I don't have a set requirement. Just trying to learn more about the system
and 25 minutes seemed excessive. I really have nothing to compare against
and have no expectations; but, it takes about 900 seconds to run the count
function in the shell. My main goal is to figure out what
can have a max of around 4
open files if there are 2000 regions per node)...
Let me also check the logs a little more carefully and get back to the
forum..
Thank you
Vidhya
On 5/26/10 9:38 AM, Jean-Daniel Cryans jdcry...@apache.org wrote:
I'm pretty sure something else is going on.
1
Invisible. What's your need?
J-D
On Wed, May 26, 2010 at 3:56 PM, Vidhyashankar Venkataraman
vidhy...@yahoo-inc.com wrote:
Is there a way to customize the compaction function (like a hook provided by
the API) or is it invisible to the user?
Thank you
Vidhya
With beefy nodes, don't be afraid of using bigger regions... and LZO.
At stumbleupon we have 1GB maxfilesize on our 13B rows table and LZO
enabled on every table. The number of regions per node is a factor of
so many things... size of rows, acces pattern, hardware, etc. FWIW, I
would say that you
HBase became an Apache Top Level project recently so everything is
moving around, and this isn't an atomic operation ;)
Thanks for reporting!
J-D
On Sat, May 29, 2010 at 9:52 AM, Charles Woerner
charleswoer...@gmail.com wrote:
Following the links from the Releases page (
OOME is a Java exception, nothing HBase specific. It means that the
JVM ran out of memory.
BTW your log wasn't attached to your email (they are usually blocked),
so please post it on a web server or pastebin it so we can help you.
J-D
On Wed, Jun 9, 2010 at 11:02 AM, Vidhyashankar Venkataraman
On Wed, Jun 9, 2010 at 11:16 AM, Vidhyashankar Venkataraman
vidhy...@yahoo-inc.com wrote:
What do you mean by pastebinning it? I will try hosting it on a webserver..
pastebin.com
I know that OOME is Java running out of heap space: Can you let me know what
are the usual causes for OOME
allow more
than around 3-4 gigs of RAM..
On 6/9/10 11:26 AM, Jean-Daniel Cryans jdcry...@apache.org wrote:
On Wed, Jun 9, 2010 at 11:16 AM, Vidhyashankar Venkataraman
vidhy...@yahoo-inc.com wrote:
What do you mean by pastebinning it? I will try hosting it on a webserver..
pastebin.com
I
Also 0.20.4 has the ExplicitColumnTracker that spins in a infinite
loop in some situations.
J-D
On Thu, Jun 10, 2010 at 3:38 PM, Ryan Rawson ryano...@gmail.com wrote:
hey,
so you have discovered a particular 'trick' about how the HBase RPC
works... at the lowest level there is only 1 socket
You can check the general health by using the webui, it runs on the
master node at port 60010.
For the errors, the context you gave is so limited that giving any
meaningful answer is impossible. Please post full logs on a web server
or on pastebin.com (or your preferred code pasting site) if it
Friso,
This is very interesting, and nobody answered probably because no one
tried tcp_tw_recycle. I personally didn't even know about that config
until a few minutes ago ;)
So from the varnish mailing list, it seems that machines behind
firewalls or NAT won't play well with that config, but I
While I'm trying to figure out what is causing the region to be
non-responsive, what's the best way to recover?
I'd recover from that the same way I'd recover from any unavailable
storage component, since that can happen with any DB/SAN/etc right?
The link between your client and HBase could
this can be achieved with much improved performance?
Thank you.
Regards,
Raghava.
On Tue, Jun 22, 2010 at 12:57 PM, Jean-Daniel Cryans
jdcry...@apache.orgwrote:
This is not super clear, some comments inline.
J-D
On Tue, Jun 22, 2010 at 12:49 AM, Raghava Mutharaju
m.vijayaragh
At first glance it looks like a double assignment of .META., and the
file it's trying to get was probably already compacted by another
region server. Check the master log to see why/how it happened.
J-D
On Fri, Jun 25, 2010 at 11:13 AM, Ted Yu yuzhih...@gmail.com wrote:
Hi,
I upgraded a 3 node
We use ganglia at StumbleUpon, to enable metrics see
http://hbase.apache.org/docs/r0.20.5/metrics.html
J-D
On Fri, Jun 25, 2010 at 11:52 AM, Palaniappan Thiyagarajan
pthiyagara...@cashedge.com wrote:
Hi,
I would like to know what kind of monitoring you are having for your
production env and
:39,898 DEBUG org.apache.hadoop.hbase.regionserver.HRegion:
Closing TED-PACKAGESUMMARY-1277485167424-0,,1277485173774: disabling
compactions flushes
Any insight ?
Thanks
On Fri, Jun 25, 2010 at 11:31 AM, Jean-Daniel Cryans jdcry...@apache.org
wrote:
At first glance it looks like a double
(sorry it took so long to answer, we were all busy with the various
meetings around the Bay Area)
I can see the issue:
2010-06-30 13:48:16,135 DEBUG master.BaseScanner
(BaseScanner.java:checkAssigned(580)) - Current assignment of
.META.,,1 is not valid; serverAddress=, startCode=0 unknown.
...
://hbase.apache.org/docs/r0.89.20100621/ for more info. Cloudera's
CDH3b2 also has everything you need.
J-D
On Thu, Jul 1, 2010 at 12:03 PM, Jean-Daniel Cryans jdcry...@apache.org wrote:
653 regions is very low, even if you had a total of 3 region servers I
wouldn't expect any problem.
So to me
the whole hbase. Anyway, I am
regenerating the data from
scratch and let's see if it will work out.
Jimmy.
--
From: Jean-Daniel Cryans jdcry...@apache.org
Sent: Thursday, July 01, 2010 2:17 PM
To: user@hbase.apache.org
Subject: Re: dilemma
-0.20.4, and restart all hbase
master and regionservers. recreate all tables, etc.essentially starting
from scratch.
Jimmy
--
From: Jean-Daniel Cryans jdcry...@apache.org
Sent: Thursday, July 01, 2010 5:10 PM
To: user@hbase.apache.org
Subject: Re
--
From: Jean-Daniel Cryans jdcry...@apache.org
Sent: Thursday, July 01, 2010 5:10 PM
To: user@hbase.apache.org
Subject: Re: dilemma of memory and CPU for hbase.
add_table.rb doesn't actually write much in the file system, all your
data is still there. It just
You won't find it, because 0.20 isn't mavenized. Trunk is, so the next
major version will be available.
J-D
On Thu, Jul 1, 2010 at 7:47 PM, Fabiano Beppler f.bepp...@gmail.com wrote:
Hi,
Does anyone know where can I find hbase-0.20.5 in a maven repository?
Thanks in advance!
Fabiano
release:
http://archive.cloudera.com/cdh/2/hadoop-0.20.1+169.89.releasenotes.html
Jean-Daniel Cryans wrote:
Yes, and to deploy the cloudera release on your cluster :)
J-D
On Thu, Jul 1, 2010 at 11:31 AM, Ferdy ferdy.gal...@kalooga.com wrote:
Allright so I will use a cloudera release. If I
You need to configure hbase.zookeeper.quorum, else the region server cannot
guess on which machine in the local network it should find ZooKeeper. This
is described in
http://hbase.apache.org/docs/r0.20.5/api/overview-summary.html#fully-distrib
Also I recommend using 0.20.5
J-D
On Tue, Jul 6,
Jamie,
Does your configuration meets the requirements?
http://hbase.apache.org/docs/r0.20.5/api/overview-summary.html#requirements
ulimit and xcievers, if not set, are usually time bombs that blow off when
the cluster is under load.
J-D
On Wed, Jul 7, 2010 at 9:11 AM, Jamie Cockrill
Which HBase version are you running? Did you even take a look at the server
logs? (this client log you pasted just informs us that the connection to
ZooKeeper was successful)
J-D
On Wed, Jul 7, 2010 at 8:49 AM, manua agarwal.m...@gmail.com wrote:
Hi All,
I have set up the Hbase in
HBase is currently faster at writing than random reading, but long
scans are faster than writing. Not sure exactly what that unnamed
article is referring to.
Also, about using any other DBMS in conjunction with HBase, I would
simply recommend using the right tool for the right job.
J-D
On Wed,
HBase probably won't support 0.21 at all, since that release is marked
unstable. HBase 0.90 will be on hadoop 0.20-append which has a
different implementation for sync (HDFS-200 instead of HDFS-265).
I personally expect that everything will be tied back together for Hadoop 0.22
J-D
On Wed, Jul
,
So, If the application is like application for teller in banking,
whether the right to use Hbase?
regards
Firdaus
On 07/07/2010 11:36 PM, Jean-Daniel Cryans wrote:
HBase is currently faster at writing than random reading, but long
scans are faster than writing. Not sure exactly what
Passing the hbase.zookeeper.quorum config will do exactly what you
need in 0.89, but I'm not sure that it will work in 0.20
J-D
On Wed, Jul 7, 2010 at 10:46 AM, Jun Li jltz922...@gmail.com wrote:
Hello,
In my current application environment, I need to have two HBase
clusters running in two
. I've upped my MAX_FILESIZE on
my table to 1GB to see if that helps (not sure if it will!).
Thanks,
Jamie
On 7 July 2010 18:12, Jean-Daniel Cryans jdcry...@apache.org wrote:
xcievers exceptions will be in the datanodes' logs, and your problem
totally looks like it. 0.20.5 will have
You configured your table to use LZO, but it's not on the classpath.
Please read and follow
http://wiki.apache.org/hadoop/UsingLzoCompression
J-D
On Wed, Jul 7, 2010 at 11:58 AM, manua agarwal.m...@gmail.com wrote:
Hi,
I have created a swap space of 1Gb, reduced the heap size to 500Mb and
remain.
On Wed, Jul 7, 2010 at 1:13 PM, nocones77-gro...@yahoo.com wrote:
From: Jean-Daniel Cryans jdcry...@apache.org
Also, about using any other DBMS in conjunction with HBase, I would
simply recommend using the right tool for the right job.
This seems like a sensible approach to me
You don't, but the LZO code is in C++ so the jar only contains the
bindings. You also need to copy the librairies like it does in
http://wiki.apache.org/hadoop/UsingLzoCompression :
cp build/native/Linux-amd64-64/lib/libgplcompression.*
hbase/lib/native/Linux-amd64-64/
But adapted to your
memory is about the same, but the memory
cache is much much bigger, which presumably is healthlier as, in
theory, that ought to relinquish memory to processes that request it.
Lets see if that does the trick!
ta
Jamie
On 7 July 2010 19:30, Jean-Daniel Cryans jdcry...@apache.org wrote
Ted Yu already answered that yesterday:
If you look at HBase.rb, line 554, you would find the bug.
Here is the correct call:
split = KeyValue.parseColumn(column.to_java_bytes)
I guess that line was copied from line 546 which isn't in for loop.
You can file a JIRA.
J-D
On Wed,
More info on this blog post:
http://www.larsgeorge.com/2010/05/hbase-file-locality-in-hdfs.html
J-D
On Thu, Jul 8, 2010 at 10:11 AM, Jean-Daniel Cryans jdcry...@apache.org wrote:
This would be done at the expense of network IO, since you will lose
locality for jobs that read/write to HBase
..Following up on this thread the
article..
Could some one elaborate why locality is lost upon restart? Is it because
of random assignment by HMaster and/or HRegionServer is stateless or other
reasons?
thanks
venkatesh
-Original Message-
From: Jean-Daniel Cryans jdcry
(changing the subject, let's not hijack threads)
will the data move over time though...for example if i have lots of access to
data in DataNode A ? without the current work that is in progress..
HBase has no control on that, but data will be moved if those regions
are used. Like the article
subject before last send
..forgot..
thanks again..
i'll be one active user with tons of Qs in the next few months :)
-Original Message-
From: Jean-Daniel Cryans jdcry...@apache.org
To: user@hbase.apache.org
Sent: Thu, Jul 8, 2010 1:58 pm
Subject: About data locality (Was: Re
It's not IO intense, it's IO latency sensitive eg. if other processes
are sucking up most of the IO bandwidth then ZK will have a hard time
taking quorum decisions.
Disks are cheap, and a single 7.2k dedicated disk can be enough.
J-D
On Thu, Jul 8, 2010 at 5:38 PM, Arun Ramakrishnan
HBASE-1621 isn't about automatic merging and it's still very experimental.
The issue with doing automatically is that you have to figure that two
regions, together, are smaller in size than the max size of a region
to split. At the same time, it's not because two regions are small
that you want
At stumbleupon, we have su.pr (url shortner / advertising platform)
that's totally based on HBase and has been in production for more than
a year. Also many other parts of our main product also rely on HBase.
J-D
On Sat, Jul 10, 2010 at 10:43 AM, S Ahmed sahmed1...@gmail.com wrote:
Its my
.
thanks in advance
venkatesh
-Original Message-
From: Jean-Daniel Cryans jdcry...@apache.org
To: user@hbase.apache.org
Sent: Sat, Jul 10, 2010 1:47 pm
Subject: Re: real world usage, any web applications built using hbase?
At stumbleupon, we have su.pr (url shortner / advertising
This is all zookeeper logging, and although it's chatty I don't see
anything wrong in it (the NodeExists is normal since in 0.20 there are
some places that instead of checking for a znode existence, we try to
create it and only if it fails we update it). Also we log everything
that comes out of
Which HBase version are you using? Currently replication is only
available in trunk (and will be available in the next 0.89 release).
The documentation is available at
src/main/java/org/apache/hadoop/hbase/replication/package.html, you
probably forgot to run the bin/replication/add_peer.rb script
This kind of issue was discussed a couple of times on this mailing
list. Basically, you can play with hbase.client.pause and
hbase.client.retries.number but you won't find this satisfactory,
which is why we opened
https://issues.apache.org/jira/browse/HBASE-2445
J-D
On Tue, Jul 13, 2010 at 10:38
Thanks for info where i can find some documentation. There is info about
zookeeper that it need running in standalone mode it is true?
Well you can run add_peer.rb when the clusters are running, but they
won't pickup the change live (that part isn't done yet). So if you run
the script while
!
-justin
On 7/13/10 10:47 AM, Jean-Daniel Cryans wrote:
This kind of issue was discussed a couple of times on this mailing
list. Basically, you can play with hbase.client.pause and
hbase.client.retries.number but you won't find this satisfactory,
which is why we opened
https
loading data
connect slave or turn on replication on existing tables with data?
W dniu 13.07.2010 19:56, Jean-Daniel Cryans pisze:
Thanks for info where i can find some documentation. There is info about
zookeeper that it need running in standalone mode it is true?
Well you can run
)
at
org.apache.hadoop.hbase.replication.regionserver.ReplicationSource.run(ReplicationSource.java:262)
W dniu 13.07.2010 20:18, Jean-Daniel Cryans pisze:
No, but you can use the new mapreduce utility
org.apache.hadoop.hbase.mapreduce.CopyTable to copy whole tables
between clusters. It's like distcp for HBase.
Oh and looking
Please use a pasting service for the log traces. I personally use pastebin.com
You probably had a GC that lasted too long, this is something out of
the control of the application (apart from trying to put as less data
in memory as possible, but you are inserting so...). Your log doesn't
contain
?
I don't work for cloudera, but IIRC the next beta for CDH3 is due for September.
Jimmy
--
From: Jean-Daniel Cryans jdcry...@apache.org
Sent: Tuesday, July 13, 2010 2:55 PM
To: user@hbase.apache.org
Subject: Re: regionserver crash under heavy
14.07.2010 00:08, Jean-Daniel Cryans pisze:
Just looked at the head of 0.20-append and I see it contains the
missing patch (was committed as part of HDFS-1057).
So that would mean that the file is just empty :) If you insert a few
rows in the shell on the master cluster, do you see them some
release of the cloudera distribution for both hadoop
and hbase.
Jimmy
--
From: Jean-Daniel Cryans jdcry...@apache.org
Sent: Tuesday, July 13, 2010 4:24 PM
To: user@hbase.apache.org
Subject: Re: regionserver crash under heavy load
Your region
Nice story Stu ;)
So the first thing you saw in the master is that it was splitting the
HLogs from the dead region servers (looks like they were kill -9 or
failed very hard during/after your mapreduce job). As for all the HDFS
error messages, it sounds like you are missing the last requirement on
/backup.tar.gz
Jinsong
--
From: Jean-Daniel Cryans jdcry...@apache.org
Sent: Wednesday, July 14, 2010 11:16 AM
To: user@hbase.apache.org
Subject: Re: regionserver crash under heavy load
So your region servers had their session expired but I don't see
Dmitry,
Your log shows this:
2010-07-12 15:10:03,299 WARN org.apache.hadoop.hbase.util.Sleeper: We slept
86246ms, ten times longer than scheduled: 1000
This is a pause that lasted more than a minute, the process was in
that state (GC, swapping, mix of all of them) for some reason and it
was
Stuart,
I answered this to your last serie of emails some hours ago, I still
stand behind my observations:
So the first thing you saw in the master is that it was splitting the
HLogs from the dead region servers (looks like they were kill -9 or
failed very hard during/after your mapreduce job).
Check xcievers too, giving more mem to HBase won't fix HDFS.
J-D
On Wed, Jul 14, 2010 at 5:53 PM, Stuart Smith stu24m...@yahoo.com wrote:
Ok, back up again.
This time, in addition to just watching restarting, the error logs led me
to this:
About your new crash:
2010-07-15 04:09:03,248 DEBUG
org.apache.hadoop.hbase.io.hfile.LruBlockCache: Cache Stats: Sizes:
Total=3.3544312MB (3517376), Free=405.42056MB (425114272),
Max=408.775MB (428631648), Counts: Blocks=0, Access=1985476, Hit=0,
Miss=1985476, Evictions=0, Evicted=0, Ratios: Hit
Nothing particular in that dump, and I'm not aware of any deadlock in
0.20.3, could we see the region server log?
Thx
J-D
On Thu, Jul 15, 2010 at 4:17 PM, Luke Forehand
luke.foreh...@networkedinsights.com wrote:
First post evar!
I have a 3 node cluster and have set
Looks like your .META. is confused about things, or your master is
editing it in a weird way. AFAIK, it's not a known issue with 0.20.5.
I would advise first scanning the .META. table and look if your rows
are changing between shell invocations (just look at the first row of
each table).
If it
u18 frequently sigsegvs on users (and then they wonder why region
servers are missing), and this is also true for Hadoop. u20 seems
stable but a lot of people still prefer u16.
J-D
On Tue, Jul 20, 2010 at 4:23 PM, Syed Wasti mdwa...@hotmail.com wrote:
Hi,
We recently upgraded our QA cluster
So you would buffer edits going to the same row? Unless you have your
own write-ahead-log, you'd likely lose data on node failure.
But WRT your question, 5 cells with different timestamps is as costly
to store/query as 5 cells with the same timestamp. The major
difference is that the former case
HBaseAdmin.getClusterStatus().getServers()
J-D
On Wed, Jul 21, 2010 at 9:56 AM, Ted Yu yuzhih...@gmail.com wrote:
Hi,
Is there API to query the number of live region servers ?
Thanks
times longer than scheduled: 1
Is it possible for HBase Master to restart dead region server in this case ?
On Wed, Jul 21, 2010 at 10:02 AM, Jean-Daniel Cryans
jdcry...@apache.orgwrote:
HBaseAdmin.getClusterStatus().getServers()
J-D
On Wed, Jul 21, 2010 at 9:56 AM, Ted Yu yuzhih
ZooKeeper is only a canary, telling the region server that it was
partionned from the cluster for longer than the default timeout
somehow, usually because of GC pauses. You should see lines like
slept for x, long than y messages before what you pasted.
J-D
On Wed, Jul 21, 2010 at 2:49 PM, Steve
Is HADOOP_CLASSPATH=${HBSE_CONF_DIR}' pointing to the right location
on every machine in the cluster? While the job is running, you can go
on one slave machine, issue a ps aux | grep java and check if the
Child tasks have the correct classpath.
J-D
On Wed, Jul 21, 2010 at 3:03 PM, HAN LIU
and
increase zookeeper timeout settings. I will give them a trial after
finishing queued data load.
Other suggestions are most welcome.
On Wed, Jul 21, 2010 at 2:55 PM, Jean-Daniel Cryans
jdcry...@apache.orgwrote:
ZooKeeper is only a canary, telling the region server
On Jul 22, 2010, at 4:43 PM, Jean-Daniel Cryans wrote:
Han,
This is bad, you must be doing something slow like creating a new
HTable for each put call. Also you need to use the write buffer
(disable auto flushing, then set the write buffer size on HTable
during the map configuration) if since
See http://hbase.apache.org/docs/r0.89.20100621/acid-semantics.html
But, Please note that this is not true _across rows_ for multirow
batch mutations. is actually untrue for 0.89 since
https://issues.apache.org/jira/browse/HBASE-2353 (wasn't reflected in
that document).
Hope that helps,
J-D
On
TableOutputFormat is really just a wrapper around a HTable, see for
yourself
http://github.com/apache/hbase/blob/0.20/src/java/org/apache/hadoop/hbase/mapreduce/TableOutputFormat.java
So there must be something else about the way you use it, or the way
you use HTable directly. Showing bits of
Ryan (who wrote HFile) did a lot of testing around block size and
didn't really see any difference when changing it. So I would
recommend that you benchmark different values with your own data/usage
pattern and see if you do have better/worse perfs.
The tradeoff for larger values is that in order
Thanks for the heads up. Do you know what happens if I set this value larger
than 5MB? We will always be scanning the data, and always in large blocks.
I have yet to calculate the typical size of a single scan but imagine that it
will usually be larger than 1MB.
I never tried that, hard
commits. There seems to be some other issue
too here.
Thanks
Karthik
On Tue, Jul 27, 2010 at 10:18 AM, Jean-Daniel Cryans
jdcry...@apache.orgwrote:
Since they are in the same batch, they could end up on the same
timestamp and one will hide the other. When not batched, there's
always a few
, 2010, at 10:13 AM, Jean-Daniel Cryans wrote:
After altering the table, issue a major compaction on it and
everything will be re-written with the new block size.
Inline.
J-D
On Thu, Jul 29, 2010 at 8:54 AM, Ferdy Galema ferdy.gal...@kalooga.com wrote:
Using Hbase 0.20.5 with Hadoop CDH2 0.20.1+169.89 I noticed something very
strange.
When overwriting a certain column in a column family with 1 VERSIONS, and
removing that value later (for example
Well it says; Times: user=0.17 sys=0.04, real=299.23 secs
So why did it take 0.04 of system time but 300 secs of real time?
That's insane. Either the region server process was completely starved
of CPU cycles (are you on EC2 or any virtualized service like that?),
or the computer was put to
I see. Usually a whole customer fits within a region. Actually, the
number of customers that doesn't fit in a single region are only two or three.
But then another question comes up. Even if a put all the data in a single
table, given that the keys are written in order, and given that several
swappiness is something else, it's good to set it at 0 when you have
enough RAM to fit everything but it will still swap when you run out
if it and it will be a big hit.
I would advise monitoring the cluster, or at least very least look at
the output of the top command while the job is running.
From
http://hbase.apache.org/docs/r0.20.6/api/overview-summary.html#overview_description
At minimum, you should set the list of servers that you want
ZooKeeper to run on using the hbase.zookeeper.quorum property. This
property defaults to localhost which is not suitable for a fully
distributed
I think that the person who wrote the header of that page meant that
the hadoop-gpl-compression project lacks fixes included in Kevin's
repo. AFAIK you can hit those if you use LZOed files as input for MR,
but I've been using the second one for more than a year without any
issue (in HBase).
J-D
Is that coming from the master? If so, it means that it was trying to
write recovered data from a failed region server and wasn't able to do
so. It sounds bad.
- Can we get full stack traces of that error?
- Did you check the datanode logs for any exception? Very often
(strong emphasis on very),
cluster restart, is there any memory of which region servers last served
which regions or some other method to improve data locality?
Nope, not yet. The new master code for 0.90 has some basics, but it's
a bit complicated and we're not there yet. It basically requires
asking the Namenode for
correctly
as per the API doc you mention.
Thanks
Jamie
On 2 August 2010 19:18, Jean-Daniel Cryans jdcry...@apache.org wrote:
Is that coming from the master? If so, it means that it was trying to
write recovered data from a failed region server and wasn't able to do
so. It sounds bad
On a small cluster like that I wouldn't bother giving 3 machines to
zookeeper since your cluster is a reliable as your master node.
Instead, make sure that your master has some redundant hardware and
put a standalone zookeeper on it.
J-D
On Tue, Aug 3, 2010 at 3:41 PM, Justin Cohen
Inline.
J-D
On Tue, Aug 3, 2010 at 2:44 PM, Vidhyashankar Venkataraman
vidhy...@yahoo-inc.com wrote:
A few issues I have been observing on changing block settings:
1. What happens if we change the block size of a column family on an
already populated database? Will this not throw apps on
Sorry took a day to answer, see inline.
J-D
On Mon, Aug 2, 2010 at 10:47 AM, Maxim Veksler ma...@vekslers.org wrote:
Hello,
We're setting up a data warehouse environment that includes Hadoop, HBase,
Hive and our own in-house MR jobs.
I would like with your permission to discuss the
The first part of the log is usually a client that died while it was
doing a request.
The second part of the log is a session expiration. The log fragment
is too small to tell if it was the region server that paused or the ZK
ensemble that was unreachable during that time... are the zk servers
Wow. I get tons of them in the logs.. And there arent that many clients that
got killed as reported by the MR job.. Is that the only case when these
errors are reported?
What about speculative execution? Or RPC timeouts (do you log that)?
Ok good, so one of the two happened then.. I will
If the inserts are coming from more than 1 client, and your are trying
to delete from only 1 client, then likely it won't work. You could try
using a pool of deleters (multiple threads that delete rows) that you
feed from the scanner. Or you could run a MapReduce that would
parallelize that for
HDFS errors + this message:
2010-08-07 08:41:40,392 INFO org.apache.zookeeper.ClientCnxn: Client session
timed out, have not heard from server in 332256ms for sessionid
points to a GC pause (that lasted about 5 minutes). Make sure you
aren't swapping, that HBase isn't CPU starved, monitor your
Hard to tell if it's decent performance. How do you define decent?
What kind of hardware are we talking about? Which version are you
using? How much memory was given to HBase?
Also did you set the write buffer on the client side on HTable? Did
you also turn off auto-flushing? Do you monitor your
-Daniel Cryans wrote:
Hard to tell if it's decent performance. How do you define decent?
I consider it descent if it is roughly the best performance one can get using
my schema on my machines
What kind of hardware are we talking about?
One machine for HBase master and 6 regionservers. Specs
? For example how did you
do the first 3 steps described on that page?
Also is there any documentation that describes the multiple-client in HBase?
On Aug 9, 2010, at 2:37 PM, Jean-Daniel Cryans wrote:
That's pretty powerful machines, I would expect more performance. You
could try using the same
I see. 0.89 is the still a developer release and I hear that it is not
stable. But it sounds really tempting because it boosts performance by a lot.
Can I trust it if my final goal is to insert about 100TB of data? What could
be the possible issues? Also when shall I expect to see a stable
The region server sets the timestamp to System.currentTimeMillis if it
wasn't set by the user.
J-D
On Tue, Aug 10, 2010 at 11:38 AM, Michael Segel
michael_se...@hotmail.com wrote:
Its possible that when you're writing a row to HBase that you do not specify
the timestamp.
My question is
I don't remember seeing anything like that, except that if your data
isn't compressible then it will probably take a little longer to
insert your data. The vmstat and top outputs are pretty useless in
this case, as we can't tell exactly what's going on in the processes.
I suggest you take a look
clients? Or maybe point me to a reference on such topics
since I am not really an expert on Java. :p
Thanks again for your reply.
Han
On Aug 9, 2010, at 4:13 PM, Jean-Daniel Cryans wrote:
I see. 0.89 is the still a developer release and I hear that it is not
stable. But it sounds really
1 - 100 of 1248 matches
Mail list logo
