batch mutation : how to delete whole row?
Hi everyone, in our test code we perform a dummy "clear" by reading all the rows and deleting them (while waiting for cassandra 0.7 & CASSANDRA-531). A couple of days ago I updated our code to perform this operation using batchMutate, but there seem to be no way to perform a deletion of the whole row, only columns. The org.apache.cassandra.thrift.Deletion object can be used with a slice predicate but if I use an empty SlicePredicate there is the obvious validation error of missing either a range or a list of column names. Is it correct that I cannot perform a row delete via batchMutation, or is there another way (apart from reading all the column names and adding multiple deletions per row)? Would it make sense to allow a Deletion object to refer to a row, or even better a RowDeletion(key) class? AFAICT there is no underlying technical blocker, but I may be wrong as usual :) -- blog en: http://www.riffraff.info blog it: http://riffraff.blogsome.com
Re: batch mutation : how to delete whole row?
This has been fixed in 0.7 (https://issues.apache.org/jira/browse/CASSANDRA-1027). Not sure this has been merged in 0.6 though. On Wed, May 26, 2010 at 9:05 AM, gabriele renzi wrote: > Hi everyone, > > in our test code we perform a dummy "clear" by reading all the rows > and deleting them (while waiting for cassandra 0.7 & CASSANDRA-531). > A couple of days ago I updated our code to perform this operation > using batchMutate, but there seem to be no way to perform a deletion > of the whole row, only columns. > > > The org.apache.cassandra.thrift.Deletion object can be used with a > slice predicate but if I use an empty SlicePredicate there is the > obvious validation error of missing either a range or a list of column > names. > > Is it correct that I cannot perform a row delete via batchMutation, or > is there another way (apart from reading all the column names and > adding multiple deletions per row)? > Would it make sense to allow a Deletion object to refer to a row, or > even better a RowDeletion(key) class? > > > AFAICT there is no underlying technical blocker, but I may be wrong as usual > :) > > > -- > blog en: http://www.riffraff.info > blog it: http://riffraff.blogsome.com >
Re: batch mutation : how to delete whole row?
You could either use 1 remove(keyspace, key, column_path, timestamp, consistency_level) per aech key, or wait till https://issues.apache.org/jira/browse/CASSANDRA-494 fixed (to use SliceRange in the Deletion) gabriele renzi wrote: > > Is it correct that I cannot perform a row delete via batchMutation, or > is there another way (apart from reading all the column names and > adding multiple deletions per row)? > Would it make sense to allow a Deletion object to refer to a row, or > even better a RowDeletion(key) class? > > > AFAICT there is no underlying technical blocker, but I may be wrong as usual > :) > >
Re: batch mutation : how to delete whole row?
On Wed, May 26, 2010 at 9:54 AM, Mishail wrote: > You could either use 1 remove(keyspace, key, column_path, timestamp, > consistency_level) per aech key, or wait till > https://issues.apache.org/jira/browse/CASSANDRA-494 fixed (to use > SliceRange in the Deletion) thanks, I'm already doing that but sending 2k commands when one would suffice is clearly a bit wasteful :) CASSANDRA-1027 seems good enough, and pretty trivial. Thanks both for your answers.
Re: Order Preserving Partitioner
Just in case you don't know: You can do range searches on keys even with Random Partitioner, you just won't get the results in order. If this is good enough for you (e.g. if you can order the results on the client, or if you just need to get the right answer, but not the right order), then you should use Random Partitioner. (I bring this up because it confused me until recently.) On Wed, May 26, 2010 at 5:14 AM, Steve Lihn wrote: > I have a question on using Order Preserving Partitioner. > > Many rowKeys in my system will be related to dates, so it seems natural to > use Order Preserving Partitioner instead of the default Random Partitioner. > However, I have been warned that special attention has to be applied for > Order Preserving Partitioner to work properly (basically to ensure a good > key distribution and avoid "hot spot") and reverting it back to Random may > not be easy. Also not every rowKey is related to dates, for these, using > Random Partitioner is okay, but there is only one place to set Partitioner. > > (Note: The intension of this warning is actually to discredit Cassandra and > persuade me not to use it.) > > It seems the choice of Partitioner is defined in the storage-conf.xml and > is a global property. My question why does it have to be a global property? > Is there a future plan to make it customizable per KeySpace (just like you > would choose hash or range partition for different table/data in RDBMS) ? > > Thanks, > Steve >
Questions regarding batch mutates and transactions
Hey guys, I originally asked this on the Hector group, but no one was sure of the answer. Can I get some feedback on this. I'd prefer to avoid having to use something like Cages if I can for most of our use cases. Long term I can see we'll need to use something like Cages, especially when it comes to complex operations such as billing. However for a majority of our uses, I think it's a bit overkill. I've used transactions heavily in the workplace on SQL based app developments. To be honest, a majority of application's I've built utilize optimistic locking, and only the atomic, consistent, and durable functionality of transactional ACID properties. To encapsulate all 3, I essentially need all writes to cassandra for a given business invocation to occur in a single write. With Spring, I would implement my own transaction manager which simply adds all mutates and delete ops to a batch mutate. When my transaction commits, I would execute the mutation on the given keyspace. Now this would only work if the following semantics apply. I've tried searching for details in Cassandra's batch mutate, but I'm not finding what I need. Here are 2 use cases as an example. Case 1: Successful update : User adds new contact Transaction Start. Biz op 1. Row is created in "contacts" and all data is added via batch mutation Biz op 2. Row is created for an SMS message is created for queueing through the SMS gateway return op 2 return op 1 Transaction Commit (batch mutate executed) Case 2. Failed update: User adds new contact Biz op 1. Row is created in "contacts" Biz op 2. Row is created for SMS message queuing. Fails due to invalid international phone number format return op 2 return op 1 Transaction is rolled back (batch mutate never executed) Now, here is where I can't find what I need in the doc. In case 1, if my mutation from biz op 2 were to fail during a batch mutate operation encapsulating all mutations, does the batch mutation as a whole not get executed, or would I still have the mutation from op 1 written to cassandra while the op 2 write fails? Thanks,
nodetool move looks stuck
I ran nodetool move on one of the nodes and it seems stuck for a few hours now. I've been able to run it successfully in the past, but this time it looks stuck. Streams shows as if there's work in progress, but the same files have been at the same position for a few hours. I've also checked the compaction manager through jmx and it's not compacting anything. Anything I can do to continue the move operation? Is this a bug? thanks $ nodetool -h 192.168.254.58 -p 9004 streams Mode: Normal Nothing streaming to /192.168.252.124 Streaming to: /192.168.254.57 /outbrain/cassandra/data/outbrain_kvdb/stream/KvImpressions-170-Filter.db 0/45587965 /outbrain/cassandra/data/outbrain_kvdb/stream/KvImpressions-170-Data.db 0/18369406636 /outbrain/cassandra/data/outbrain_kvdb/stream/KvRatings-142-Index.db 0/6163411 /outbrain/cassandra/data/outbrain_kvdb/stream/KvRatings-142-Filter.db 0/85645 /outbrain/cassandra/data/outbrain_kvdb/stream/KvRatings-142-Data.db 0/37985032 /outbrain/cassandra/data/outbrain_kvdb/stream/KvAds-108-Index.db 0/77307418 /outbrain/cassandra/data/outbrain_kvdb/stream/KvAds-108-Filter.db 0/1436005 /outbrain/cassandra/data/outbrain_kvdb/stream/KvAds-108-Data.db 0/704736670 /outbrain/cassandra/data/Keyspace1/stream/Standard1-40-Index.db 0/230280 /outbrain/cassandra/data/Keyspace1/stream/Standard1-40-Filter.db 0/20605 /outbrain/cassandra/data/Keyspace1/stream/Standard1-40-Data.db 0/202808000 /outbrain/cassandra/data/Keyspace1/stream/Standard1-41-Index.db 0/23912 /outbrain/cassandra/data/Keyspace1/stream/Standard1-41-Filter.db 0/20605 /outbrain/cassandra/data/Keyspace1/stream/Standard1-41-Data.db 0/21434973 /outbrain/cassandra/data/Keyspace1/stream/Standard1-42-Index.db 0/965028 /outbrain/cassandra/data/Keyspace1/stream/Standard1-42-Filter.db 0/20605 /outbrain/cassandra/data/Keyspace1/stream/Standard1-42-Data.db 0/865380205 Not receiving any streams. $ nodetool -h 192.168.254.57 -p 9004 streams Mode: Normal Nothing streaming to /192.168.254.58 Streaming from: /192.168.254.58 outbrain_kvdb: /outbrain/cassandra/data/outbrain_kvdb/KvImpressions-tmp-1081-Data.db 0/18369406636 outbrain_kvdb: /outbrain/cassandra/data/outbrain_kvdb/KvRatings-tmp-167-Index.db 0/6163411 outbrain_kvdb: /outbrain/cassandra/data/outbrain_kvdb/KvRatings-tmp-167-Filter.db 0/85645 outbrain_kvdb: /outbrain/cassandra/data/outbrain_kvdb/KvRatings-tmp-167-Data.db 0/37985032 outbrain_kvdb: /outbrain/cassandra/data/outbrain_kvdb/KvAds-tmp-123-Index.db 0/77307418 outbrain_kvdb: /outbrain/cassandra/data/outbrain_kvdb/KvAds-tmp-123-Filter.db 0/1436005 outbrain_kvdb: /outbrain/cassandra/data/outbrain_kvdb/KvAds-tmp-123-Data.db 0/704736670 Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-21-Index.db 0/230280 Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-21-Filter.db 0/20605 Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-21-Data.db 0/202808000 Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-22-Index.db 0/23912 Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-22-Filter.db 0/20605 Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-22-Data.db 0/21434973 Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-23-Index.db 0/965028 Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-23-Filter.db 0/20605 Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-23-Data.db 0/865380205
Re: Questions regarding batch mutates and transactions
The summary of your question is: is batch_mutate atomic in the general sense, meaning when used with multiple keys, multiple column families etc, correct? On Wed, May 26, 2010 at 12:45 PM, Todd Nine wrote: > Hey guys, > I originally asked this on the Hector group, but no one was sure of the > answer. Can I get some feedback on this. I'd prefer to avoid having to use > something like Cages if I can for most of our use cases. Long term I can > see we'll need to use something like Cages, especially when it comes to > complex operations such as billing. However for a majority of our uses, I > think it's a bit overkill. I've used transactions heavily in the workplace > on SQL based app developments. To be honest, a majority of application's > I've built utilize optimistic locking, and only the atomic, consistent, and > durable functionality of transactional ACID properties. > > To encapsulate all 3, I essentially need all writes to cassandra for a > given business invocation to occur in a single write. With Spring, I would > implement my own transaction manager which simply adds all mutates and > delete ops to a batch mutate. When my transaction commits, I would execute > the mutation on the given keyspace. Now this would only work if the > following semantics apply. I've tried searching for details in Cassandra's > batch mutate, but I'm not finding what I need. Here are 2 use cases as an > example. > > Case 1: Successful update : User adds new contact > > Transaction Start. > Biz op 1. Row is created in "contacts" and all data is added via batch > mutation > Biz op 2. Row is created for an SMS message is created for queueing > through the SMS gateway > return op 2 > return op 1 > Transaction Commit (batch mutate executed) > > Case 2. Failed update: User adds new contact > > Biz op 1. Row is created in "contacts" > Biz op 2. Row is created for SMS message queuing. Fails due to invalid > international phone number format > return op 2 > return op 1 > Transaction is rolled back (batch mutate never executed) > > > Now, here is where I can't find what I need in the doc. In case 1, if my > mutation from biz op 2 were to fail during a batch mutate operation > encapsulating all mutations, does the batch mutation as a whole not get > executed, or would I still have the mutation from op 1 written to cassandra > while the op 2 write fails? > > Thanks, >
Moving/copying columns in between ColumnFamilies
Hey All, Assume I have two ColumnFamilies in the same keyspace and I want to move or copy a range of columns (defined by a keyrange) into another columnfamily. Do you think it's somehow possible and doable with the current support of the API, if so how? Best Regards, Utku
RE: Moving/copying columns in between ColumnFamilies
There are no single API call to achieve this. It’s read and write, plus a delete (if move) API calls I guess. From: Utku Can Topçu [mailto:u...@topcu.gen.tr] Sent: Wednesday, May 26, 2010 9:09 PM To: user@cassandra.apache.org Subject: Moving/copying columns in between ColumnFamilies Hey All, Assume I have two ColumnFamilies in the same keyspace and I want to move or copy a range of columns (defined by a keyrange) into another columnfamily. Do you think it's somehow possible and doable with the current support of the API, if so how? Best Regards, Utku
Re: Moving/copying columns in between ColumnFamilies
Sorry I now realized that I used the wrong terminology. What I really meant was, moving or copying the ROWS defined by a KeyRange in between ColumnFamilies. Do you think it's doable with an efficient way? On Wed, May 26, 2010 at 3:14 PM, Dop Sun wrote: > There are no single API call to achieve this. > > > > It’s read and write, plus a delete (if move) API calls I guess. > > > > *From:* Utku Can Topçu [mailto:u...@topcu.gen.tr] > *Sent:* Wednesday, May 26, 2010 9:09 PM > *To:* user@cassandra.apache.org > *Subject:* Moving/copying columns in between ColumnFamilies > > > > Hey All, > > Assume I have two ColumnFamilies in the same keyspace and I want to move or > copy a range of columns (defined by a keyrange) into another columnfamily. > > Do you think it's somehow possible and doable with the current support of > the API, if so how? > > Best Regards, > Utku >
RE: Moving/copying columns in between ColumnFamilies
In Thrift API, I guess you need to use read/ insert and then delete to implement the move action. If you can shut the Cassandra down, maybe you can try to sstable2json to export data out, and json2sstable to import back to different column family file? I did not do it before, but I guess it may work if they have same schema definition. From: Utku Can Topçu [mailto:u...@topcu.gen.tr] Sent: Wednesday, May 26, 2010 9:18 PM To: user@cassandra.apache.org Subject: Re: Moving/copying columns in between ColumnFamilies Sorry I now realized that I used the wrong terminology. What I really meant was, moving or copying the ROWS defined by a KeyRange in between ColumnFamilies. Do you think it's doable with an efficient way? On Wed, May 26, 2010 at 3:14 PM, Dop Sun wrote: There are no single API call to achieve this. It’s read and write, plus a delete (if move) API calls I guess. From: Utku Can Topçu [mailto:u...@topcu.gen.tr] Sent: Wednesday, May 26, 2010 9:09 PM To: user@cassandra.apache.org Subject: Moving/copying columns in between ColumnFamilies Hey All, Assume I have two ColumnFamilies in the same keyspace and I want to move or copy a range of columns (defined by a keyrange) into another columnfamily. Do you think it's somehow possible and doable with the current support of the API, if so how? Best Regards, Utku
Re: Avro Example Code
I've got a mostly working Avro server and client for HBase at http://github.com/hammer/hbase-trunk-with-avro and http://github.com/hammer/pyhbase. If you replace "scan" with "slice", it shouldn't be too much different for Cassandra... On Mon, May 17, 2010 at 10:31 AM, Wellman, David wrote: > I spent the weekend working with avro and some java junit tests. I still > have a lot of learning to do, but if others would like to use, add to or > improve upon the tests then I would appricate the feedback and help. > > David Wellman > > > On May 17, 2010, at 10:16 AM, "Eric Evans" wrote: > > On Fri, 2010-05-14 at 14:52 -0600, David Wellman wrote: >> >>> Does anyone have a good link or example code that we can use to spike on >>> Avro with Cassandra? >>> >> >> If you're using Python, the best place to look is the functional tests >> (see test/system), otherwise, Patrick's quick start >> (http://bit.ly/32T6Mk). >> >> As Gary mentioned already, it's still very rough. After all the recent >> changes in trunk (dropping the keyspace arg, binary keys, etc), it's >> just barely back to a state where you can read/write, and only then via >> get/insert (i.e. no slicing, batch_mutate, etc). >> >> -- >> Eric Evans >> eev...@rackspace.com >> >>
Re: Avro Example Code
Fantastic! Thank you. On May 26, 2010, at 8:38 AM, Jeff Hammerbacher wrote: > I've got a mostly working Avro server and client for HBase at > http://github.com/hammer/hbase-trunk-with-avro and > http://github.com/hammer/pyhbase. If you replace "scan" with "slice", it > shouldn't be too much different for Cassandra... > > On Mon, May 17, 2010 at 10:31 AM, Wellman, David wrote: > I spent the weekend working with avro and some java junit tests. I still > have a lot of learning to do, but if others would like to use, add to or > improve upon the tests then I would appricate the feedback and help. > > David Wellman > > > On May 17, 2010, at 10:16 AM, "Eric Evans" wrote: > > On Fri, 2010-05-14 at 14:52 -0600, David Wellman wrote: > Does anyone have a good link or example code that we can use to spike on Avro > with Cassandra? > > If you're using Python, the best place to look is the functional tests > (see test/system), otherwise, Patrick's quick start > (http://bit.ly/32T6Mk). > > As Gary mentioned already, it's still very rough. After all the recent > changes in trunk (dropping the keyspace arg, binary keys, etc), it's > just barely back to a state where you can read/write, and only then via > get/insert (i.e. no slicing, batch_mutate, etc). > > -- > Eric Evans > eev...@rackspace.com > >
using more than 50% of disk space
We're investigating Cassandra, and we are looking for a way to get Cassandra use more than 50% of it's data disks. Is this possible? For major compactions, it looks like we can use more than 50% of the disk if we use multiple similarly sized column families. If we had 10 column families of the same size, we could use 90% of the disk, since a major compaction would only need as much free space as the largest column family (in reality we would use less). Is that right? For bootstrapping new nodes, it looks like adding a new node will require that an existing node does anti-compaction. This anti-compaction could take nearly 50% of the disk. Is there a way around this? Is there anything else that would prevent us from using more than 50% of the data disk. Thanks, Sean
RE: using more than 50% of disk space
See https://issues.apache.org/jira/browse/CASSANDRA-579 for some background here: I was just about to start working on this one, but it won't make it in until 0.7. -Original Message- From: "Sean Bridges" Sent: Wednesday, May 26, 2010 11:50am To: user@cassandra.apache.org Subject: using more than 50% of disk space We're investigating Cassandra, and we are looking for a way to get Cassandra use more than 50% of it's data disks. Is this possible? For major compactions, it looks like we can use more than 50% of the disk if we use multiple similarly sized column families. If we had 10 column families of the same size, we could use 90% of the disk, since a major compaction would only need as much free space as the largest column family (in reality we would use less). Is that right? For bootstrapping new nodes, it looks like adding a new node will require that an existing node does anti-compaction. This anti-compaction could take nearly 50% of the disk. Is there a way around this? Is there anything else that would prevent us from using more than 50% of the data disk. Thanks, Sean
Two threads inserting columns into same key followed by read gets unexpected results
Hi, I'm seeing a problem with inserting columns into one key using multiple threads and I'm not sure if it's a bug or if it's my misunderstanding of how insert/get_slice should work. My setup is that I have two separate client processes, each with a single thread, writing concurrently to Cassandra and each process is doing the same thing: read columns from one key and then use one of the column names as the basis for another row key, and then insert into that new row key a column whose name is unique to the process. Each process then immediately reads back the columns on that row key. It's the reading back part that's showing inconsistent behavior: most of the time if Thread1 writes Column1 and Thread2 writes Column2 at about the same time, things are consistent: Thread1 sees either Column1 or it sees Column1 and Column 2 (same for Thread2). Both of those cases are expected and indeed comprise 99% of the results. Sometimes, however, Thread1 sees NO columns, and sometimes it sees Column2 (instead of Column1, as I'd expect). Debug logs show that in failing cases, Thread1 completes its insert/get_slice sequence BEFORE Thread2 starts its insert/get_slice, but (in this example) Thread2 would get back NO columns. I turned on debug level of logging on Cassandra and I saw that in at least one case the two client requests are coming in at the exact same millisecond so I'm wondering if there's some concurrency issue on the server. I'm running linux with Java build 1.6.0_20-b02. Cassandra is version 0.6.1 and I have one node in the cluster. The consistency level for writing is ALL and for reading it's ONE. The clients are using the Java Hector interface. Both of the two client processes are on the same machine as the Cassandra server (running on a dual-core processor). This fails with 0.6.2 code also. Am I wrong in thinking that an insert on a column with consistency level ALL followed immediately by a get_slice should include that column? Thanks Scott McCarty
Re: Order Preserving Partitioner
Correct me if I'm wrong here. Even though you can get your results with Random Partitioner, it's a lot less efficient if you're going across different machines to get your results. If you're doing a lot of range queries, it makes sense to have things ordered sequentially so that if you do need to go to disk, the reads will be faster, rather than lots of random reads across your system. It's also my understanding that if you go with the OPP, you could hash your key yourself using md5 or sha-1 to effectively get random partitioning. So it's a bit of a pain, but not impossible to do a split between OPP and RP for your different columnfamily/keyspaces. On May 26, 2010, at 2:32 AM, David Boxenhorn wrote: > Just in case you don't know: You can do range searches on keys even with > Random Partitioner, you just won't get the results in order. If this is good > enough for you (e.g. if you can order the results on the client, or if you > just need to get the right answer, but not the right order), then you should > use Random Partitioner. > > (I bring this up because it confused me until recently.) > > On Wed, May 26, 2010 at 5:14 AM, Steve Lihn wrote: > I have a question on using Order Preserving Partitioner. > > Many rowKeys in my system will be related to dates, so it seems natural to > use Order Preserving Partitioner instead of the default Random Partitioner. > However, I have been warned that special attention has to be applied for > Order Preserving Partitioner to work properly (basically to ensure a good key > distribution and avoid "hot spot") and reverting it back to Random may not be > easy. Also not every rowKey is related to dates, for these, using Random > Partitioner is okay, but there is only one place to set Partitioner. > > (Note: The intension of this warning is actually to discredit Cassandra and > persuade me not to use it.) > > It seems the choice of Partitioner is defined in the storage-conf.xml and is > a global property. My question why does it have to be a global property? Is > there a future plan to make it customizable per KeySpace (just like you would > choose hash or range partition for different table/data in RDBMS) ? > > Thanks, > Steve >
Re: using more than 50% of disk space
So after CASSANDRA-579, anti compaction won't be done on the source node, and we can use more than 50% of the disk space if we use multiple column families? Thanks, Sean On Wed, May 26, 2010 at 10:01 AM, Stu Hood wrote: > See https://issues.apache.org/jira/browse/CASSANDRA-579 for some > background here: I was just about to start working on this one, but it won't > make it in until 0.7. > > > -Original Message- > From: "Sean Bridges" > Sent: Wednesday, May 26, 2010 11:50am > To: user@cassandra.apache.org > Subject: using more than 50% of disk space > > We're investigating Cassandra, and we are looking for a way to get > Cassandra > use more than 50% of it's data disks. Is this possible? > > For major compactions, it looks like we can use more than 50% of the disk > if > we use multiple similarly sized column families. If we had 10 column > families of the same size, we could use 90% of the disk, since a major > compaction would only need as much free space as the largest column family > (in reality we would use less). Is that right? > > For bootstrapping new nodes, it looks like adding a new node will require > that an existing node does anti-compaction. This anti-compaction could > take > nearly 50% of the disk. Is there a way around this? > > Is there anything else that would prevent us from using more than 50% of > the > data disk. > > Thanks, > > Sean > > >
Subscribe
Sent from my iPhone
Doing joins between column familes
So I am not sure if you guys are familiar with OCM . Basically it is
an ORM for Cassandra. Been testing it
So I have created model that has the following object relationship.
OCM generates the code from this that allows me to do easy
programmatic query from Java to Cassandra.
Object1-(Many2Many)->Object2-(Many2Many)->Object3-(Many2Many)->Object4-
(Many2Many)->Node
So my app gets the NODE and tries to query the dependency relationship
from Node->Object4->Object3->Object2->Object1.
I have compared the performance between Cassandra(with OCM) vs DB2.
The result is not very encouraging since the DB2 performance is
showing at least 3X faster than Cassandra. DB2 basically is just a
single call with a number of inner joins ..
Looking at the code, I think we might get a better performance if
somehow we can do the joins between objects within Cassandra server
rather than the client side. Right now , I am basically doing the
following.
Node node = new Node(connection,"nodeidentifier");
node.loadInfo(); // going to the wire ...?
node.loadObject4(); // this goes to the wire too ..
object4Keys = node.getObject4().getColumns.keys();
while(object4Keys.hasNextElement)
{
object4Key = object4Key.nextElement();
object4 = node.getObject4().get(object4Key);
object4.loadInfo();// this goes to the wire
too ..
object4.loadObject3(); // this goes to the wire too ..
object3Keys = object4.getObject3().getColumns.keys();
while(object3Keys.hasNextElement)
{
object3Key = object3Key.nextElement();
object3 = node.getObject4().get(object4Key);
object3.loadInfo(); // this goes to the wire too ..
object3.loadObject2();// this goes to the wire too ..
...
..
... until you get object1//
}
I think there is a lot of going back and forth between the client and
cassandra and if we can only move the relationship joins to Cassandra
server I think we can minimize the latency and improve the overall
performance on the query.
Is there a way to do a join across ColumnFamilies in Cassandra ?
Re: nodetool move looks stuck
Are there any exceptions in the log like the one in https://issues.apache.org/jira/browse/CASSANDRA-1019 ? If so you'll need to restart the moving node and try again. On Wed, May 26, 2010 at 3:54 AM, Ran Tavory wrote: > I ran nodetool move on one of the nodes and it seems stuck for a few hours > now. > I've been able to run it successfully in the past, but this time it looks > stuck. > Streams shows as if there's work in progress, but the same files have been > at the same position for a few hours. > I've also checked the compaction manager through jmx and it's not compacting > anything. > Anything I can do to continue the move operation? Is this a bug? > thanks > > $ nodetool -h 192.168.254.58 -p 9004 streams > Mode: Normal > Nothing streaming to /192.168.252.124 > Streaming to: /192.168.254.57 > /outbrain/cassandra/data/outbrain_kvdb/stream/KvImpressions-170-Filter.db > 0/45587965 > /outbrain/cassandra/data/outbrain_kvdb/stream/KvImpressions-170-Data.db > 0/18369406636 > /outbrain/cassandra/data/outbrain_kvdb/stream/KvRatings-142-Index.db > 0/6163411 > /outbrain/cassandra/data/outbrain_kvdb/stream/KvRatings-142-Filter.db > 0/85645 > /outbrain/cassandra/data/outbrain_kvdb/stream/KvRatings-142-Data.db > 0/37985032 > /outbrain/cassandra/data/outbrain_kvdb/stream/KvAds-108-Index.db > 0/77307418 > /outbrain/cassandra/data/outbrain_kvdb/stream/KvAds-108-Filter.db > 0/1436005 > /outbrain/cassandra/data/outbrain_kvdb/stream/KvAds-108-Data.db > 0/704736670 > /outbrain/cassandra/data/Keyspace1/stream/Standard1-40-Index.db 0/230280 > /outbrain/cassandra/data/Keyspace1/stream/Standard1-40-Filter.db 0/20605 > /outbrain/cassandra/data/Keyspace1/stream/Standard1-40-Data.db > 0/202808000 > /outbrain/cassandra/data/Keyspace1/stream/Standard1-41-Index.db 0/23912 > /outbrain/cassandra/data/Keyspace1/stream/Standard1-41-Filter.db 0/20605 > /outbrain/cassandra/data/Keyspace1/stream/Standard1-41-Data.db 0/21434973 > /outbrain/cassandra/data/Keyspace1/stream/Standard1-42-Index.db 0/965028 > /outbrain/cassandra/data/Keyspace1/stream/Standard1-42-Filter.db 0/20605 > /outbrain/cassandra/data/Keyspace1/stream/Standard1-42-Data.db > 0/865380205 > Not receiving any streams. > $ nodetool -h 192.168.254.57 -p 9004 streams > Mode: Normal > Nothing streaming to /192.168.254.58 > Streaming from: /192.168.254.58 > outbrain_kvdb: > /outbrain/cassandra/data/outbrain_kvdb/KvImpressions-tmp-1081-Data.db > 0/18369406636 > outbrain_kvdb: > /outbrain/cassandra/data/outbrain_kvdb/KvRatings-tmp-167-Index.db 0/6163411 > outbrain_kvdb: > /outbrain/cassandra/data/outbrain_kvdb/KvRatings-tmp-167-Filter.db 0/85645 > outbrain_kvdb: > /outbrain/cassandra/data/outbrain_kvdb/KvRatings-tmp-167-Data.db 0/37985032 > outbrain_kvdb: > /outbrain/cassandra/data/outbrain_kvdb/KvAds-tmp-123-Index.db 0/77307418 > outbrain_kvdb: > /outbrain/cassandra/data/outbrain_kvdb/KvAds-tmp-123-Filter.db 0/1436005 > outbrain_kvdb: > /outbrain/cassandra/data/outbrain_kvdb/KvAds-tmp-123-Data.db 0/704736670 > Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-21-Index.db > 0/230280 > Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-21-Filter.db > 0/20605 > Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-21-Data.db > 0/202808000 > Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-22-Index.db > 0/23912 > Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-22-Filter.db > 0/20605 > Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-22-Data.db > 0/21434973 > Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-23-Index.db > 0/965028 > Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-23-Filter.db > 0/20605 > Keyspace1: /outbrain/cassandra/data/Keyspace1/Standard1-tmp-23-Data.db > 0/865380205 > > > -- Jonathan Ellis Project Chair, Apache Cassandra co-founder of Riptano, the source for professional Cassandra support http://riptano.com
Re: Order Preserving Partitioner
I don't think that queries on a key range are valid unless you are using OPP. As far as hashing the key for OPP goes, I take it to be the same a not using OPP. It's really a matter of where it gets done, but it has much the same effect. (I think) Jonathan On Wed, May 26, 2010 at 12:51 PM, Peter Hsu wrote: > Correct me if I'm wrong here. Even though you can get your results with > Random Partitioner, it's a lot less efficient if you're going across > different machines to get your results. If you're doing a lot of range > queries, it makes sense to have things ordered sequentially so that if you > do need to go to disk, the reads will be faster, rather than lots of random > reads across your system. > It's also my understanding that if you go with the OPP, you could hash your > key yourself using md5 or sha-1 to effectively get random partitioning. So > it's a bit of a pain, but not impossible to do a split between OPP and RP > for your different columnfamily/keyspaces. > On May 26, 2010, at 2:32 AM, David Boxenhorn wrote: > > Just in case you don't know: You can do range searches on keys even with > Random Partitioner, you just won't get the results in order. If this is good > enough for you (e.g. if you can order the results on the client, or if you > just need to get the right answer, but not the right order), then you should > use Random Partitioner. > > (I bring this up because it confused me until recently.) > > On Wed, May 26, 2010 at 5:14 AM, Steve Lihn wrote: >> >> I have a question on using Order Preserving Partitioner. >> >> Many rowKeys in my system will be related to dates, so it seems natural to >> use Order Preserving Partitioner instead of the default Random Partitioner. >> However, I have been warned that special attention has to be applied for >> Order Preserving Partitioner to work properly (basically to ensure a good >> key distribution and avoid "hot spot") and reverting it back to Random may >> not be easy. Also not every rowKey is related to dates, for these, using >> Random Partitioner is okay, but there is only one place to set Partitioner. >> >> (Note: The intension of this warning is actually to discredit Cassandra >> and persuade me not to use it.) >> >> It seems the choice of Partitioner is defined in the storage-conf.xml and >> is a global property. My question why does it have to be a global property? >> Is there a future plan to make it customizable per KeySpace (just like you >> would choose hash or range partition for different table/data in RDBMS) ? >> >> Thanks, >> Steve > > >
Re: Doing joins between column familes
On Wed, May 26, 2010 at 7:45 PM, Dodong Juan wrote: > > So I am not sure if you guys are familiar with OCM . Basically it is an ORM > for Cassandra. Been testing it > In case anyone is interested I have posted a reply on the OCM issue tracker where this was also raised. http://github.com/charliem/OCM/issues/closed#issue/5/comment/254717 Charlie M
Re: Error reporting Key cache hit rate with cfstats or with JMX
If I disable row cache the numbers look good - key cache hit rate is > 0, so it seems to be related to row cache. Interestingly, after running for a really long time and with both row and keys caches I do start to see Key cache hit rate > 0 but the numbers are so small that it doesn't make sense. I have capacity for 10M keys and 10M rows, the number of cached keys is ~5M and very similarly the number of cached rows is also ~5M, however the hit rates are very different, 0.7 for rows and 0.006 for keys. I'd expect the keys hit rate to be identical since none of them reached the limit yet. Key cache capacity: 1000 Key cache size: 5044097 Key cache hit rate: 0.0062089764058896576 Row cache capacity: 1000 Row cache size: 5057231 Row cache hit rate: 0.7361241352465543 On Tue, May 25, 2010 at 3:43 PM, Jonathan Ellis wrote: > What happens if you disable row cache? > > On Tue, May 25, 2010 at 4:53 AM, Ran Tavory wrote: > > It seems there's an error reporting the Key cache hit rate. The value is > > always 0.0 and I have a feeling it's incorrect. This is seen both by > using > > notetool cfstats as well as accessing JMX directly > > > (org.apache.cassandra.db:type=Caches,keyspace=outbrain_kvdb,cache=KvAdsKeyCache > > RecentHitRate) > >> RowsCached="1000" > > KeysCached="1000"/> > > Column Family: KvAds > > SSTable count: 7 > > Space used (live): 1288942061 > > Space used (total): 1559831566 > > Memtable Columns Count: 73698 > > Memtable Data Size: 17121092 > > Memtable Switch Count: 33 > > Read Count: 3614433 > > Read Latency: 0.068 ms. > > Write Count: 3503269 > > Write Latency: 0.024 ms. > > Pending Tasks: 0 > > Key cache capacity: 1000 > > Key cache size: 619624 > > Key cache hit rate: 0.0 > > Row cache capacity: 1000 > > Row cache size: 447154 > > Row cache hit rate: 0.8460295730014572 > > Compacted row minimum size: 387 > > Compacted row maximum size: 31430 > > Compacted row mean size: 631 > > The Row cache hit rate looks good, 0.8 but Key cache hit rate always > seems > > to be 0.0 while the number of unique keys stays about 619624 for quite a > > while. > > Is it a real caching problem or just a reporting glitch? > > > > -- > Jonathan Ellis > Project Chair, Apache Cassandra > co-founder of Riptano, the source for professional Cassandra support > http://riptano.com >
Re: Doing joins between column familes
I wrote some Iterable<*> methods to do this for column families that share key structure with OPP. It is on the hector examples page. Caveat emptor. It does iterative chunking of the working set for each column family, so that you can set the nominal transfer size when you construct the Iterator/Iterable. I've been very happy with the performance of it, even over large ranges of keys. This is with OrderPreservingPartitioner because of other requirements, so it may not be a good example for comparison with a random partitioner, which is preferred. Doing joins as such on the server works against the basic design of Cassandra. The server does a few things very well only because it isn't overloaded with extra faucets and kitchen sinks. However, I'd like to be able to load auxiliary classes into the server runtime in a modular way, just for things like this. Maybe we'll get that someday. My impression is that there is much more common key structure in a workable Cassandra storage layout than in a conventional ER model. This is the nature of the beast when you are organizing your information more according to access patterns than fully normal relationships. That is one of the fundamental design trade-offs of using a hash structure over a schema. Having something that lets you deploy a fully normal schema on a hash store can be handy, but it can also obscure the way that your application indirectly exercises the storage layer. The end-result may be that the layout is less friendly to the underlying mechanisms of Cassandra. I'm not saying that it is bad to have a tool to do this, only that it can make it easy to avoid thinking about Cassandra storage in terms of what it really is. There may be ways to optimize the OCM queries, but that takes you down the road of query optimization, which can be quite nebulous. My gut instinct is to focus more on the layout, using aggregate keys and common key structure where you can, so that you can take advantage of the parallel queries more of the time. On Wed, May 26, 2010 at 3:13 PM, Charlie Mason wrote: > On Wed, May 26, 2010 at 7:45 PM, Dodong Juan wrote: >> >> So I am not sure if you guys are familiar with OCM . Basically it is an ORM >> for Cassandra. Been testing it >> > > In case anyone is interested I have posted a reply on the OCM issue > tracker where this was also raised. > > http://github.com/charliem/OCM/issues/closed#issue/5/comment/254717 > > > Charlie M >
Best Timestamp?
What is the best timestamp to use while using Cassandra with C#? I have been using DateTime.Now.Ticks, but I have seen others using different things. Thanks.
Re: Best Timestamp?
On 26 May 2010 22:42, Steven Haar wrote: > What is the best timestamp to use while using Cassandra with C#? I have > been using DateTime.Now.Ticks, but I have seen others using different > things. > The standard that most clients seem to use is epoch-microseconds, or microseconds since midnight GMT 1/1/1970 The trick is to get all clients to use the same value, so it only ever increases :) Mark
Re: Best Timestamp?
Right, in C# this would be (not the most efficient way, but you get the idea): long timestamp = (DateTime.Now.Ticks - new DateTime(1970, 1, 1).Ticks)/10; On Wed, May 26, 2010 at 4:50 PM, Mark Robson wrote: > On 26 May 2010 22:42, Steven Haar wrote: > >> What is the best timestamp to use while using Cassandra with C#? I have >> been using DateTime.Now.Ticks, but I have seen others using different >> things. >> > > The standard that most clients seem to use is epoch-microseconds, or > microseconds since midnight GMT 1/1/1970 > > The trick is to get all clients to use the same value, so it only ever > increases :) > > Mark >
Re: Best Timestamp?
On 26 May 2010 22:56, Miguel Verde wrote: > Right, in C# this would be (not the most efficient way, but you get the > idea): > long timestamp = (DateTime.Now.Ticks - new DateTime(1970, 1, 1).Ticks)/10; > > > Yeah, you're fine provided: a) All your client applications (which perform writes) are consistent and b) Your infrastructure's clocks are fairly in sync. The latter point is important to note, as a single app server node which has a clock which is accidentally far in the future could wreck your database. Mark
Re: Error reporting Key cache hit rate with cfstats or with JMX
It sure sounds like you're seeing the "my row cache contains the entire hot data set, so the key cache only gets the cold reads" effect. On Wed, May 26, 2010 at 2:54 PM, Ran Tavory wrote: > If I disable row cache the numbers look good - key cache hit rate is > 0, so > it seems to be related to row cache. > Interestingly, after running for a really long time and with both row and > keys caches I do start to see Key cache hit rate > 0 but the numbers are so > small that it doesn't make sense. > I have capacity for 10M keys and 10M rows, the number of cached keys is ~5M > and very similarly the number of cached rows is also ~5M, however the hit > rates are very different, 0.7 for rows and 0.006 for keys. I'd expect the > keys hit rate to be identical since none of them reached the limit yet. > Key cache capacity: 1000 > Key cache size: 5044097 > Key cache hit rate: 0.0062089764058896576 > Row cache capacity: 1000 > Row cache size: 5057231 > Row cache hit rate: 0.7361241352465543 > > > On Tue, May 25, 2010 at 3:43 PM, Jonathan Ellis wrote: >> >> What happens if you disable row cache? >> >> On Tue, May 25, 2010 at 4:53 AM, Ran Tavory wrote: >> > It seems there's an error reporting the Key cache hit rate. The value is >> > always 0.0 and I have a feeling it's incorrect. This is seen both by >> > using >> > notetool cfstats as well as accessing JMX directly >> > >> > (org.apache.cassandra.db:type=Caches,keyspace=outbrain_kvdb,cache=KvAdsKeyCache >> > RecentHitRate) >> > > > RowsCached="1000" >> > KeysCached="1000"/> >> > Column Family: KvAds >> > SSTable count: 7 >> > Space used (live): 1288942061 >> > Space used (total): 1559831566 >> > Memtable Columns Count: 73698 >> > Memtable Data Size: 17121092 >> > Memtable Switch Count: 33 >> > Read Count: 3614433 >> > Read Latency: 0.068 ms. >> > Write Count: 3503269 >> > Write Latency: 0.024 ms. >> > Pending Tasks: 0 >> > Key cache capacity: 1000 >> > Key cache size: 619624 >> > Key cache hit rate: 0.0 >> > Row cache capacity: 1000 >> > Row cache size: 447154 >> > Row cache hit rate: 0.8460295730014572 >> > Compacted row minimum size: 387 >> > Compacted row maximum size: 31430 >> > Compacted row mean size: 631 >> > The Row cache hit rate looks good, 0.8 but Key cache hit rate always >> > seems >> > to be 0.0 while the number of unique keys stays about 619624 for quite a >> > while. >> > Is it a real caching problem or just a reporting glitch? >> >> >> >> -- >> Jonathan Ellis >> Project Chair, Apache Cassandra >> co-founder of Riptano, the source for professional Cassandra support >> http://riptano.com > > -- Jonathan Ellis Project Chair, Apache Cassandra co-founder of Riptano, the source for professional Cassandra support http://riptano.com
Thoughts on adding complex queries to Cassandra
Are there any thoughts on adding a more complex query to Cassandra? At a high level what I'm wondering is: Would it be possible/desirable/in keeping with the Cassandra plan, to add something like a javascript blob on to a get range slice etc, that does some further filtering on the results before returning them. The goal being to trade off some CPU on Cassandra for network bandwidth. -JD
RE: Thoughts on adding complex queries to Cassandra
I'm very curious on this topic as well. Mainly, I'd like to know is this functionality handled through Map/Reduce HADOOP operations? Nick From: Jeremy Davis [mailto:jerdavis.cassan...@gmail.com] Sent: Wednesday, May 26, 2010 3:31 PM To: user@cassandra.apache.org Subject: Thoughts on adding complex queries to Cassandra Are there any thoughts on adding a more complex query to Cassandra? At a high level what I'm wondering is: Would it be possible/desirable/in keeping with the Cassandra plan, to add something like a javascript blob on to a get range slice etc, that does some further filtering on the results before returning them. The goal being to trade off some CPU on Cassandra for network bandwidth. -JD
Cassandra's 2GB row limit and indexing
Hi all,
I'm currently looking at new database options for a URL shortener in order
to scale well with increased traffic as we add new features. Cassandra seems
to be a good fit for many of our requirements, but I'm struggling a bit to
find ways of designing certain indexes in Cassandra due to its 2GB row
limit.
The easiest example of this is that I'd like to create an index by the
domain that shortened URLs are linking to, mostly for spam control so it's
easy to grab all the links to any given domain. As far as I can tell the
typical way to do this in Cassandra is something like: -
DOMAIN = { //columnfamily
thing.com { //row key
timestamp: "shorturl567", //column name: value
timestamp: "shorturl144",
timestamp: "shorturl112",
...
}
somethingelse.com {
timestamp: "shorturl817",
...
}
}
The values here are keys for another columnfamily containing various data on
shortened URLs.
The problem with this approach is that a popular domain (e.g. blogspot.com)
could be used in many millions of shortened URLs, so would have that many
columns and hit the row size limit mentioned at
http://wiki.apache.org/cassandra/CassandraLimitations.
Does anyone know an effective way to design this type of one-to-many index
around this limitation (could be something obvious I'm missing)? If not, are
the changes proposed for
https://issues.apache.org/jira/browse/CASSANDRA-16likely to make this
type of design workable?
Thanks in advance for any advice,
Richard
Re: Cassandra's 2GB row limit and indexing
The example is a little confusing.
.. but ..
1) "sharding"
You can square the capacity by having a 2-level map.
CF1->row->value->CF2->row->value
This means finding some natural subgrouping or hash that provides a
good distribution.
2) "hashing"
You can also use some additional key hashing to spread the rows over a
wider space:
Find a delimiter that works for you and identify the row that owns it
by "domain" + "delimiter" + hash(domain) modulo some divisor, for
example.
3) "overflow"
You can implement some overflow logic to create overflow rows which
act like (2), but is less sparse
while count(columns) for candidate row > some threshold, try row +
"delimiter" + subrow++
This is much easier when you are streaming data in, as opposed to
poking the random value here and there
Just some ideas. I'd go with 2, and find a way to adjust the modulo to
minimize the row spread. 2) isn't guaranteed to provide uniformity,
but 3) isn't guaranteed to provide very good performance. Perhaps a
combination of them both? The count is readily accessible, so it may
provide for some informed choices at run time. I'm assuming your
column sizes are fairly predictable.
Has anybody else tackled this before?
On Wed, May 26, 2010 at 8:52 PM, Richard West wrote:
> Hi all,
>
> I'm currently looking at new database options for a URL shortener in order
> to scale well with increased traffic as we add new features. Cassandra seems
> to be a good fit for many of our requirements, but I'm struggling a bit to
> find ways of designing certain indexes in Cassandra due to its 2GB row
> limit.
>
> The easiest example of this is that I'd like to create an index by the
> domain that shortened URLs are linking to, mostly for spam control so it's
> easy to grab all the links to any given domain. As far as I can tell the
> typical way to do this in Cassandra is something like: -
>
> DOMAIN = { //columnfamily
> thing.com { //row key
> timestamp: "shorturl567", //column name: value
> timestamp: "shorturl144",
> timestamp: "shorturl112",
> ...
> }
> somethingelse.com {
> timestamp: "shorturl817",
> ...
> }
> }
>
> The values here are keys for another columnfamily containing various data on
> shortened URLs.
>
> The problem with this approach is that a popular domain (e.g. blogspot.com)
> could be used in many millions of shortened URLs, so would have that many
> columns and hit the row size limit mentioned at
> http://wiki.apache.org/cassandra/CassandraLimitations.
>
> Does anyone know an effective way to design this type of one-to-many index
> around this limitation (could be something obvious I'm missing)? If not, are
> the changes proposed for https://issues.apache.org/jira/browse/CASSANDRA-16
> likely to make this type of design workable?
>
> Thanks in advance for any advice,
>
> Richard
>
Re: Anyone using hadoop/MapReduce integration currently?
2010/5/26 Utku Can Topçu > Hi Jeremy, > > > > Why are you using Cassandra versus using data stored in HDFS or HBase? > - I'm thinking of using it for realtime streaming of user data. While > streaming the requests, I'm also using Lucandra for indexing the data in > realtime. It's a better option when you compare it with HBase or the native > HDFS flat files, because of low latency in writes. i'm interested in realtime index with lucandra. but how to intersect posting list from multiple terms with cansandra. if through the network, i think it is very inefficient > > > > Is there anything holding you back from using it (if you would like to > use it but currently cannot)? > > My answer to this would be: > - The current integration only supports the whole range of the CF to be > input for the map phase, it would be way much better if the InputFormat had > means of support for a KeyRange. > > Best Regards, > Utku > > > On Tue, May 25, 2010 at 6:35 PM, Jeremy Hanna > wrote: > >> I'll be doing a presentation on Cassandra's (0.6+) hadoop integration next >> week. Is anyone currently using MapReduce or the initial Pig integration? >> >> (If you're unaware of such integration, see >> http://wiki.apache.org/cassandra/HadoopSupport) >> >> If so, could you post to this thread on how you're using it or planning on >> using it (if not covered by the shroud of secrecy)? >> >> e.g. >> What is the use case? >> >> Why are you using Cassandra versus using data stored in HDFS or HBase? >> >> Are you using a separate Hadoop cluster to run the MR jobs on, or perhaps >> are you running the Job Tracker and Task Trackers on Cassandra nodes? >> >> Is there anything holding you back from using it (if you would like to use >> it but currently cannot)? >> >> Thanks! > > >
Re: Error reporting Key cache hit rate with cfstats or with JMX
so the row cache contains both rows and keys and if I have large enough row cache (in particular if row cache size equals key cache size) then it's just wasteful to keep another key cache and I should eliminate the key-cache, correct? On Thu, May 27, 2010 at 1:21 AM, Jonathan Ellis wrote: > It sure sounds like you're seeing the "my row cache contains the > entire hot data set, so the key cache only gets the cold reads" > effect. > > On Wed, May 26, 2010 at 2:54 PM, Ran Tavory wrote: > > If I disable row cache the numbers look good - key cache hit rate is > 0, > so > > it seems to be related to row cache. > > Interestingly, after running for a really long time and with both row and > > keys caches I do start to see Key cache hit rate > 0 but the numbers are > so > > small that it doesn't make sense. > > I have capacity for 10M keys and 10M rows, the number of cached keys is > ~5M > > and very similarly the number of cached rows is also ~5M, however the hit > > rates are very different, 0.7 for rows and 0.006 for keys. I'd expect the > > keys hit rate to be identical since none of them reached the limit yet. > > Key cache capacity: 1000 > > Key cache size: 5044097 > > Key cache hit rate: 0.0062089764058896576 > > Row cache capacity: 1000 > > Row cache size: 5057231 > > Row cache hit rate: 0.7361241352465543 > > > > > > On Tue, May 25, 2010 at 3:43 PM, Jonathan Ellis > wrote: > >> > >> What happens if you disable row cache? > >> > >> On Tue, May 25, 2010 at 4:53 AM, Ran Tavory wrote: > >> > It seems there's an error reporting the Key cache hit rate. The value > is > >> > always 0.0 and I have a feeling it's incorrect. This is seen both by > >> > using > >> > notetool cfstats as well as accessing JMX directly > >> > > >> > > (org.apache.cassandra.db:type=Caches,keyspace=outbrain_kvdb,cache=KvAdsKeyCache > >> > RecentHitRate) > >> >>> > RowsCached="1000" > >> > KeysCached="1000"/> > >> > Column Family: KvAds > >> > SSTable count: 7 > >> > Space used (live): 1288942061 > >> > Space used (total): 1559831566 > >> > Memtable Columns Count: 73698 > >> > Memtable Data Size: 17121092 > >> > Memtable Switch Count: 33 > >> > Read Count: 3614433 > >> > Read Latency: 0.068 ms. > >> > Write Count: 3503269 > >> > Write Latency: 0.024 ms. > >> > Pending Tasks: 0 > >> > Key cache capacity: 1000 > >> > Key cache size: 619624 > >> > Key cache hit rate: 0.0 > >> > Row cache capacity: 1000 > >> > Row cache size: 447154 > >> > Row cache hit rate: 0.8460295730014572 > >> > Compacted row minimum size: 387 > >> > Compacted row maximum size: 31430 > >> > Compacted row mean size: 631 > >> > The Row cache hit rate looks good, 0.8 but Key cache hit rate always > >> > seems > >> > to be 0.0 while the number of unique keys stays about 619624 for quite > a > >> > while. > >> > Is it a real caching problem or just a reporting glitch? > >> > >> > >> > >> -- > >> Jonathan Ellis > >> Project Chair, Apache Cassandra > >> co-founder of Riptano, the source for professional Cassandra support > >> http://riptano.com > > > > > > > > -- > Jonathan Ellis > Project Chair, Apache Cassandra > co-founder of Riptano, the source for professional Cassandra support > http://riptano.com >
Continuously increasing RAM usage
We're seeing RAM usage continually climb until eventually, cassandra becomes unresponsive. The JVM isn't OOM'ing. It has only committed 14/24GB of memory. So, I am assuming that the memory usage is related to mmap'd IO. Fair assumption? I tried setting the IO mode to standard, but it seemed to be a little slower and couldn't get the machine to come back online with adequate read performance, so I set it back. I'll have to write a solid cache warming script if I'm going to try that again. Any other ideas for what might be causing the issue? Is there something I should monitor or look at next time it happens? Thanks
Re: Anyone using hadoop/MapReduce integration currently?
On Tue, May 25, 2010 at 6:35 PM, Jeremy Hanna wrote: > What is the use case? we end up with messed up data in the database, we run a mapreduce job to find irregular data from time to time. > Why are you using Cassandra versus using data stored in HDFS or HBase? as of now our mapreduce task is only used for "fixing" cassandra so the question is useless :) > Are you using a separate Hadoop cluster to run the MR jobs on, or perhaps are > you running the Job Tracker and Task Trackers on Cassandra nodes? separate > Is there anything holding you back from using it (if you would like to use it > but currently cannot)? It would be nice if the output of the mapreduce job was a MutationOutputFormat in which we could write insert/delete, but I recall there is something on jira already albeit not sure if it was merged.
