Best practice for sorting on frequent updated column?
I need to sort data on a frequent updated column, such as like count of an item. The common way of getting data sorted in Cassandra is to have the column to be sorted on as clustering key. However, whenever such column is updated, we need to delete the row of old value and insert the new one, which not only can generate a lot of tombstones, but also require a read-before-write if we don't know the original value (such as using counter table to maintain the count and propagate it to the table that needs to sort on the count). I was wondering what is best practice for such use case? I'm currently using DSE search to handle it but I would like to see a Cassandra only solution. Thanks.
Is compound index a planned feature in 3.0?
Compound index in MongoDB is really useful for qiery that involves filtering/sorting on multiple columns. I was wondering if Cassandra 3.0 is supposed to implement this feature. When I read through JIRA, I only found feature like CASSANDRA-6048 https://issues.apache.org/jira/browse/CASSANDRA-6048, which allows using multiple single column indexes in a query by joining predicates. Compound index is more query driven and is closer to current application-maintained index table, which may provide better performance than single column index and can greatly simplify index maintenance during updates than index table. Any idea? Ziju
Re: Is compound index a planned feature in 3.0?
The global index JIRA actually mentions compound index but it seems that there is no JIRA created for this feature? Anyway, I think I should wait for 3.0 and see what does it bring to index. Thanks. On Fri, Dec 26, 2014 at 6:09 PM, DuyHai Doan doanduy...@gmail.com wrote: Many JIRA related to index are opened for 3.x Global indices: https://issues.apache.org/jira/browse/CASSANDRA-6477 Functional index: https://issues.apache.org/jira/browse/CASSANDRA-7458 Partial index: https://issues.apache.org/jira/browse/CASSANDRA-7391 On Fri, Dec 26, 2014 at 10:49 AM, ziju feng pkdog...@gmail.com wrote: Compound index in MongoDB is really useful for qiery that involves filtering/sorting on multiple columns. I was wondering if Cassandra 3.0 is supposed to implement this feature. When I read through JIRA, I only found feature like CASSANDRA-6048 https://issues.apache.org/jira/browse/CASSANDRA-6048, which allows using multiple single column indexes in a query by joining predicates. Compound index is more query driven and is closer to current application-maintained index table, which may provide better performance than single column index and can greatly simplify index maintenance during updates than index table. Any idea? Ziju
Store counter with non-counter column in the same column family?
I was wondering if there is plan to allow creating counter column and standard column in the same table. Here is my use case: I want to use counter to count how many users like a given item in my application. The like count needs to be returned along with details of item in query. To support querying items in different ways, I use both application-maintained denormalized index tables and DSE search for indexing. (DSE search is also used for text searching) Since current counter implementation doesn't allow having counter columns and non-counter columns in the same table, I have to propagate the current count from counter table to the main item table and index tables, so that like counts can be returned by those index tables without sending extra requests to counter table and DSE search is able to build index on like count column in the main item table to support like count related queries (such as sorting by like count). IMHO, the only way to sync data between counter table and normal table within a reasonable time (sub-seconds) currently is to read the current value from counter table right after the update. However it suffers from several issues: 1. Read-after-write may not return the correct count when replication factor 1 unless consistency level ALL/LOCAL_ALL is used 2. There are two extra non-parallelizable round-trips between the application server and cassandra, which can have great impact on performance. If it is possible to store counter in standard column family, only one write will be needed to update like count in the main table. Counter value will also be eventually synced between replicas so that there is no need for application to use extra mechanism like scheduled task to get the correct counts. A related issue is lifting the limitation of not allowing updating counter columns and normal columns in one batch, since it is quite common to not only have a counter for statistics but also store the details, such as storing the relation of which user likes which items in my user case. Any idea?
Re: Store counter with non-counter column in the same column family?
I just skimmed through JIRA https://issues.apache.org/jira/browse/CASSANDRA-4775 and it seems there has been some effort to make update idempotent. Perhaps the problem can be fixed in the near future? Anyway, what is the current best practice for such use case? (Counting and displaying counts in different queries) I don't need a 100% accurate count and strong consistency. Performance and application complexity is my main concern. Thanks On Mon, Dec 22, 2014 at 10:37 PM, Ryan Svihla rsvi...@datastax.com wrote: You can cheat it by using the non counter column as part of your primary key (clustering column specifically) but the cases where this could work are limited and the places this is a good idea are even more rare. As for using counters in batches are already a not well regarded concept and counter batches have a number of troubling behaviors, as already stated increments aren't idempotent and batch implies retry. As for DSE search its doing something drastically different internally and the type of counting its doing is many orders of magnitude faster ( think bitmask style matching + proper async 2i to minimize fanout cost) Generally speaking counting accurately while being highly available creates an interesting set of logical tradeoffs. Example what do you do if you're not able to communicate between two data centers, but both are up and serving likes quite happily? Is your counting down? Do you keep counting but serve up different answers? More accurately since problems are rarely data center to data center but more frequently between replicas, how much availability are you willing to give up in exchange for a globally accurate count? On Dec 22, 2014 6:00 AM, DuyHai Doan doanduy...@gmail.com wrote: It's not possible to mix counter and non counter columns because currently the semantic of counter is only increment/decrement (thus NOT idempotent) and requires some special handling compared to other C* columns. On Mon, Dec 22, 2014 at 11:33 AM, ziju feng pkdog...@gmail.com wrote: I was wondering if there is plan to allow creating counter column and standard column in the same table. Here is my use case: I want to use counter to count how many users like a given item in my application. The like count needs to be returned along with details of item in query. To support querying items in different ways, I use both application-maintained denormalized index tables and DSE search for indexing. (DSE search is also used for text searching) Since current counter implementation doesn't allow having counter columns and non-counter columns in the same table, I have to propagate the current count from counter table to the main item table and index tables, so that like counts can be returned by those index tables without sending extra requests to counter table and DSE search is able to build index on like count column in the main item table to support like count related queries (such as sorting by like count). IMHO, the only way to sync data between counter table and normal table within a reasonable time (sub-seconds) currently is to read the current value from counter table right after the update. However it suffers from several issues: 1. Read-after-write may not return the correct count when replication factor 1 unless consistency level ALL/LOCAL_ALL is used 2. There are two extra non-parallelizable round-trips between the application server and cassandra, which can have great impact on performance. If it is possible to store counter in standard column family, only one write will be needed to update like count in the main table. Counter value will also be eventually synced between replicas so that there is no need for application to use extra mechanism like scheduled task to get the correct counts. A related issue is lifting the limitation of not allowing updating counter columns and normal columns in one batch, since it is quite common to not only have a counter for statistics but also store the details, such as storing the relation of which user likes which items in my user case. Any idea?
Re: Store counter with non-counter column in the same column family?
Thanks for the advise, I'll definitely take a look at how Spark works and how it can help with counting. One last question: My current implementation of counting is 1) increment counter 2) read counter immediately after the write 3) write counts to multiple tables for different query paths and solr. If I switch to Spark, do I still needs to use counter or counting will be done by spark on regular table? On Tue, Dec 23, 2014 at 11:31 AM, Ryan Svihla rsvi...@datastax.com wrote: increment wouldn't be idempotent from the client unless you knew the count at the time of the update (which you could do with LWT but that has pretty harsh performance), that particular jira is about how they're laid out and avoiding race conditions between nodes, which was resolved in 2.1 beta 1 (which is now in officially out in the 2.1.x branch) General improvements on counters in 2.1 are laid out here http://www.datastax.com/dev/blog/whats-new-in-cassandra-2-1-a-better-implementation-of-counters As for best practice the answer is multiple tables for multiple query paths, or you can use something like solr or spark, take a look at the spark cassandra connector for a good way to count on lots of data from lots of different query paths https://github.com/datastax/spark-cassandra-connector. On Mon, Dec 22, 2014 at 9:22 PM, ziju feng pkdog...@gmail.com wrote: I just skimmed through JIRA https://issues.apache.org/jira/browse/CASSANDRA-4775 and it seems there has been some effort to make update idempotent. Perhaps the problem can be fixed in the near future? Anyway, what is the current best practice for such use case? (Counting and displaying counts in different queries) I don't need a 100% accurate count and strong consistency. Performance and application complexity is my main concern. Thanks On Mon, Dec 22, 2014 at 10:37 PM, Ryan Svihla rsvi...@datastax.com wrote: You can cheat it by using the non counter column as part of your primary key (clustering column specifically) but the cases where this could work are limited and the places this is a good idea are even more rare. As for using counters in batches are already a not well regarded concept and counter batches have a number of troubling behaviors, as already stated increments aren't idempotent and batch implies retry. As for DSE search its doing something drastically different internally and the type of counting its doing is many orders of magnitude faster ( think bitmask style matching + proper async 2i to minimize fanout cost) Generally speaking counting accurately while being highly available creates an interesting set of logical tradeoffs. Example what do you do if you're not able to communicate between two data centers, but both are up and serving likes quite happily? Is your counting down? Do you keep counting but serve up different answers? More accurately since problems are rarely data center to data center but more frequently between replicas, how much availability are you willing to give up in exchange for a globally accurate count? On Dec 22, 2014 6:00 AM, DuyHai Doan doanduy...@gmail.com wrote: It's not possible to mix counter and non counter columns because currently the semantic of counter is only increment/decrement (thus NOT idempotent) and requires some special handling compared to other C* columns. On Mon, Dec 22, 2014 at 11:33 AM, ziju feng pkdog...@gmail.com wrote: I was wondering if there is plan to allow creating counter column and standard column in the same table. Here is my use case: I want to use counter to count how many users like a given item in my application. The like count needs to be returned along with details of item in query. To support querying items in different ways, I use both application-maintained denormalized index tables and DSE search for indexing. (DSE search is also used for text searching) Since current counter implementation doesn't allow having counter columns and non-counter columns in the same table, I have to propagate the current count from counter table to the main item table and index tables, so that like counts can be returned by those index tables without sending extra requests to counter table and DSE search is able to build index on like count column in the main item table to support like count related queries (such as sorting by like count). IMHO, the only way to sync data between counter table and normal table within a reasonable time (sub-seconds) currently is to read the current value from counter table right after the update. However it suffers from several issues: 1. Read-after-write may not return the correct count when replication factor 1 unless consistency level ALL/LOCAL_ALL is used 2. There are two extra non-parallelizable round-trips between the application server and cassandra, which can have great impact on performance. If it is possible to store counter in standard column family, only one write
Plan to implement server side synchronization of denormalized data ?
Hi all, I was wondering if there is any plan to support syncing change automatically between entity table and tables that contain denormalized data on server side? I think many use cases in Cassandra require some level of denormalization. However, there is currently little support for denormalization from server side. Denormalization has to be done by driver or even application, which leads to two issues: 1. Application complexity: As far as I know, there is no drivers support propagating changes of main entity to denormalized ones, user will have to handle data synchronization themselves. There can be a lot of codes to write and it's quite hard to get it done right, considering things like what consistency level to use, sync vs async update, reverse index table, etc. 2. Data consistency: Suppose there is an entity table: Create table entity( id text primary key, name text, value text) and an index table for 'name', which also stores 'value' for denormalization: Create table name_idx( name text, id text, value text) When a request to update 'value' is sent to the application, it needs to update both entity and name_idx tables. Suppose another request to update 'name' is sent at the same time, the application will need to delete the original row from name_idx and create a new row based on the new name. However, if the 1st request read (it has to retrieve the value of 'name' in order to update name_idx) before the 2nd request finishes, its update statement will generate a row in name_idx with the original name, which leads to inconsistent data. CAS may help here, but when the number of concurrent requests is large and there are more index tables, CAS could fail frequently. Since secondary index has limitation in both performance and query flexibility (no order by, for example), it can help the application a lot if Cassandra support server maintained (just like the secondary index) index tables on a main table. One possible syntax can be 'CREATE VIEW view_name ON table_name' and assume each column in the view would have the same name as in the main table as convention, so that user can create different views based on their query requirements. Thanks, Ziju
Document of WRITETIME function needs update
Hi, I found that the WRITETIME function on counter column returns date/time in milliseconds instead of microseconds, which is not mentioned in the document http://www.datastax.com/documentation/cql/3.1/cql/cql_using/use_writetime.html. It will be helpful to clarify the difference in the document. One side question: I denormalize the counter column value to regular tables using read-after-write in QUORUM consistency from counter table and update the regular tables using counter column's write time to resolve write conflict. Is this a valid use case? Thanks, Ziju.
Re: Does the default LIMIT applies to automatic paging?
Thank you all for your answers and clarification. The reason I mentioned the 1 rows LIMIT is not only because it is the default LIMIT in cqlsh, but also because I found it on the CQL document http://www.datastax.com/documentation/cql/3.1/cql/cql_reference/select_r.html, specifically the Specifying rows returned using LIMIT section. Perhaps the document needs some updates to clarify a bit about what applies to the drivers and what applies to cqlsh? On Wed, Jun 25, 2014 at 12:21 AM, Sylvain Lebresne sylv...@datastax.com wrote: On Tue, Jun 24, 2014 at 1:03 AM, ziju feng pkdog...@gmail.com wrote: I was wondering if the default 1 rows LIMIT applies to automatic pagination in C* 2.0 (I'm using Datastax driver). There is no 1 rows LIMIT in CQL. cqlsh does apply a default LIMIT if you don't provide for convenience sake, but it's a cqlsh thing. Therefore, there is no default limit with the java driver (neither with or without automatic pagination). -- Sylvain
Re: Does the default LIMIT applies to automatic paging?
Does that mean the iterator will give me all the data instead of 1 rows? On Mon, Jun 23, 2014 at 10:20 PM, DuyHai Doan doanduy...@gmail.com wrote: With the Java Driver, set the fetchSize and use ResultSet.iterator Le 24 juin 2014 01:04, ziju feng pkdog...@gmail.com a écrit : Hi All, I have a wide row table that I want to iterate through all rows under a specific partition key. The table may contains around one million rows per partition I was wondering if the default 1 rows LIMIT applies to automatic pagination in C* 2.0 (I'm using Datastax driver). If so, what is best way to retrieve all rows of a given partition? Should I use a super large LIMIT value or should I manually page through the table? Thanks, Ziju
Does the default LIMIT applies to automatic paging?
Hi All, I have a wide row table that I want to iterate through all rows under a specific partition key. The table may contains around one million rows per partition I was wondering if the default 1 rows LIMIT applies to automatic pagination in C* 2.0 (I'm using Datastax driver). If so, what is best way to retrieve all rows of a given partition? Should I use a super large LIMIT value or should I manually page through the table? Thanks, Ziju
Retrieve counter value after update
Hi All, I was wondering if there is a planned feature in Cassandra to return the current counter value after the update statement? Our project is using counter column to count and since counter column cannot reside in the same table with regular columns, we have to denormalize the counter value as integer into other tables that need to display the value. Our current way of denormalization is to read the current value and writetime from the counter table after the update and then batch update other tables with the value and timestamp (to resolve wrtie conflict). I don't know if this is a common requirement but I think if update to counter table can return the current value and timestamp (or counter column can reside in regular table in the first place), we can save this extra read, which can reduce cluster load and update latency. Thanks, Ziju
Re: Data modeling for Pinterest-like application
I was thinking to use counter type a separate pin counter table and, when I need to update the like count, I would use read-after-write to get the current value and timestamp and then denormalize into pin's detail table and board tables. Is it a viable solution in this case? Thanks -- View this message in context: http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Data-modeling-for-Pinterest-like-application-tp7594481p7594539.html Sent from the cassandra-u...@incubator.apache.org mailing list archive at Nabble.com.
Data modeling for Pinterest-like application
Hello, I'm working on data modeling for a Pinterest-like project. There are basically two main concepts: Pin and Board, just like Pinterest, where pin is an item containing an image, description and some other information such as a like count, and each board should contain a sorted list of Pins. The board can be modeled with primary key (board_id, created_at, pin_id) where created_at is used to sort the pins of the board by date. The problem is whether I should denormalize details of pins into the board table or just retrieve pins by page (page size can be 10~20) and then multi-get by pin_ids to obtain details. Since there are some boards that are accessed very often (like the home board), denormalization seems to be a reasonable choice to enhance read performance. However, we then have to update not only the pin table be also each row in the board table that contains the pin whenever a pin is updated, which sometimes could be quite frequent (such as updating the like count). Since a pin may be contained by many boards (could be thousands), denormalization seems to bring a lot of load on the write side as well as application code complexity. Any suggestion to whether our data model should go denormalized or the normalized/multi-get way which then perhaps need a separate cached layer for read? Thanks, Ziju
Re: Data modeling for Pinterest-like application
Thanks for your answer, I really like the frequency of update vs read way of thinking. A related question is whether it is a good idea to denormalize on read-heavy part of data while normalize on other less frequently-accessed data? Our app will have a limited number of system managed boards that are viewed by every user so it makes sense to denormalize and propagate updates of pins to these boards. We will also have a like board for each user containing pins that they like, which can be somewhat private and only viewed by the owner. Since a pin can be potentially liked by thousands of user, if we also denormalize the like board, everytime that pin is liked by another user we would have to update the like count in thousands of like boards. Does normalize work better in this case or cassandra can handle this kind of write load? -- View this message in context: http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/Data-modeling-for-Pinterest-like-application-tp7594481p7594517.html Sent from the cassandra-u...@incubator.apache.org mailing list archive at Nabble.com.
How to guarantee consistency between counter and materialized view?
Hi all, Is there any way to guarantee a counter's value in materialized views, which could be some other column families with different row keys and with counter's value de-normalized, in sync with the value in its counter column family? Since batch can only work as either non-counter or counter-only batch, I can't have updating the batch table and the materialized view tables in one batch. In the case that the client fails right after updating the counter table and right before updating the materialized views, there could be inconsistent data. Thanks, Ziju