Hi,
consider the schema
pk1 text,
ck1 text
v1 text,
v2 text.
PRIMARY KEY(pk1,ck1)

1. insert into ks.tablename(pk1,ck1,v1,v2) values('PK1,'CK1','a','a');
2. delete from ks.tablename where pk1='PK2' and ck1='CK2';
3. insert into ks.tablename(pk1,ck1) values('PK3,'CK3');
4. insert into ks.tablename(pk1,ck1,v1) values('PK4,'CK4','a');

3rd case is "insert of the form when ONLY primary key values are specified"

if you are sure, case 3 will never occur from your application, you can
check on length of "next"(as in the code snippet),
next.length() will be greater than zero in case 1,4
next.length() will be equal to zero in case 2,3

Thus, inspite of 3 being an insert, in the code snippet, it might appear to
be a delete.


Rephrasing
"If you are sure that your application will NOT do an insert of the form
when ONLY primary key values are specified, you can check the length of
next, to indicate whether it is an insert/update(where atleast one non
primary key column value is inserted) or a delete if length is zero."
If you are sure case 3 will never occur,
then checking the next.length(), you can decide whether it is an
insert/update(length > 0) OR delete(length == 0)

I would urge you to try the snippet once on you own, to see what kind of
data it produces in *next*. You could dump the output of next in a column
for audit table, to see that output.


Regards
Siddharth Verma

On Wed, Oct 5, 2016 at 1:23 AM, Kant Kodali <k...@peernova.com> wrote:

> Hi Siddharth,
>
> I don't quite follow the assumption "If you are sure that your
> application will NOT do an insert of the form when ONLY primary key values
> are specified, you can check the length of next, to indicate whether it is
> an insert/update(where atleast one non primary key column value is
> inserted) or a delete if length is zero.". Could you please provide an
> example ?
>
> Thanks,
> kant
>
>
>
> On Tue, Oct 4, 2016 12:34 PM, siddharth verma sidd.verma29.l...@gmail.com
> wrote:
>
>> Hi,
>> I am not sure whether it will help you or not.
>> Code snippet :
>> public Collection<Mutation> augment(Partition update)
>> {
>> ...
>>     StringBuilder next=new StringBuilder();
>>     SearchIterator<Clustering, Row> searchIterator =
>> update.searchIterator(ColumnFilter.all(update.metadata()),false);
>>         while(searchIterator.hasNext()){
>>             next.append(searchIterator.next(Clustering.EMPTY).
>> toString()+"\001");
>>         }
>> ...
>> //next carries non primary key column values
>> }
>>
>> If you are sure that your application will NOT do an insert of the form
>> when ONLY primary key values are specified, you can check the length of
>> next, to indicate whether it is an insert/update(where atleast one non
>> primary key column value is inserted) or a delete if length is zero.
>>
>> The code snippet is to the best of my knowledge, however, kindly try it
>> once at your end, as this was part of some legacy code, and I am not
>> completely sure about it.
>>
>> Here, if the assumption stated above holds true, you could avoid a
>> cassandra select for that key.
>>
>> Thanks
>> Siddharth Verma
>>
>>
>> On Wed, Oct 5, 2016 at 12:20 AM, Kant Kodali <k...@peernova.com> wrote:
>>
>> Thanks a lot, This helps me to make a decision on not to write one for
>> the performance reasons you pointed out!
>>
>>
>>
>> On Tue, Oct 4, 2016 11:42 AM, Eric Stevens migh...@gmail.com wrote:
>>
>> You would have to perform a SELECT on the row in the trigger code in
>> order to determine if there was underlying data.  Cassandra is in essence
>> an append-only data store, when an INSERT or UPDATE is executed, it has no
>> idea if there is already a row underlying it, and for write performance
>> reasons it also doesn't care.
>>
>> Note that if you do this, you're going to introduce a giant bottleneck in
>> your write path and increase the IO cost of writes.  You'll also probably
>> have some race conditions such that if two writes to the same row happen in
>> quick succession your trigger might not notice that one of them is writing
>> to the same row as the other. You might need to resort to CAS operations to
>> overcome that, along with its associated overhead.  But all that said, it
>> should be possible, though you'll have to write it for yourself in your
>> trigger code.
>>
>>
>>
>> On Tue, Oct 4, 2016 at 12:29 PM Kant Kodali <k...@peernova.com> wrote:
>>
>> Hi all,
>>
>> How to write a trigger in Cassandra to detect updates? My requirement is
>> that I want a trigger to alert me only when there is an update to an
>> existing row and looks like given the way INSERT and Update works this
>> might be hard to do because INSERT will just overwrite if there is an
>> existing row and Update becomes new insert where there is no row that
>> belongs to certain partition key. is there a way to solve this problem?
>>
>> Thanks,
>>
>> kant
>>
>>
>>

Reply via email to