I am working on a self-bi application where users can upload their own datasets which are stored in Cassandra tables that are created dynamically. The data is extracted from files that the user can upload. So, each dataset is written into its own Cassandra table modeled based on column headers in the uploaded file while indexing the dimensions.
Once the data is uploaded, the users are allowed to build reports, analyze, etc., from within the application. I need a way to allow users to merge/join data from two or more datasets/tables based on matching keys and write the result into a new Cassandra table. Once a dataset/table is created, it will stay immutable and data is only read from it.
user table 1
username
email
employee id
user table 2
employee id
manager
I need to merge data in user table 1 and user table 2 on matching employee id and write to new table that is created dynamically.
new table
username
email
employee id
manager
What would be the best way to do this?
The only option that you have is to do the join in your application code. There are just few details to suggest a proper solution.
Please add details about table keys, usage patterns... in general, in cassandra you model from usage point of view, i.e. starting with queries that you'll execute on data.
In order to merge 2 tables on this pattern, you have to do it into application, creating the third table (target table ) and fill it with data from both tables. You have to make sure that you read the data in pages to not OOM, it really depends on size of the data.
Another alternative is to build the joins into Spark, but maybe is too over-engineering in your case.
You can have merge table with primary key of user so that merged data goes in one row and that should be unique since it is one time action.
Than when user clicks you can go through one table in batches with fetch size (for java you can check query options but that is a way to have a fixed window which will be loaded and when reached move to next fetch size of elements). Lets say you have fetch size of 1000 items, iterate over them from one table and find matches in second table, and after 1000 is reached place batch of 1000 inserts to new table.
If that is time consuming you can as suggested use some other tool like Apache Spark or Spring Batch and do that in background informing user that it will take place.
Related
I am currently working on attempting to transition a large database of testing records from Excel to Access to allow for better relational analysis between different groups; however, with how our team completes tests the same tests are repeated on a cycle making it difficult to capture a unique ID -
I have considered assigning one for each test but then realized the ID would still repeat when the test is ran again. I have considered using the timeframe of the review but that would also be repeated unless I build a separate table for each individual test. I have considered using the issue number that would be assigned to items requiring action, but this would not apply to all rows therefore access would not allow it.
Within our current database we capture test name, type, timeframe reviewed, start and end dates, result types, descriptions and root cause, and issue identifiers if remediation is required.
Does anyone have any suggestions on how I might transition this data into Access without losing the Primary Key feature on certain tables?
I would suggest creating a master table to identify your 'test types' (the things that can be run multiple times). You populate this table with the unique entries in the test name column in your Excel data. This column in your Excel data then becomes a 'foreign key' pointing to your master 'test types' table.
Ideally I would recommend using an autonumber as the primary key in the master table and then replacing the test name column in your Excel data with the numeric values generated when you populate the 'test types' table.
You would then import your Excel data into a table that is recording 'Test Run Instances'. Again I would set an autonumber field as the primary key on this table. Other columns in your Excel data that contain repeating data would make candidates for other 'master' tables (e.g. you talk about 'result types' - is there a set list of these that could be moved out to a master table?).
I have a table Called 'usertab' to store user details such as:
userid uuid,
firstname text,
lastname text
email text
gender int
image text
Most of the other tables contains userid as a field for referencing 'usertab',
but when I retrieve data from other table, I need to execute another select query to get user details.
So if 10,000 or more data retrieved, same number of select query executed for getting user details. This makes our system slow.
So we add usertab fields such as firstname,lastname, gender, image in other tables in addition to userid field.
So on data retrieval, the system become fast, but we faced another problem. If any changes in usertab table such as change in firstname, lastname, gender or image, we need to update other tables that contains user details. If we consider huge amount of data in other tables, how can I handle this?
We are using lucene index and C#.
Cassandra writes significantly faster and more efficient than reads.
That's why cassandra prefer Denationalization over normalization
Denormalization is the concept that a data model should be designed so that a given query can be served from the results from one row and query. Instead of doing multiple reads from multiple tables and rows to gather all the required data for a response, instead modify your application logic to insert the required data multiple times into every row that might need it in the future This way, all required data can be available in just one read which prevents multiple lookups.
When executing multiple update you can use executeAsync.
Session allows asynchronous execution of statements (for any type of statement: simple, bound or batch) by exposing the ExecuteAsync method.
//Execute a statement asynchronously using await
var rs = await session.ExecuteAsync(statement);
Source : https://www.hakkalabs.co/articles/cassandra-data-modeling-guide
I am working on creating cassandra tables to support various queries my application makes. So far it's great, except I'm having a problem with one particular case:
We have a tinder-like interface where a user can vote "up" or "down" on a photo. I would like to dynamically serve the top 5 photos per query, sorted according to a custom "score", which excludes photos that a user has already voted on.
The pool of photos and the votes per photo are potentially > 10,000 each.
I was thinking of creating a table with a column of "votes: set" containing user ids, but it doesn't help because you can't query by NOT CONTAINS.
Is there a way to perform this query efficiently in Cassandra?
I have two tables which both include a date field. Currently I have two portals, one for each table (occurrence).
Is it was possible to display the results of both of these in one portal, and sort by the date?
Technically a portal can only display records from one table. If you need to join two tables then you have to do this manually or change the design and use one table instead of two (since you want them in the same portal, then the tables are similar to some degree; maybe this similarity can go into its own table).
Sometimes developers use the so-called virtual table technique: they create a table with, say, a field with the record number and a bunch of calculated fields that pick their values from elsewhere, for example, from prefilled global variables. They create a portal to this table, set up the relationship to display the required number of records, and write the code to fill these variables. This way they can show data that isn't stored in any table, combine tables, etc. But it's an arcane technique, I would recommend it only as the last resort.
Each user of my system can have contacts. Each contact has details like Name, Address, Email, Phone, etc.
Do you think is a good idea to store this contacts in Azure Tables? I am worried about the following:
How do I search for a specific field (like Email or Phone)?
How do I get only the contacts belonging to a specific user?
How do I sort the contacts by a field?
I think that contacts could be a good candidate for storing in Table Storage - but only if you can partition on the owning person and never really need to search or aggregate across multiple owning users.
One possible design for this is:
store the contacts once with the owning user as partition key and some unique field for row key, but with the fields as columns within each row.
How do I search for a specific field (like Email or Phone)?
You can then ask table storage to search within a partition - it will then do a scan within that partition - which shouldn't be particularly large or slow for any single partition.
How do I get only the contacts belonging to a specific user?
This is just a simple query by partition key only
How do I sort the contacts by a field?
All results from table storage are sorted by (partitionkey, rowkey) so to sort the contacts for a user, you'll need to query for all of them, and then sort them within your web or worker role.
Other designs are, of course, possible -
e.g. you could store each contact in multiple rows in multiple tables - this would then allow you to have pre-formed sort orders within the table storage.
e.g. you could use separate tables instead of separate partitionkeys for each user - this has the advantage that when you delete a user, you can delete the entire table belonging to that user.
Note... while it's possible to use table storage for this... actually I almost always seem to end up back in SQL Azure at the moment - it's just so much more powerful and predictable (IMO). When the team deliver secondary indexing, then I might be tempted to use it for more of my data.