Hi All
Thanks for pay attention to my mail, I had meet a problem, can you please
give me some advise or suggestion.
my problem as bellow :
1. There was many content( docs, picture, topic ...) in my web applications.
2. Each content has a accessCount value, work for record how many times the
content being viewed. we store the accessCount value in Cassandra.
3. Every times user view the content, the accessCount will increase 1. Every
increase, I need to do 3 steps, first read current value, then increase it,
at last store it back to Cassandra.
4. User access was happed on many different webservers , so If 2 user access
same content at same times, then accessCount maybe will incorrect, becuase I
maybe lost one access count.
As I know, Cassandra have not inc/dec method current, I have to design some
way to make accessCount correct as bellow, it not very well , so please let
me know if you has more better way, thanks.
Way 1. Got help from memcached, becuase memcached has inc/dec/CAP method, so
I can using memcached to maintains the accessCount value, and have a backend
cronjob to sync the value to cassandra with fixed time period ( eg: sync it
every 5s ).
This way can make things easy, but have risk if the memcached server was
down, will lost 5s counts, another risk is, the backend cronjob was a single
failuer point .
Way 2. Create a ColumnFamily as access queue, every times, user access
action will being pushed into this queue with timeUUID, a backend job will
batch read all access action, count it with inmemory loop. After got the
result, then store it back to Cassandra.
This way was all based on Cassandra, the backend job still was a single
failuer point.
Way 3. Like RDBMS's optimistic lock, if remove operation can return the
number of column being removed or update , then bellow code should can work:
while(true){
int count = readAccessCount( contentID ) ; // get the
value, if the value not exist, will loop until to get it
if( deleteOldAccessCount( contentID, count ) != 1 ){ // remove
old count, if it already removed, then conflicte happend, will get the value
again.
writeAccessCount( ++count ); // after remove old value,
insert new value
break ;
}
}
but dur to my test, even the remove target not exist, remove operation still
sucess have not any return. Is there some work arround to make optimistic
lock?
I think somebody shold meet same problem as me, can you tell us your
solutions?
Thanks
-Santal
