is possible that memstore cannot get enough memory when more RAM give to block cache so it flush more frequently, but I really need more reading to understand when memstore will flush. And one thing I noticed is when I restart hbase for the very first test, the performance is best, then the second time, it is slower, both read and write, and slower and slower in the following tests and get to a stable point after about 3 or 4 times, in each run I will read 5,000,000 rows and update 5,000,000 rows. There are too many factors affect the read/write OPS in hbase... My purpose is to find a proper way to evaluate performance, since we are going to change something in hbase and it is good to have a base benchmark so we can compare the performance after change. So I must make sure the perf test itself make sense and should be trusted. I saw an entry in the log may help to see the cache settings in my system: hfile.LruBlockCache: Total=373.54 MB, free=13.13 MB, max=386.68 MB, blocks=5655, accesses=17939947, hits=14065015, hitRatio=78.40%, , cachingAccesses=17934857, cachingHits=14064420, cachingHitsRatio=78.42%, evictions=15646, evicted=3861343, evictedPerRun=246.7942657470703 My testing environment is a workstation with 12 core CPU, 96G memory and 1.7 T disk. But it is a shared workstation, many users share it and I started hbase in standalone mode with hbase-site.xml as below : configuration property namehbase.rootdir/name valuehdfs://localhost:24400/hbase/value /property property namehbase.zookeeper.property.dataDir/name valuehdfs://localhost:24400/zookeeper/value /property property namehbase.master.port/name value24560/value /property property namehbase.master.info.port/name value24561/value /property property namehbase.regionserver.port/name value24562/value /property property namehbase.regionserver.info.port/name value24563/value /property property namehbase.zookeeper.peerport/name value24567/value /property property namehbase.zookeeper.leaderport/name value24568/value /property property namehbase.zookeeper.property.clientPort/name value24570/value /property property namehbase.rest.port/name value24571/value /property property namehbase.client.scanner.caching/name value100/value /property property namehbase.client.scanner.timeout.period/name value6/value /property property namehbase.bulkload.staging.dir/name valuehdfs://localhost:24400/hbase-staging/value /property property namehbase.snapshot.enabled/name valuetrue/value /property property namehbase.master.distributed.log.splitting/name valuefalse/value /property property namezookeeper.session.timeout/name value9000/value:-) I just want to make sure never timeout here, since I get timeout so many times... /property property namehfile.block.cache.size/name value0.4/value /property /configuration [liuliumi@ YCSB]$ free
how to explain read/write performance change after modifying the hfile.block.cache.size?
Hello, all, I am playing with YCSB to test HBase performance. I am using HBase 0.98.5. I tried to adjust the hfile.block.cache.size to see the difference, when I set hfile.block.cache.size to 0, read performance is very bad, but write performance is very very very good; when I set hfile.block.cache.size to 0.4, read is better, but write performance drop dramatically. I disable the client side writebuffer already. This is hard to understand for me: The HBase guide just said hfile.block.cache.size setting is about how much memory used as block cache used by StoreFile. I have no idea of how HBase works internally. Typically, it is easy to understand that increase the size of cache should help the read, but why it will harm the write operation? The write performance down from 30,000 to 4,000 for your reference, just by changing the hfile.block.cache.size from 0 to 0.4. Could anyone give me a brief explanation about this observation or give me some advices about what to study to understand what is block cache used for? Another question: HBase write will first write to WAL then to memstore. Will the write to WAL go to disk directly before hbase write memstore, a sync operation or it is possible that write to WAL is still buffered somewhere when hbase put the data into the memstore? Reading src code may cost me months, so a kindly reply will help me a lot... ... Thanks very much! Best Regards, Ming
RE: how to explain read/write performance change after modifying the hfile.block.cache.size?
Thank you Ted, It is a great explanation. You are always very helpful ^_^ I will study the link carefully. Thanks, Ming -Original Message- From: Ted Yu [mailto:yuzhih...@gmail.com] Sent: Friday, November 21, 2014 1:32 AM To: user@hbase.apache.org Subject: Re: how to explain read/write performance change after modifying the hfile.block.cache.size? When block cache size increases from 0 to 0.4, the amount of heap given to memstore decreases. This would slow down the writes. Please see: http://hbase.apache.org/book.html#store.memstore For your second question, see this thread: http://search-hadoop.com/m/DHED4TEvBy1/lars+hbase+hflushsubj=Re+Clarifications+on+HBase+Durability Cheers On Thu, Nov 20, 2014 at 8:05 AM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com wrote: Hello, all, I am playing with YCSB to test HBase performance. I am using HBase 0.98.5. I tried to adjust the hfile.block.cache.size to see the difference, when I set hfile.block.cache.size to 0, read performance is very bad, but write performance is very very very good; when I set hfile.block.cache.size to 0.4, read is better, but write performance drop dramatically. I disable the client side writebuffer already. This is hard to understand for me: The HBase guide just said hfile.block.cache.size setting is about how much memory used as block cache used by StoreFile. I have no idea of how HBase works internally. Typically, it is easy to understand that increase the size of cache should help the read, but why it will harm the write operation? The write performance down from 30,000 to 4,000 for your reference, just by changing the hfile.block.cache.size from 0 to 0.4. Could anyone give me a brief explanation about this observation or give me some advices about what to study to understand what is block cache used for? Another question: HBase write will first write to WAL then to memstore. Will the write to WAL go to disk directly before hbase write memstore, a sync operation or it is possible that write to WAL is still buffered somewhere when hbase put the data into the memstore? Reading src code may cost me months, so a kindly reply will help me a lot... ... Thanks very much! Best Regards, Ming
RE: how to explain read/write performance change after modifying the hfile.block.cache.size?
on the table/column family you're writing to (like bloom filters, etc). That would help us diagnose this. -- Lars From: Ted Yu yuzhih...@gmail.com To: user@hbase.apache.org user@hbase.apache.org Sent: Thursday, November 20, 2014 9:32 AM Subject: Re: how to explain read/write performance change after modifying the hfile.block.cache.size? When block cache size increases from 0 to 0.4, the amount of heap given to memstore decreases. This would slow down the writes. Please see: http://hbase.apache.org/book.html#store.memstore For your second question, see this thread: http://search-hadoop.com/m/DHED4TEvBy1/lars+hbase+hflushsubj=Re+Clarifications+on+HBase+Durability Cheers On Thu, Nov 20, 2014 at 8:05 AM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com wrote: Hello, all, I am playing with YCSB to test HBase performance. I am using HBase 0.98.5. I tried to adjust the hfile.block.cache.size to see the difference, when I set hfile.block.cache.size to 0, read performance is very bad, but write performance is very very very good; when I set hfile.block.cache.size to 0.4, read is better, but write performance drop dramatically. I disable the client side writebuffer already. This is hard to understand for me: The HBase guide just said hfile.block.cache.size setting is about how much memory used as block cache used by StoreFile. I have no idea of how HBase works internally. Typically, it is easy to understand that increase the size of cache should help the read, but why it will harm the write operation? The write performance down from 30,000 to 4,000 for your reference, just by changing the hfile.block.cache.size from 0 to 0.4. Could anyone give me a brief explanation about this observation or give me some advices about what to study to understand what is block cache used for? Another question: HBase write will first write to WAL then to memstore. Will the write to WAL go to disk directly before hbase write memstore, a sync operation or it is possible that write to WAL is still buffered somewhere when hbase put the data into the memstore? Reading src code may cost me months, so a kindly reply will help me a lot... ... Thanks very much! Best Regards, Ming
RE: Is it possible that HBase update performance is much better than read in YCSB test?
Thank you Andrew, this is an excellent answer, I get it now. I will try your hbase client for a 'fair' test :-) Best Regards, Ming -Original Message- From: Andrew Purtell [mailto:apurt...@apache.org] Sent: Thursday, November 13, 2014 2:08 AM To: user@hbase.apache.org Cc: DeRoo, John Subject: Re: Is it possible that HBase update performance is much better than read in YCSB test? Try this HBase YCSB client instead: https://github.com/apurtell/ycsb/tree/new_hbase_client The HBase YCSB driver in the master repo holds on to one HTable instance per driver thread. We accumulate writes into a 12MB write buffer before flushing them en masse. This is why the behavior you are seeing confounds your expectations. It's not correct behavior IMHO. YCSB wants to measure the round trip of every op, not the non-cost of local caching. Worse, if we have a lot of driver threads accumulating 12MB of edits more or less at the same rate, then we will flush these buffers more or less at the same time and stampede the cluster, which leads to deep valleys in observed write performance of 30-60 seconds or longer. On Tue, Nov 11, 2014 at 8:40 PM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com wrote: Hi, all, I am trying to use YCSB to test on our HBase 0.98.5 instance and got a strange result: update is 6x better than read. It is just an exercise, so the HBase is running in a workstation in standalone mode. I modified the workloada shipped with YCSB into two new workloads: workloadr and workloadu, where workloadr is do 100% read operation and workloadu is do 100% update operation. At the bottom is the workloadr and workloadu config files for your reference. I found out that the read performance is much worse than the update performance, read is about 6000: YCSB Client 0.1 Command line: -db com.yahoo.ycsb.db.HBaseClient -P workloads/workloadr -p columnfamily=family -s -t [OVERALL], RunTime(ms), 16565.0 [OVERALL], Throughput(ops/sec), 6036.824630244491 And the update performance is about 36000, 6x better than read. YCSB Client 0.1 Command line: -db com.yahoo.ycsb.db.HBaseClient -P workloads/workloadu -p columnfamily=family -s -t [OVERALL], RunTime(ms), 2767.0 [OVERALL], Throughput(ops/sec), 36140.22406938923 Is this possible? IMHO, read should be faster than update. Maybe I am wrong in the workload file? Or there is a possibility that update is faster than read? I don't find a YCSB mailing list, if anyone knows, please give me a link, so I can also ask question on that mailing list. But is it possible that put is faster than get in hbase? If not, the result must be wrong and I need to debug the YCSB code to figure out what is going wrong. Workloadr: recordcount=10 operationcount=10 workload=com.yahoo.ycsb.workloads.CoreWorkload readallfields=true readproportion=1 updateproportion=0 scanproportion=0 insertproportion=0 requestdistribution=zipfian workloadu: recordcount=10 operationcount=10 workload=com.yahoo.ycsb.workloads.CoreWorkload readallfields=true readproportion=0 updateproportion=1 scanproportion=0 insertproportion=0 requestdistribution=zipfian Thanks, Ming -- Best regards, - Andy Problems worthy of attack prove their worth by hitting back. - Piet Hein (via Tom White)
Is it possible that HBase update performance is much better than read in YCSB test?
Hi, all, I am trying to use YCSB to test on our HBase 0.98.5 instance and got a strange result: update is 6x better than read. It is just an exercise, so the HBase is running in a workstation in standalone mode. I modified the workloada shipped with YCSB into two new workloads: workloadr and workloadu, where workloadr is do 100% read operation and workloadu is do 100% update operation. At the bottom is the workloadr and workloadu config files for your reference. I found out that the read performance is much worse than the update performance, read is about 6000: YCSB Client 0.1 Command line: -db com.yahoo.ycsb.db.HBaseClient -P workloads/workloadr -p columnfamily=family -s -t [OVERALL], RunTime(ms), 16565.0 [OVERALL], Throughput(ops/sec), 6036.824630244491 And the update performance is about 36000, 6x better than read. YCSB Client 0.1 Command line: -db com.yahoo.ycsb.db.HBaseClient -P workloads/workloadu -p columnfamily=family -s -t [OVERALL], RunTime(ms), 2767.0 [OVERALL], Throughput(ops/sec), 36140.22406938923 Is this possible? IMHO, read should be faster than update. Maybe I am wrong in the workload file? Or there is a possibility that update is faster than read? I don't find a YCSB mailing list, if anyone knows, please give me a link, so I can also ask question on that mailing list. But is it possible that put is faster than get in hbase? If not, the result must be wrong and I need to debug the YCSB code to figure out what is going wrong. Workloadr: recordcount=10 operationcount=10 workload=com.yahoo.ycsb.workloads.CoreWorkload readallfields=true readproportion=1 updateproportion=0 scanproportion=0 insertproportion=0 requestdistribution=zipfian workloadu: recordcount=10 operationcount=10 workload=com.yahoo.ycsb.workloads.CoreWorkload readallfields=true readproportion=0 updateproportion=1 scanproportion=0 insertproportion=0 requestdistribution=zipfian Thanks, Ming
RE: is there a HBase 0.98 hdfs directory structure introduction?
Thanks Ted for the short but very useful reply! ^_^
It is clear now.
-Original Message-
From: Ted Yu [mailto:yuzhih...@gmail.com]
Sent: Monday, November 03, 2014 11:30 AM
To: user@hbase.apache.org
Subject: Re: is there a HBase 0.98 hdfs directory structure introduction?
In 0.98, you would find your table under the following directory:
$roodir/{namespace}/table
If you don't specify namespace at table creation time, 'default' namespace
would be used.
Cheers
On Sun, Nov 2, 2014 at 7:16 PM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com
wrote:
Hi, all,
I have a program to calculate the disk usage of hbase per table in
hbase 0.94. I used to use the hadoop fs -du command against
directory $roodir/table as the size a table uses, as described in
HBase's ref
guide: http://hbase.apache.org/book/trouble.namenode.html .
However, when we upgraded to HBase 0.98, the directory structure
changed a lot. Yes, I can use ls to find the table directory and
modify the program myself, but I wish there will be a good reference
to learn more details about the change. The document on hbase official
web site seems not updated. So can anyone help to briefly introduce
the new directory structure or give me a link? It will be good to know
what each directory is for.
Thanks,
Ming
is there a HBase 0.98 hdfs directory structure introduction?
Hi, all, I have a program to calculate the disk usage of hbase per table in hbase 0.94. I used to use the hadoop fs -du command against directory $roodir/table as the size a table uses, as described in HBase's ref guide: http://hbase.apache.org/book/trouble.namenode.html . However, when we upgraded to HBase 0.98, the directory structure changed a lot. Yes, I can use ls to find the table directory and modify the program myself, but I wish there will be a good reference to learn more details about the change. The document on hbase official web site seems not updated. So can anyone help to briefly introduce the new directory structure or give me a link? It will be good to know what each directory is for. Thanks, Ming
when will hbase create the zookeeper znode 'root-region-server’ is created? Hbase 0.94
Hello,
I am trying to debug a coprocessor code on hbase 0.94.24, which seems to work
well on 0.94.5, but I cannot make it work on 0.94.24.
Here is the copy of some coprocessor init code:
public class TestEndpoint implements TestIface, HTableWrapper {
…
@Override
public void start(CoprocessorEnvironment env) throws IOException {
this.env = env;
conf = env.getConfiguration();
HBaseAdmin admin = new HBaseAdmin(conf);
if (!admin.tableExists(SOME_TABLE)) {
//do something if this table is not there
…
When hbase starts, in the log file I noticed that when regionserver load this
coprocessor, it will hang there inside the admin.tableExists() function. That
API will try to access the zookeeper znode ‘root-region-server’, so I start the
‘hbase zkcli’ and run ‘ls /hbase’ at that time, and I found the znode
‘root-region-server’ is not created. Since the coprocessor want to access a
table, it must look up the –ROOT- region, which location is saved in that
znode, but that znode is not there. Then it hangs there for ever. If I disable
this coprocessor, hbase can start good and I can see ‘root-region-server’ znode
created there.
This coprocessor code is claimed to work well on 0.94.5, so I am wondering if
there is something changed about the sequence of ‘load coprocessor’ and the
‘create root-region-server znode’ in hbase 0.94 serials after 0.94.5?
So my basic question is : when znode ‘root-region-server’ is created? Who
create it? And is there any fixed timing sequence of this initialization
between the coprocessor loading time?
By the way, I cannot find anywhere to download a 0.94.5 hbase source code, can
anyone tell me if there is somewhere I can find it?
I know old version is obsolete , but this is not for production, but for
research, so please help me if you have any idea. Thanks very much in advance.
Thanks,
Ming
RE: when will hbase create the zookeeper znode 'root-region-server’ is created? Hbase 0.94
Thanks Ted and Sean, It is great to find the archives, why I did not find them for so long time... :-) The original coprocessor author reply me yesterday, it is my fault. In fact, that coprocessor is loaded after the hbase startup, not during hbase regionserver startup time. In their client application code , they invoke addCoprocessor() to load the coprocessor at run-time. So I should not put them into the hbase-site.xml. When loaded later, the HBaseAdmin.tableExist() can work, since at that time, all initialization is done. So this is not a version issue, it can work in 0.94.5 and 0.94.24 as well. Thank you all for the help. Ming -Original Message- From: Ted Yu [mailto:yuzhih...@gmail.com] Sent: Thursday, October 16, 2014 10:29 PM To: user@hbase.apache.org Subject: Re: when will hbase create the zookeeper znode 'root-region-server’ is created? Hbase 0.94 Ming: The tar ball in the archive contains source code. See example below: $ tar tzvf hbase-0.94.5.tar.gz | grep '\.java' | grep Assignment -rw-r--r-- 0 jenkins jenkins 47982 Feb 7 2013 hbase-0.94.5/src/test/java/org/apache/hadoop/hbase/master/TestAssignmentManager.java -rw-r--r-- 0 jenkins jenkins 136645 Feb 7 2013 hbase-0.94.5/src/main/java/org/apache/hadoop/hbase/master/AssignmentManager.java FYI On Thu, Oct 16, 2014 at 7:18 AM, Sean Busbey bus...@cloudera.com wrote: On Thu, Oct 16, 2014 at 2:18 AM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com wrote: By the way, I cannot find anywhere to download a 0.94.5 hbase source code, can anyone tell me if there is somewhere I can find it? I know old version is obsolete , but this is not for production, but for research, so please help me if you have any idea. Thanks very much in advance. In most cases you can get old versions of the HBase source from the ASF archives: http://archive.apache.org/dist/hbase/hbase-0.94.5/ -- Sean
Why hbase need manual split?
Hi, all, As I understand, HBase will automatically split a region when the region is too big. So in what scenario, user needs to do a manual split? Could someone kindly give me some examples that user need to do the region split explicitly via HBase Shell or Java API? Thanks very much. Regards, Ming
RE: Why hbase need manual split?
Thanks Arun, and John, Both of your scenarios make a lot of sense to me. But for the sequence-based key case, I am still confused. It is like an append-only operation, so new data are always written into the same region, but that region will eventually reach the hbase.hregion.max.filesize and be automatically split, why still need a manual split? If we set the hbase.hregion.max.filesize to a not too big value, then a region will never grow too big? And I think I need to first understand how HBase do the auto split internally ( I am very new to HBase). Given a region with start key A, and end key B. When split, how HBase do split internally? Split in the middle of key range? Original region is in range [A,B], so split to [A, B-A/2] and [B-A/2+1, B] ? Then if most of the row key are in a small range [A, C], while C is very close to B-A/2, then I can see a problem of auto split. Is this true? Can HBase do split in other ways? Thanks, Ming -Original Message- From: john guthrie [mailto:graf...@gmail.com] Sent: Wednesday, August 06, 2014 6:01 PM To: user@hbase.apache.org Subject: Re: Why hbase need manual split? i had a customer with a sequence-based key (yes, he knew all the downsides for that). being able to split manually meant he could split a region that got too big at the end vice right down the middle. with a sequentially increasing key, splitting the region in half left one region half the desired size and likely to never be added to On Wed, Aug 6, 2014 at 2:44 AM, Arun Allamsetty arun.allamse...@gmail.com wrote: Hi Ming, The reason why we have it is because the user can decide where each key goes. I can think multiple scenarios off the top of my head where it would be useful and others can correct me if I am wrong. 1. Cases where you cannot have row keys which are equally lexically distributed, leading in unequal loads on the regions. In such cases, we can set key ranges to be assigned to different regions so that we can have a more equal distribution. 2. The second scenario I am thinking of may be wrong and if it is, it'll clear my misconceptions. In case you cannot denormalize your data and you have to perform joins on certain range of row keys which are lexically similar. So we split them and they would be assigned to the same region server (right?) and the join would be performed locally. Cheers, Arun Sent from a mobile device. Please don't mind the typos. On Aug 6, 2014 12:30 AM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com wrote: Hi, all, As I understand, HBase will automatically split a region when the region is too big. So in what scenario, user needs to do a manual split? Could someone kindly give me some examples that user need to do the region split explicitly via HBase Shell or Java API? Thanks very much. Regards, Ming
RE: Why hbase need manual split?
Thanks John, This is a very good answer, now I understand why you use manual split, thanks. And I have a typo in my previous post, The C is very close to A not to B-A/2. So every split in middle of key range will result a big region and a small region. So very bad. So HBase only do auto split in middle of key range? Or there exist other algorithm here. Any help will be very appreciated! Best Regards, Ming -Original Message- From: john guthrie [mailto:graf...@gmail.com] Sent: Wednesday, August 06, 2014 6:35 PM To: user@hbase.apache.org Subject: Re: Why hbase need manual split? to be honest, we were doing manual splits for the main reason that we wanted to make sure it was done on our schedule. but it also occurred to me that the automatic splits, at least by default, split the region in half. normally the idea is that both new halves continue to grow, but with a sequentially increasing key that won't be true. so if you're splitting in half you want your region split size to be twice your desired region size so that when a split does occur the older half of the region is the size you want it. manual splitting lets you split at the end hope this helps, and hope i'm not wrong, john On Wed, Aug 6, 2014 at 6:25 AM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com wrote: Thanks Arun, and John, Both of your scenarios make a lot of sense to me. But for the sequence-based key case, I am still confused. It is like an append-only operation, so new data are always written into the same region, but that region will eventually reach the hbase.hregion.max.filesize and be automatically split, why still need a manual split? If we set the hbase.hregion.max.filesize to a not too big value, then a region will never grow too big? And I think I need to first understand how HBase do the auto split internally ( I am very new to HBase). Given a region with start key A, and end key B. When split, how HBase do split internally? Split in the middle of key range? Original region is in range [A,B], so split to [A, B-A/2] and [B-A/2+1, B] ? Then if most of the row key are in a small range [A, C], while C is very close to B-A/2, then I can see a problem of auto split. Is this true? Can HBase do split in other ways? Thanks, Ming -Original Message- From: john guthrie [mailto:graf...@gmail.com] Sent: Wednesday, August 06, 2014 6:01 PM To: user@hbase.apache.org Subject: Re: Why hbase need manual split? i had a customer with a sequence-based key (yes, he knew all the downsides for that). being able to split manually meant he could split a region that got too big at the end vice right down the middle. with a sequentially increasing key, splitting the region in half left one region half the desired size and likely to never be added to On Wed, Aug 6, 2014 at 2:44 AM, Arun Allamsetty arun.allamse...@gmail.com wrote: Hi Ming, The reason why we have it is because the user can decide where each key goes. I can think multiple scenarios off the top of my head where it would be useful and others can correct me if I am wrong. 1. Cases where you cannot have row keys which are equally lexically distributed, leading in unequal loads on the regions. In such cases, we can set key ranges to be assigned to different regions so that we can have a more equal distribution. 2. The second scenario I am thinking of may be wrong and if it is, it'll clear my misconceptions. In case you cannot denormalize your data and you have to perform joins on certain range of row keys which are lexically similar. So we split them and they would be assigned to the same region server (right?) and the join would be performed locally. Cheers, Arun Sent from a mobile device. Please don't mind the typos. On Aug 6, 2014 12:30 AM, Liu, Ming (HPIT-GADSC) ming.l...@hp.com wrote: Hi, all, As I understand, HBase will automatically split a region when the region is too big. So in what scenario, user needs to do a manual split? Could someone kindly give me some examples that user need to do the region split explicitly via HBase Shell or Java API? Thanks very much. Regards, Ming
