You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Jason Tang <ar...@gmail.com> on 2012/04/25 15:18:47 UTC
Cassandra search performance
Hi
We have the such CF, and use secondary index to search for simple data
"status", and among 1,000,000 row records, we have 200 records with status
we want.
But when we start to search, the performance is very poor, and check with
the command "./bin/nodetool -h localhost -p 8199 cfstats" , Cassandra read
1,000,000 records, and "Read Latency" is 0.2 ms, so totally it used 200
seconds.
It use lots of CPU, and check the stack, all thread in Cassandra is read
from socket.
So I wonder, how to really use index to find the 200 records instead of
scan all rows. (Supper Column?)
*ColumnFamily: queue*
* Key Validation Class: org.apache.cassandra.db.marshal.BytesType*
* Default column value validator:
org.apache.cassandra.db.marshal.BytesType*
* Columns sorted by: org.apache.cassandra.db.marshal.BytesType*
* Row cache size / save period in seconds / keys to save : 0.0/0/all*
* Row Cache Provider:
org.apache.cassandra.cache.ConcurrentLinkedHashCacheProvider*
* Key cache size / save period in seconds: 0.0/0*
* GC grace seconds: 0*
* Compaction min/max thresholds: 4/32*
* Read repair chance: 0.0*
* Replicate on write: false*
* Bloom Filter FP chance: default*
* Built indexes: [queue.idxStatus]*
* Column Metadata:*
* Column Name: status (737461747573)*
* Validation Class: org.apache.cassandra.db.marshal.AsciiType*
* Index Name: idxStatus*
* Index Type: KEYS*
*
*
BRs
//Jason
Re: Cassandra search performance
Posted by Dave Brosius <db...@mebigfatguy.com>.
When you build a secondary index on a field, what happens is a second
(hidden) column family is created for you, with the key for that column
family being the value of the index field, and the columns being the
keys of the entries in your column family that have that value.
So in your case, you made a bogus column that had all the same values,
and index that. So what that did was create a hidden column family with
one row. the key had the bogus value, and you had n columns that held
the keys for all your records.
When you did the query, since you did bogus_column EQ bogusvalue, you
basically got all the keys from that one index row.
You then loaded all rows in your column family, and in memory did the
real query you wanted with the inequality based on time (00:00 to 23:59)
As you can see this index was of no value, as it didn't pare down your
set of rows at all.
What you probably want to do is change your data model. Your original
column family should probably have a key that represents some time based
element 'bucket'. For instance, you might want to have rows that
represents 1 hour, and all the columns in that row are values for
various points of time in that one hour.
Then when you want to do searches for data for time ranges, you know
what keys you should query to get the values for those times. And then
what columns in those rows to get.
Of course you haven't really laid out what you are trying to do, so this
explanation is kind of fuzzy, so...
On 05/12/2012 06:18 AM, Jason Tang wrote:
> I try to search one column, this column store the time as the type
> Long, 1,000,000 data equally distributed in 24 hours, I only want to
> search certain time rang, eg from 01:30 to 01:50 or 08:00 to 12:00,
> but something stranger happened.
>
> Search 00:00 to 23:59 limit 100
> It took less then 1 second scan 100 record
>
> Search 00:00 to 00:20 limit 100
> It took more then one minute scan around 2,400 recods
>
> So the result shows it seems cassandra scan one by one to match the
> condition, and the data is not ordered in sequence.
>
> One more thing, to have equal condition, I make a redundant column to
> have equal condition, the value is same for all records.
> The search condition like get record where equal='equal' and time >
> 00:00 and time < 00:20
>
> Is it the expected behavior of secondary index or I didn't use it correct.
>
> Because I used to have another test, I have one string column most of
> it is string 'true' and I add 100 'false' among 1,000,000 'true' , it
> shows it only scan 100 records.
>
> So how can I exam what happened inside cassadra, and where I can find
> out the detail of how secondary works?
>
> 在 2012年5月8日星期二，Maxim Potekhin 写道：
>
> Thanks for the comments, much appreciated.
>
> Maxim
>
>
> On 5/7/2012 3:22 AM, David Jeske wrote:
>> On Sun, Apr 29, 2012 at 4:32 PM, Maxim Potekhin <potekhin@bnl.gov
>> <javascript:_e({}, 'cvml', 'potekhin@bnl.gov');>> wrote:
>>
>> Looking at your example,as I think you understand, you forgo
>> indexes by
>> combining two conditions in one query, thinking along the
>> lines of what is
>> often done in RDBMS. A scan is expected in this case, and
>> there is no
>> magic to avoid it.
>>
>>
>> This sounds like a mis-understanding of how RDBMSs work. If you
>> combine two conditions in a single SQL query, the SQL execution
>> optimizer looks at the cardinality of any indicies. If it can
>> successfully predict that one of the conditions significantly
>> reduces the set of rows that would be considered (such as a
>> status match having 200 hits vs 1M rows in the table), then it
>> selects this index for the first-iteration, and each index hit
>> causes a record lookup which is then tested for the other
>> conditions. (This is one of several query-execution types RDBMS
>> systems use)
>>
>> I'm no Cassandra expert, so I don't know what it does WRT
>> index-selection, but from the page written on secondary indicies,
>> it seems like if you just query on status, and do the other
>> filtering yourself it'll probably do what you want...
>>
>> http://www.datastax.com/dev/blog/whats-new-cassandra-07-secondary-indexes
>>
>>
>> However, if this query is important, you can easily index on
>> two conditions,
>> using a composite type (look it up), or string concatenation
>> for quick and
>> easy solution.
>>
>>
>> This is not necessarily a good idea. Creating a composite index
>> explodes the index size unnecessarily. If a condition can reduce
>> a query to 200 records, there is no need to have a composite
>> index including another condition.
>
Re: Cassandra search performance
Posted by Jason Tang <ar...@gmail.com>.
I try to search one column, this column store the time as the type Long,
1,000,000 data equally distributed in 24 hours, I only want to search
certain time rang, eg from 01:30 to 01:50 or 08:00 to 12:00, but something
stranger happened.
Search 00:00 to 23:59 limit 100
It took less then 1 second scan 100 record
Search 00:00 to 00:20 limit 100
It took more then one minute scan around 2,400 recods
So the result shows it seems cassandra scan one by one to match the
condition, and the data is not ordered in sequence.
One more thing, to have equal condition, I make a redundant column to have
equal condition, the value is same for all records.
The search condition like get record where equal='equal' and time > 00:00
and time < 00:20
Is it the expected behavior of secondary index or I didn't use it correct.
Because I used to have another test, I have one string column most of it is
string 'true' and I add 100 'false' among 1,000,000 'true' , it shows it
only scan 100 records.
So how can I exam what happened inside cassadra, and where I can find out
the detail of how secondary works?
在 2012年5月8日星期二，Maxim Potekhin 写道：
> Thanks for the comments, much appreciated.
>
> Maxim
>
>
> On 5/7/2012 3:22 AM, David Jeske wrote:
>
> On Sun, Apr 29, 2012 at 4:32 PM, Maxim Potekhin <potekhin@bnl.gov<javascript:_e({}, 'cvml', 'potekhin@bnl.gov');>
> > wrote:
>
>> Looking at your example,as I think you understand, you forgo indexes by
>> combining two conditions in one query, thinking along the lines of what is
>> often done in RDBMS. A scan is expected in this case, and there is no
>> magic to avoid it.
>>
>
> This sounds like a mis-understanding of how RDBMSs work. If you combine
> two conditions in a single SQL query, the SQL execution optimizer looks at
> the cardinality of any indicies. If it can successfully predict that one of
> the conditions significantly reduces the set of rows that would be
> considered (such as a status match having 200 hits vs 1M rows in the
> table), then it selects this index for the first-iteration, and each index
> hit causes a record lookup which is then tested for the other conditions.
> (This is one of several query-execution types RDBMS systems use)
>
> I'm no Cassandra expert, so I don't know what it does WRT
> index-selection, but from the page written on secondary indicies, it seems
> like if you just query on status, and do the other filtering yourself it'll
> probably do what you want...
>
> http://www.datastax.com/dev/blog/whats-new-cassandra-07-secondary-indexes
>
>
>> However, if this query is important, you can easily index on two
>> conditions,
>> using a composite type (look it up), or string concatenation for quick and
>> easy solution.
>>
>
> This is not necessarily a good idea. Creating a composite index explodes
> the index size unnecessarily. If a condition can reduce a query to 200
> records, there is no need to have a composite index including another
> condition.
>
>
>
Re: Cassandra search performance
Posted by Maxim Potekhin <po...@bnl.gov>.
Thanks for the comments, much appreciated.
Maxim
On 5/7/2012 3:22 AM, David Jeske wrote:
> On Sun, Apr 29, 2012 at 4:32 PM, Maxim Potekhin <potekhin@bnl.gov
> <ma...@bnl.gov>> wrote:
>
> Looking at your example,as I think you understand, you forgo
> indexes by
> combining two conditions in one query, thinking along the lines of
> what is
> often done in RDBMS. A scan is expected in this case, and there is no
> magic to avoid it.
>
>
> This sounds like a mis-understanding of how RDBMSs work. If you
> combine two conditions in a single SQL query, the SQL execution
> optimizer looks at the cardinality of any indicies. If it can
> successfully predict that one of the conditions significantly reduces
> the set of rows that would be considered (such as a status match
> having 200 hits vs 1M rows in the table), then it selects this index
> for the first-iteration, and each index hit causes a record lookup
> which is then tested for the other conditions. (This is one of
> several query-execution types RDBMS systems use)
>
> I'm no Cassandra expert, so I don't know what it does WRT
> index-selection, but from the page written on secondary indicies, it
> seems like if you just query on status, and do the other filtering
> yourself it'll probably do what you want...
>
> http://www.datastax.com/dev/blog/whats-new-cassandra-07-secondary-indexes
>
> However, if this query is important, you can easily index on two
> conditions,
> using a composite type (look it up), or string concatenation for
> quick and
> easy solution.
>
>
> This is not necessarily a good idea. Creating a composite index
> explodes the index size unnecessarily. If a condition can reduce a
> query to 200 records, there is no need to have a composite index
> including another condition.
unsibscribe
Posted by Maxim Potekhin <po...@bnl.gov>.
Re: Cassandra search performance
Posted by David Jeske <da...@gmail.com>.
On Sun, Apr 29, 2012 at 4:32 PM, Maxim Potekhin <po...@bnl.gov> wrote:
> Looking at your example,as I think you understand, you forgo indexes by
> combining two conditions in one query, thinking along the lines of what is
> often done in RDBMS. A scan is expected in this case, and there is no
> magic to avoid it.
>
This sounds like a mis-understanding of how RDBMSs work. If you combine two
conditions in a single SQL query, the SQL execution optimizer looks at the
cardinality of any indicies. If it can successfully predict that one of the
conditions significantly reduces the set of rows that would be considered
(such as a status match having 200 hits vs 1M rows in the table), then it
selects this index for the first-iteration, and each index hit causes a
record lookup which is then tested for the other conditions. (This is one
of several query-execution types RDBMS systems use)
I'm no Cassandra expert, so I don't know what it does WRT index-selection,
but from the page written on secondary indicies, it seems like if you just
query on status, and do the other filtering yourself it'll probably do what
you want...
http://www.datastax.com/dev/blog/whats-new-cassandra-07-secondary-indexes
> However, if this query is important, you can easily index on two
> conditions,
> using a composite type (look it up), or string concatenation for quick and
> easy solution.
>
This is not necessarily a good idea. Creating a composite index explodes
the index size unnecessarily. If a condition can reduce a query to 200
records, there is no need to have a composite index including another
condition.
Re: Cassandra search performance
Posted by Maxim Potekhin <po...@bnl.gov>.
Jason,
I'm using plenty of secondary indexes with no problem at all.
Looking at your example,as I think you understand, you forgo indexes by
combining two conditions in one query, thinking along the lines of what is
often done in RDBMS. A scan is expected in this case, and there is no
magic to avoid it.
However, if this query is important, you can easily index on two conditions,
using a composite type (look it up), or string concatenation for quick and
easy solution. Which is, you _create an additional column_ which contains a
combination of the two you want to use in a query. Then index on it.
Problem solved.
The composite solution is more elegant but what I describe works in
simple cases.
It works for me.
Maxim
On 4/25/2012 10:45 AM, Jason Tang wrote:
> 1.0.8
>
> 在 2012年4月25日 下午10:38，Philip Shon <philip.shon@gmail.com
> <ma...@gmail.com>>写 道：
>
> what version of cassandra are you using. I found a big performance
> hit when querying on the secondary index.
>
> I came across this bug in versions prior to 1.1
>
> https://issues.apache.org/jira/browse/CASSANDRA-3545
>
> Hope that helps.
>
> 2012/4/25 Jason Tang <ares.tang@gmail.com
> <ma...@gmail.com>>
>
> And I found, if I only have the search condition "status", it
> only scan 200 records.
>
> But if I combine another condition "partition" then it scan
> all records because "partition" condition match all records.
>
> But combine with other condition such as "userName", even all
> "userName" is same in the 1,000,000 records, it only scan 200
> records.
>
> So it impacted by scan execution plan, if we have several
> search conditions, how it works? Do we have the similar
> execution plan in Cassandra?
>
>
> 在 2012年4月25日 下午9:18，Jason Tang <ares.tang@gmail.com
> <ma...@gmail.com>>写 道：
>
> Hi
>
> We have the such CF, and use secondary index to search for
> simple data "status", and among 1,000,000 row records, we
> have 200 records with status we want.
>
> But when we start to search, the performance is very poor,
> and check with the command "./bin/nodetool -h localhost -p
> 8199 cfstats" , Cassandra read 1,000,000 records, and
> "Read Latency" is 0.2 ms, so totally it used 200 seconds.
>
> It use lots of CPU, and check the stack, all thread in
> Cassandra is read from socket.
>
> So I wonder, how to really use index to find the 200
> records instead of scan all rows. (Supper Column?)
>
> /ColumnFamily: queue/
> /Key Validation Class:
> org.apache.cassandra.db.marshal.BytesType/
> /Default column value validator:
> org.apache.cassandra.db.marshal.BytesType/
> /Columns sorted by: org.apache.cassandra.db.marshal.BytesType/
> /Row cache size / save period in seconds / keys to save :
> 0.0/0/all/
> /Row Cache Provider:
> org.apache.cassandra.cache.ConcurrentLinkedHashCacheProvider/
> /Key cache size / save period in seconds: 0.0/0/
> /GC grace seconds: 0/
> /Compaction min/max thresholds: 4/32/
> /Read repair chance: 0.0/
> /Replicate on write: false/
> /Bloom Filter FP chance: default/
> /Built indexes: [queue.idxStatus]/
> /Column Metadata:/
> /Column Name: status (737461747573)/
> /Validation Class: org.apache.cassandra.db.marshal.AsciiType/
> /Index Name: idxStatus/
> /Index Type: KEYS/
> /
> /
> BRs
> //Jason
>
>
>
>
Re: Cassandra search performance
Posted by Jason Tang <ar...@gmail.com>.
1.0.8
在 2012年4月25日 下午10:38，Philip Shon <ph...@gmail.com>写道：
> what version of cassandra are you using. I found a big performance hit
> when querying on the secondary index.
>
> I came across this bug in versions prior to 1.1
>
> https://issues.apache.org/jira/browse/CASSANDRA-3545
>
> Hope that helps.
>
> 2012/4/25 Jason Tang <ar...@gmail.com>
>
>> And I found, if I only have the search condition "status", it only scan
>> 200 records.
>>
>> But if I combine another condition "partition" then it scan all records
>> because "partition" condition match all records.
>>
>> But combine with other condition such as "userName", even all "userName"
>> is same in the 1,000,000 records, it only scan 200 records.
>>
>> So it impacted by scan execution plan, if we have several search
>> conditions, how it works? Do we have the similar execution plan in
>> Cassandra?
>>
>>
>> 在 2012年4月25日 下午9:18，Jason Tang <ar...@gmail.com>写道：
>>
>> Hi
>>>
>>> We have the such CF, and use secondary index to search for simple
>>> data "status", and among 1,000,000 row records, we have 200 records with
>>> status we want.
>>>
>>> But when we start to search, the performance is very poor, and check
>>> with the command "./bin/nodetool -h localhost -p 8199 cfstats" , Cassandra
>>> read 1,000,000 records, and "Read Latency" is 0.2 ms, so totally it used
>>> 200 seconds.
>>>
>>> It use lots of CPU, and check the stack, all thread in Cassandra is
>>> read from socket.
>>>
>>> So I wonder, how to really use index to find the 200 records instead
>>> of scan all rows. (Supper Column?)
>>>
>>> *ColumnFamily: queue*
>>> * Key Validation Class: org.apache.cassandra.db.marshal.BytesType*
>>> * Default column value validator:
>>> org.apache.cassandra.db.marshal.BytesType*
>>> * Columns sorted by: org.apache.cassandra.db.marshal.BytesType*
>>> * Row cache size / save period in seconds / keys to save :
>>> 0.0/0/all*
>>> * Row Cache Provider:
>>> org.apache.cassandra.cache.ConcurrentLinkedHashCacheProvider*
>>> * Key cache size / save period in seconds: 0.0/0*
>>> * GC grace seconds: 0*
>>> * Compaction min/max thresholds: 4/32*
>>> * Read repair chance: 0.0*
>>> * Replicate on write: false*
>>> * Bloom Filter FP chance: default*
>>> * Built indexes: [queue.idxStatus]*
>>> * Column Metadata:*
>>> * Column Name: status (737461747573)*
>>> * Validation Class: org.apache.cassandra.db.marshal.AsciiType*
>>> * Index Name: idxStatus*
>>> * Index Type: KEYS*
>>> *
>>> *
>>> BRs
>>> //Jason
>>>
>>
>>
>
Re: Cassandra search performance
Posted by Philip Shon <ph...@gmail.com>.
what version of cassandra are you using. I found a big performance hit
when querying on the secondary index.
I came across this bug in versions prior to 1.1
https://issues.apache.org/jira/browse/CASSANDRA-3545
Hope that helps.
2012/4/25 Jason Tang <ar...@gmail.com>
> And I found, if I only have the search condition "status", it only scan
> 200 records.
>
> But if I combine another condition "partition" then it scan all records
> because "partition" condition match all records.
>
> But combine with other condition such as "userName", even all "userName"
> is same in the 1,000,000 records, it only scan 200 records.
>
> So it impacted by scan execution plan, if we have several search
> conditions, how it works? Do we have the similar execution plan in
> Cassandra?
>
>
> 在 2012年4月25日 下午9:18，Jason Tang <ar...@gmail.com>写道：
>
> Hi
>>
>> We have the such CF, and use secondary index to search for simple data
>> "status", and among 1,000,000 row records, we have 200 records with status
>> we want.
>>
>> But when we start to search, the performance is very poor, and check
>> with the command "./bin/nodetool -h localhost -p 8199 cfstats" , Cassandra
>> read 1,000,000 records, and "Read Latency" is 0.2 ms, so totally it used
>> 200 seconds.
>>
>> It use lots of CPU, and check the stack, all thread in Cassandra is
>> read from socket.
>>
>> So I wonder, how to really use index to find the 200 records instead of
>> scan all rows. (Supper Column?)
>>
>> *ColumnFamily: queue*
>> * Key Validation Class: org.apache.cassandra.db.marshal.BytesType*
>> * Default column value validator:
>> org.apache.cassandra.db.marshal.BytesType*
>> * Columns sorted by: org.apache.cassandra.db.marshal.BytesType*
>> * Row cache size / save period in seconds / keys to save : 0.0/0/all
>> *
>> * Row Cache Provider:
>> org.apache.cassandra.cache.ConcurrentLinkedHashCacheProvider*
>> * Key cache size / save period in seconds: 0.0/0*
>> * GC grace seconds: 0*
>> * Compaction min/max thresholds: 4/32*
>> * Read repair chance: 0.0*
>> * Replicate on write: false*
>> * Bloom Filter FP chance: default*
>> * Built indexes: [queue.idxStatus]*
>> * Column Metadata:*
>> * Column Name: status (737461747573)*
>> * Validation Class: org.apache.cassandra.db.marshal.AsciiType*
>> * Index Name: idxStatus*
>> * Index Type: KEYS*
>> *
>> *
>> BRs
>> //Jason
>>
>
>
Re: Cassandra search performance
Posted by Jason Tang <ar...@gmail.com>.
And I found, if I only have the search condition "status", it only scan 200
records.
But if I combine another condition "partition" then it scan all records
because "partition" condition match all records.
But combine with other condition such as "userName", even all "userName" is
same in the 1,000,000 records, it only scan 200 records.
So it impacted by scan execution plan, if we have several search
conditions, how it works? Do we have the similar execution plan in
Cassandra?
在 2012年4月25日 下午9:18，Jason Tang <ar...@gmail.com>写道：
> Hi
>
> We have the such CF, and use secondary index to search for simple data
> "status", and among 1,000,000 row records, we have 200 records with status
> we want.
>
> But when we start to search, the performance is very poor, and check
> with the command "./bin/nodetool -h localhost -p 8199 cfstats" , Cassandra
> read 1,000,000 records, and "Read Latency" is 0.2 ms, so totally it used
> 200 seconds.
>
> It use lots of CPU, and check the stack, all thread in Cassandra is read
> from socket.
>
> So I wonder, how to really use index to find the 200 records instead of
> scan all rows. (Supper Column?)
>
> *ColumnFamily: queue*
> * Key Validation Class: org.apache.cassandra.db.marshal.BytesType*
> * Default column value validator:
> org.apache.cassandra.db.marshal.BytesType*
> * Columns sorted by: org.apache.cassandra.db.marshal.BytesType*
> * Row cache size / save period in seconds / keys to save : 0.0/0/all*
> * Row Cache Provider:
> org.apache.cassandra.cache.ConcurrentLinkedHashCacheProvider*
> * Key cache size / save period in seconds: 0.0/0*
> * GC grace seconds: 0*
> * Compaction min/max thresholds: 4/32*
> * Read repair chance: 0.0*
> * Replicate on write: false*
> * Bloom Filter FP chance: default*
> * Built indexes: [queue.idxStatus]*
> * Column Metadata:*
> * Column Name: status (737461747573)*
> * Validation Class: org.apache.cassandra.db.marshal.AsciiType*
> * Index Name: idxStatus*
> * Index Type: KEYS*
> *
> *
> BRs
> //Jason
>