Well, the problem is still there, i.e. I tried to add one more index and the
3-node cluster is just going spastic, becomes unresponsive etc. These boxes
have plenty of CPU and memory.
--
View this message in context:
:)
On Thu, Mar 31, 2011 at 3:06 PM, buddhasystem lt;potek...@bnl.govgt;
wrote:
I just configured a cluster of two nodes -- do these token values make
sense?
The reason I'm asking that so far I don't see load balancing to be
happening, judging from performance.
Address Status
I'm rebalancing a cluster of 2 nodes at this point. Netstats on the source
node reports progress of the stream, whereas on the receving end netstats
states that progress = 0. Did anyone see that?
Do I need both nodes listed as seeds in cassandra.yaml?
TIA/
--
View this message in context:
I just configured a cluster of two nodes -- do these token values make sense?
The reason I'm asking that so far I don't see load balancing to be
happening, judging from performance.
Address Status State LoadOwnsToken
Yup, I screwed up the token setting, my bad.
Now, I moved the tokens. I still observe that read latency deteriorated with
3 machines vs original one. Replication factor is 1, Cassandra version 0.7.2
(didn't have time to upgrade as I need results by this weekend).
Key and row caching was disabled
Hello Saurabh,
I have a similar situation, with a more complex data model, and I do an
equivalent of map-reduce by hand. The redeeming value is that you have
complete freedom in how you hash, and you design the way you store indexes
and similar structures. If there is a pattern in data store, you
aaron morton wrote:
Also a node is be responsible for storing it's token range and acting as a
replica for other token ranges. So reducing the token range may not have a
dramatic affect on the storage requirements.
Aaron,
is there a way to configure wimpy nodes such that the replicas
Jonathan,
for all of us just tinker with test clusters, building confidence in the
product, it would be nice to be able to do same with nodetool, without
jconsole, just my 0.5 penny. Thanks.
Jonathan Ellis-3 wrote:
From the next paragraph of the same wiki page:
SSTables that are
Jonathan, wide rows have been discussed. I thought that the limit on number
of columns is way bigger than 45k. What can one expect in reality?
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/0-7-2-choking-on-a-5-MB-column-tp6198387p6198548.html
I see. I'm doing something even more drastic then, because I'm only inserting
one row in this case, and just use cf.insert(), without batch mutator. It
didn't occur to me that was a bad idea.
So I take it, this method will fail. Hmm.
--
View this message in context:
understand your reference to the OOP in the context of a reading 100
columns from a row.
Aaron
On 19 Mar 2011, at 16:22, buddhasystem wrote:
gt; As I'm working on this further, I want to understand this:
gt;
gt; Is it advantageous to flatten data in blocks (strings) each
This has been discussed once, but I don't remember the outcome. I insert a
row and then delete the key immediately. I then run nodetool compact. In
cassanra-cli, list cf still return 1 empty row. This is not a showstopper
but damn unpretty. Is there a way to make deleted rows go, immediately?
--
Is there is noticeable difference in speed between reading the whole row
through Pycassa, vs a range of columns? Both rows and columns are pretty
slim.
--
View this message in context:
As I'm working on this further, I want to understand this:
Is it advantageous to flatten data in blocks (strings) each containing a
series of objects, if I know that a serial object read is often likely, but
don't want to resort to OPP? I worked out the optimal granularity, it seems.
Is it better
Hello, in the instructions, I need to link concurrent_reads to number of
drives. Is this related to number of physical drives that I have in my
RAID0, or something else?
--
View this message in context:
Thanks to all for replying, but frankly I didn't get the answer I wanted.
Does the number of disks apply to number of spindles in RAID0? Or
something else like a separate disk for commitlog and for data?
--
View this message in context:
Thanks Peter, I can see it better now.
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Does-concurrent-reads-relate-to-number-of-drives-in-RAID0-tp6182346p6183051.html
Sent from the cassandra-u...@incubator.apache.org mailing list archive at
Where and how do I choose it?
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Does-concurrent-reads-relate-to-number-of-drives-in-RAID0-tp6182346p6183069.html
Sent from the cassandra-u...@incubator.apache.org mailing list archive at
Nabble.com.
Dear All,
this is from my new Cassandra server. It obviously uses hyperthreading, I
just don't know how to translate this to concurrent readers and writers in
cassandra.yaml -- can somebody take a look and tell me what number of cores
I need to assume for concurrent_reads and concurrent_writes. Is
Hello Peter, thanks for the note.
I'm not looking for anything fancy. It's just when I'm looking at the
following bit of Pycassa docs, it's not 100% clear to me that it won't
overwrite the entire row for the key, if I want to simply add an extra
column {'foo':'bar'} to the already existing row. I
Thanks for clarification, Tyler, sorry again for the basic question. I've
been doing straight inserts from Oracle so far but now I need to update rows
with new columns.
--
View this message in context:
Thanks! Docs say it's good to set it to 8*Ncores, are saying you see 8 cores
in this output? I know I need to go way above default 32 with this setup.
--
View this message in context:
Sorry for the rather primitive question, but it's not clear to me if I need
to fetch the whole row, add a column as a dictionary entry and re-insert it
if I want to expand the row by one column. Help will be appreciated.
--
View this message in context:
Thanks. Can you give me a pycassa example, if possible?
Thanks!
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Is-column-update-column-atomic-or-row-atomic-tp6174445p6174487.html
Sent from the cassandra-u...@incubator.apache.org mailing list
Tyler, as a collateral issue - I've been wondering for a while what advantage
if any it buys me, if I declare a value 'long' (which it roughly is) as
opposed to passing around strings. String is flattened onto a replica of
itself, I assume? No conversion? Maybe it even means better speed.
Thanks,
I'm doing insertion with a pycassa client. It seems to work in most cases,
but sometimes, when I go to Cassandra-cli, and query with key and column
that I inserted, I get null whereas I shouldn't. What could be causes for
that?
--
View this message in context:
thresholds: 4/32
Read repair chance: 1.0
Built indexes: []
I pretty much went with the default settings, and the column name is
'CATALOG'.
Maxim
Tyler Hobbs-2 wrote:
On Thu, Feb 24, 2011 at 2:27 PM, buddhasystem potek...@bnl.gov wrote:
I'm doing insertion with a pycassa
Thanks! You are right. I see exception but have no idea what went wrong.
ERROR [ReadStage:14] 2011-02-24 21:51:29,374 AbstractCassandraDaemon.java
(line 113) Fatal exception in thread Thread[ReadStage:14,5,main]
java.io.IOError: java.io.EOFException
at
FWIW, for me the advantage of homebrew indexes is that they can be a lot more
sophisticated than the standard -- I can hash combinations of column values
to whatever I want. I also put counters on column values in the index, so
there is lots of functionality. Of course, I can do it because my
I know that theoretically it should not (apart from compaction issues), but
maybe somebody has experience showing otherwise:
My test cluster now has 250GB of data and will have 1.5TB in its
reincarnation. If all these data is in a single CF -- will it cause read or
write performance problems?
There was a discussion here on how well (or not so well) the Super CFs are
supported. I now need to make a strategic decision as to how I plan my data.
What's the consensus -- will the super CF be there 3 years out?
TIA
Maxim
--
View this message in context:
Well I know the cache is there for a reason, I just can't explain the factor
of 4 when I run my queries on a hot vs cold cache. My queries are actually a
chain of one on an inverted index, which produces a tuple of keys to be used
in the main query. The inverted index query should be downright
I've been too smart for my own good trying to type columns, on the theory
that it would later increase performance by having more efficient
comparators in place. So if a string represents an integer, I would convert
it to an integer and declare the column as such. Same for LONG.
What I found is
is pretty close (just a size check).
If you meant that the conversion is killing performance on your
client, you should switch to a more performant client language. :)
On Fri, Feb 18, 2011 at 9:56 PM, buddhasystem potek...@bnl.gov wrote:
I've been too smart for my own good trying to type
I sidestep this problem by using a Python script (pycassa-based) where I
configure my CFs. This way, it's reproducible and documented.
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/create-additional-secondary-index-tp6033574p6033683.html
Sent
Hello,
we are acquiring new hardware for our cluster and will be installing it
soon. It's likely that I won't need to rely on secondary index
functionality, as data will be write-once read-many and I can get away with
inverse index creation at load time, plus I have some more complex indexing
in
Thank you! It's just that 7.1 seems the bleeding edge now (a serious bug
fixed today). Would you still trust it as a production-level service? I'm
just slightly concerned. I don't want to create a perception among our IT
that the product is not ready for prime time.
--
View this message in
Thank you Attila!
We will indeed have a few months of breaking in. I suppose I'll
keep my fingers crossed and see that 0.7.X is very stable. So I'll
deploy 0.7.1 -- I will need to apply all the patches, there is no
cumulative download, is that correct?
Attila Babo wrote:
0.6.8 is stable and
I've been thinking about this as well. I'm migrating data from a large Oracle
database, and the RDBMS columns names are descriptive (good) and long (bad).
For now I just keep them when populating Cassandra, but I can shave off
about 30% of storage by hashing names. I don't need any automation and
I asked a similar question (but didn't receive an answer). I'm trying to see
if a large number of CFs might be beneficial. One thing I can think about is
the size of extra storage needed for compaction -- obviously it will be
smaller in case of many smaller CFs.
--
View this message in context:
Does it also mean that the whole row will be deserialized when a query comes
just for one column?
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Calculating-the-size-of-rows-in-KBs-tp6011243p6017870.html
Sent from the
Jonathan, what if the data is really homogeneous, but over a long period of
time. I decided that the users who hit the database for recent past should
have a better ride. Splitting into a separate CF also has costs, right?
In fact, if I were to go this way, do you think I can crank down the key
One of my nodes is 76% full. I know that one of CFs represents 90% of the
data, others are really minor. Can I still compact under these conditions?
Will it crash and lose the data? Will it try to create one very large file
out of fragments, for that dominating CF?
TIA
--
View this message in
Seeing that discussion here about indexes not supported in superCFs, and less
than clear future of superCFs altogether, I was thinking about getting a
modicum of same functionality with serialized objects inside columns. This
way the column key becomes sort of analog of supercolumn key, and I
Thanks for the comment! In my case, I want to store various time slices as
indexes, so the content can be serialized as comma-separated concatenation
of unique object IDs. Example: on 20101204, multiple clouds experienced a
variety of errors in job execution. In addition, multiple users ran (or
Hello,
one node in my 3-machine cluster cannot perform compaction. I tried multiple
times, it ran out of heap space once and I increased it. Now I'm getting the
dump below (after it does run for a few minutes). I hope somebody can shed a
little light on what' going on, because I'm at a loss and
Thanks Jonathan -- does it mean that the machine is experiencing IO problems?
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Java-bombs-during-compaction-please-help-tp6001773p6002320.html
Sent from the cassandra-u...@incubator.apache.org
Hello,
If the amount of data is _that_ small, you'll have a much easier life with
MySQL, which supports the join procedure -- because that's exactly what
you want to achieve.
asil klin wrote:
Hi all,
I want to procure the intersection of columns set of two rows (from 2
different column
Just wanted to see if someone with experience in running an actual service
can advise me:
how often do you run nodetool compact on your nodes? Do you stagger it in
time, for each node? How badly is performance affected?
I know this all seems too generic but then again no two clusters are
Thanks Edward. In our usage scenario, there is never downtime, it's a global
24/7 operation.
What is impacted the worst, the read or write?
How does a node handle compaction when there is a spike of writes coming to
it?
Edward Capriolo wrote:
On Sat, Feb 5, 2011 at 11:59 AM, buddhasystem
Jonathan,
what's the implementation of that? I.e. is is a product of indexes or nested
loops?
Thanks,
Maxim
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/order-of-index-expressions-tp5995909p5996488.html
Sent from the
Even when storage is in NFS, Cassandra can still be quite useful as a file
catalog. Your physical storage can change, move etc. Therefore, it's a good
idea to provide mapping of logical names to physical store points (which in
fact can be many). This is a standard technique used in mass storage.
FWIW, I'm working on migrating a large amount of data out of Oracle into my
test cluster. The data has been warehoused as CSV files on Amazon S3. Having
that in place allows me to not put extra load on the production service when
doing many repeated tests. I then parse the data using CSV Python
CouchDB
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Using-Cassandra-to-store-files-tp5988698p5989122.html
Sent from the cassandra-u...@incubator.apache.org mailing list archive at
Nabble.com.
Dude, are you asking me to unsubscribe?
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Slow-network-writes-tp5985757p5991488.html
Sent from the cassandra-u...@incubator.apache.org mailing list archive at
Nabble.com.
How often and by what criteria is the commit log compacted/truncated?
Thanks,
Maxim
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Commit-log-compaction-tp5985221p5985221.html
Sent from the cassandra-u...@incubator.apache.org mailing list
Thank you. So what is exactly the condition that causes the older commit log
files to actually be removed? I observe that indeed they are rotated out
when the threshold is reached, but then new ones a placed in the directory
and the older ones are still there.
Thanks,
Maxim
--
View this
Thanks. Just wanted to note that counting the number of rows where foo=bar is
a fairly ubiquitous task in db applications. In case of big data,
trafficking all these data to client just to count something isn't optimal
at all.
Maxim
--
View this message in context:
Thanks. Yes I know it's by no means trivial. I thought in case there was an
index on the column on which I want to place condition, the index machinery
itself can do the counting (i.e. when the index is updated, the counter is
incremented). It doesn't seem too orthogonal to the current
Oleg,
I just wanted to add that I confirmed the importance of that rule of thumb
the hard way. I created two extra CFs and was able to reliably crash the
nodes during writes. I guess for the final setting I'll rely on results of
my testing.
But it's also important to not cause the swap death of
Thanks.
Maxim
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/How-do-I-get-0-7-1-tp5986927p5986927.html
Sent from the cassandra-u...@incubator.apache.org mailing list archive at
Nabble.com.
Jonathan,
where do I find that contrib/stress?
Maxim
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Slow-network-writes-tp5985757p5986937.html
Sent from the cassandra-u...@incubator.apache.org mailing list archive at
Nabble.com.
Stephen, sorry I didn't understand your missive.
Maxim
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/How-do-I-get-0-7-1-tp5986927p5987184.html
Sent from the cassandra-u...@incubator.apache.org mailing list archive at
Nabble.com.
Most if not all modern web application frameworks support sessions. This
applies to Django (with which I have most experience and also run it with
X.509 security layer) but also to Ruby on Rails and Pylons.
So, why would you re-invent the wheel? Too messy. It's all out there for you
to use.
For completeness:
http://stackoverflow.com/questions/3746685/running-django-site-in-multiserver-environment-how-to-handle-sessions
http://docs.djangoproject.com/en/dev/topics/http/sessions/#using-cached-sessions
I guess your approach does make sense, one only wishes that the servlet in
question
When I do a lot of inserts into my cluster (10k at a time) I get timeouts
from Thrift, the TScoket.py module.
What do I do?
Thanks,
Maxim
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/TSocket-timing-out-tp5973548p5973548.html
Sent from the
As far as I know, there are no aggregate operations built into Cassandra,
which means you'll have to retrieve all of the data to count it in the
client. I had a thread on this topic 2 weeks ago. It's pretty bad.
--
View this message in context:
Sorry Aaron but this doesn't help. As I said, machine is dead, kaput,
finished. So I can't do decommission. I can remove token to any other
node, but -- the dead machine is going to hang around in my ring reports
like a zombie.
--
View this message in context:
It does remove tokens, and the ring shows that the problematic node owns 0
tokens, which is OK. However, it's still there, listed.
It's not a bug but kind of like a feature -- you can move that node back in
two days later and move tokens in same or different way.
What I wish happened was that
Will it work for a billion rows? Because that's where eventually I'll end up
being.
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Using-Cassandra-for-storing-large-objects-tp5965418p5966284.html
Sent from the
I would ask myself a different question, which is what media-hosting sites
use (YouTube and all others). Cassandra still may have its usefulness here
as a mapper between a logical id and physical file location.
--
View this message in context:
OK, after running repair and waiting overnight the rebalancing worked and
now 3 nodes share the load as I expected. However, one node that is broken
is still listed in the ring. I have no intention of reviving it. What's the
optimal way to get rid of it as far as the ring configuration is
I was moving a node and at some point it started streaming data to 2 other
nodes. Later, that node keeled over and let's assume I can't fix it for the
next 3 days and just want to move tokens on the remaining three to even out
and see if I can live with it.
But I can't do that! The node that was
Having separate columns for Year, Month etc seems redundant. It's tons more
efficient to keep say UTC time in POSIX format (basically integer). It's
easy to convert back and forth.
If you want to get a range of dates, in that case you might use Order
Preserving Partitioner, and sort out which
Bump. I still don't know what is the best things to do, plz help.
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Node-going-down-when-streaming-data-what-next-tp5962944p5964231.html
Sent from the cassandra-u...@incubator.apache.org mailing list
I used the term sharding a bit frivolously. Sorry. It's just splitting
semantically homogenious data among CFs doesn't scale too well, as each CF
is allocated a piece of memory on the server.
--
View this message in context:
Hello,
from what I know, you don't really have to restart simultaneously,
although of course you don't want to wait.
I finally decided to use removetoken command to actually scratch out the
sickly node from the cluster. I'll bootstrap is later when it's fixed.
--
View this message in
Sorry if this sounds silly, but I can't get my brain around this one: if all
nodes contain replicas, why does the cluster stream data every time I more
or remove a token? If the data is already there, what needs to be streamed?
Thanks
Maxim
--
View this message in context:
Thanks, I'll look at the configuration again.
In the meantime, I can't move the first node in the ring (after I removed
the previous node's token) -- it throws an exception and says data is being
streamed to it -- however, this is not what netstats says! Weirdness
continues...
Maxim
--
View
Thanks! It doesn't seem to have any effect on GCing dropped CFs, though.
Maxim
--
View this message in context:
http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Forcing-GC-w-o-jconsole-tp5956747p5960100.html
Sent from the cassandra-u...@incubator.apache.org mailing list
Oleg,
I'm a novice at this, but for what it's worth I can't imagine you can have a
_sustained_ 1kHz insertion rate on a single machine which also does some
reads. If I'm wrong, I'll be glad to learn that I was. It just doesn't seem
to square with a typical seek time on a hard drive.
Maxim
--
I'm trying re-partition my 4-node cluster to make the load exactly 25% on
each node.
As per recipes found in documentation, I calculate:
for x in xrange(4):
... print 2**127/4*x
...
0
42535295865117307932921825928971026432
85070591730234615865843651857942052864
Correction -- what I meant to say that I do see announcements about streaming
in the output, but these are stuck at 0%.
--
View this message in context:
My situation is similar to one described at this link:
http://stackoverflow.com/questions/4155696/how-to-trigger-manual-java-gc-from-linux-console-with-no-x11
I'm trying the following command but it fails (connection refused)
java -jar cmdline-jmxclient-0.10.3.jar - localhost:8081
OK, so I'm looking at this page:
http://wiki.apache.org/cassandra/MemtableSSTable
This looks promising:
A compaction marker is also added to obsolete sstables so they can be
deleted on startup if the server does not perform a GC before being
restarted.
So it would seem that if I restart the
Thanks for the note,
yes, I do know what files I don't need anymore. And, I do realize the
difference between grace period of CFs, and garbage collection (or at least
I hope I do).
On the face value, documentation wasn't precise enough about JVM GC taking
care of dropped CFs. I understand this
Thanks Aaron. As I remarked earlier (and it seems it not uncommon) none of
the nodes have X11 installed (I think I could arrange this, but it's a bit
of a hassle). So if I understand correctly, jconsole is a X11 app, and I'm
out of luck with that.
I would agree with you that having a proper
Greetings --
if I use multiple secondary indexes in the query, what will Cassandra do?
Some examples say it will index on first EQ and then loop on others. Does it
ever do a proper index product to avoid inner loops?
Thanks
Maxim
--
View this message in context:
Greetings,
I just used teh nodetool to force a major compaction on my cluster. It seems
like the cfs currently in service were indeed compacted, while the old test
materials (which I dropped from CLI) were still there as tombstones.
Is that the expected behavior? Hmm...
TIA.
--
View this
Thanks!
What's strange anyhow is that the GC period for these cfs expired some days
ago. I thought that a compaction would take care of these tombstones. I used
nodetool to compact.
--
View this message in context:
90 matches
Mail list logo
