With Cassandra,
I want to represent all users objects with a unique uuid, but also contain a set of zero or more secondary user keys to map to a user. Each secondary key should map to one and only one user(id). Because I need to be able to quick lookup of secondarykey to find a user, I maintain a separate lookup table, instead of a secondary INDEX.
I've modelled the data like this, but I am open to alternatives:
CREATE TABLE users (
userid uuid PRIMARY KEY,
name text,
secondarykeys set<text>
);
CREATE TABLE user_secondarykeys (
secondarykey text,
userid uuid,
PRIMARY KEY(secondarykey)
);
A typical use case is this:
I got this user with a secondary key mail:andreas#example.org, and I would like to see if there exists any user with that secondary key, and if it do not exists, I would like to create a new user object.
I can look for the secondary key:
SELECT * FROM "user_secondarykeys" WHERE secondarykey = "mail:andreas#example.org";
and if I do not find any matches, I can insert a new user:
BEGIN BATCH
INSERT INTO users (userid, name, secondarykeys) VALUES (77059e45-5fac-460b-9c4f-47528c292be0, "Andreas", {'mail:andreas#example.org'});
INSERT INTO user_secondarykeys (secondarykey, userid) VALUES ('mail:andreas#example.org', 77059e45-5fac-460b-9c4f-47528c292be0);
APPLY BATCH;
My problem is that this can lead to inconsistent data, because a user can be inserted with that secondary key in the meantime between my select and my inserts.
I'm thinking that if I can make my INSERT transaction fail if the secondary key already exists in user_secondarykeys, that would work, because it should then also revert the insert into the users table, because of the atomic property of the transaction. However, I do not know any ways to make the INSERT fail if the secondary key exists. If I add IF NOT EXISTS to the second insert, it will not revert the trasaction it will just avoid inserting into user_secondarykeys, but it will still insert into users.
Any suggestions on how to implement this use case in a reliable way is appreciated. Thanks.
At first, I think that your model is pretty complicated, and I'm not sure if I understand correctly all of your requirements.
So if you get at first this secondary key, and then you have to decide what to do - add user or not - then the following will work for you:
Instead of checking user_secondarykeys table with SELECT statement for occurrence of particular secondary key, go with the following:
INSERT INTO user_secondarykeys (secondarykey, userid) VALUES ('mail:andreas#example.org', 77059e45-5fac-460b-9c4f-47528c292be0) IF NOT EXISTS;
So if it applies, it means that this secondary key is not connected with any user - so there are two cases: user doesn't exists or user exists and someone want's to add new secondary key for him. The following will do the job in both cases:
INSERT INTO users(userid, name, secondarykeys) VALUES(77059e45-5fac-460b-9c4f-47528c292be0, 'Andreas', secondarykeys = secondarykeys + 'mail:andreas#example.org')
Because inserts/updates in Cassandra are idempotent(except counters), this will work even if there will be already an user with that id in users table - this should just add another secondary key for him.
Pros of this solution are that you will remove this gap in time which can make you 'inconsistent'. You have a guarantee that no one will insert two users with the same secondary key. You specified that user can have no secondary keys at all - in this situation you can add him straight to the users table.
I'm thinking that if I can make my INSERT transaction fail if the secondary key already exists in user_secondarykeys, that would work, because it should then also revert the insert into the users table, because of the atomic property of the transaction. However, I do not know any ways to make the INSERT fail if the secondary key exists. If I add IF NOT EXISTS to the second insert, it will not revert the trasaction it will just avoid inserting into user_secondarykeys, but it will still insert into users.
Since Cassandra 2.0.6 you can use a conditional statements inside a batch, and if any of conditions will be not met then all instructions in that batch won't fire. This sounds great but there is a limitation - all of the statements inside batch have to operate on the single, same partition. According to this, it is impossible to make cross partition/table conditional insert/update/delete. So in your case this:
BEGIN BATCH
INSERT INTO users (userid, name, secondarykeys) VALUES (77059e45-5fac-460b-9c4f-47528c292be0, "Andreas", {'mail:andreas#example.org'});
INSERT INTO user_secondarykeys (secondarykey, userid) VALUES ('mail:andreas#example.org', 77059e45-5fac-460b-9c4f-47528c292be0) IF NOT EXISTS;
APPLY BATCH;
would not even pass the query validation, because you try here to operate on two different tables.
I'm not sure if this will be suitable for other of your requirements, I would need more information about your queries and the velocity/volume of the data. For sure there are other ways for modeling this.
It would greatly simplify the problem if every user would have to have at least one specified secondary key(e.g. email would be a great unique key for your users table), but that's are your requirements, so unless you can't change them there is no discussion.
Hope this will help you a bit.
Good luck!
Related
In Cassandra, I understand that tables are supposed to be created according to what needs to be queried. For example, I currently have a Users and Users_By_Status table.
##Users##
CREATE TABLE Users (
user_id uuid,
name text,
password text,
status int,
username text,
PRIMARY KEY (user_id)
);
CREATE INDEX user_username_idx ON Users (username);
##Users_By_Status##
CREATE TABLE Users_By_Status (
username text,
status int,
user_id uuid,
PRIMARY KEY (username, status, user_id)
);
In this case, if a user leaves, their record won't be deleted. Instead, status will be changed from 1 to 0.
If I insert data into the Users table, do I need to manually insert the data into Users_By_Status table too? What happens if I update the status in Users? Do I need to manually update the record in Users_By_Status table too?
I have a feeling I'm understanding Cassandra wrongly. Appreciate all the help I can get.
Shortly answer: yes, in your case you need to delete manually.
In cassandra db you need to write more code in your app layer to handle cenarios like that.
But we have other options like materialized view or BATCH Statements.
For your solution, i think that materialized view is the best option. You can create a Materialized view from your table Users. Like this:
CREATE MATERIALIZED VIEW Users_By_Status
AS SELECT username, status, userid
FROM Users
PRIMARY KEY(username, status, userid);
And yes, when you update table users, the update will happen in the Materialized View Users_By_Status too.
Reference: https://docs.datastax.com/en/cql-oss/3.3/cql/cql_using/useCreateMV.html
Do I need to manually update the record in Users_By_Status table too?
So CoutinhoBR alluded to it, but I'll come right out and say it. You cannot update primary key values in Cassandra. So that's where a DELETE is required to get the old status value out of there, and then a write for the new one.
I have a table called Media, where the primary key is the mediaId. I have an additional table called Media_Comments which has a commentId as the primary key, and a mediaId attribute that stores the mediaId that that comment is linked to. Same with Media_Likes, I have a primary key of mediaId and sort key of userId. I want to handle a case where a Media item is deleted by a user, which will then cause a mass deletion of all comments and likes of that Media item. I am currently writing this code in a lambda using Node.js.
I tried using a regular delete based on the condition of where 'mediaId = :mediaId', but it was complaining about needing a primary key for the table. Unfortunately, many times when I want a media item deleted I won't have specific key items available to fulfill that condition. I looked into trying to delete by a certain index, like setting a GSI on the mediaId in each table and deleting by that, unfortunately that does not seem to be an option either.
Basically, am I missing something? Is there actually a way to delete by an index? And if not, what would be the best way to do this? Setting a TTL for each item in dynamodb that is affiliated with the Media item? Or is there another recommended way to handle this problem?
Any help is greatly appreciated, thank you.
When I tried to retrieve table using contains keyword it prompts "Cannot use CONTAINS relation on non collection column col1" but when I tried to create table using
CREATE TABLE test (id int,address map<text, int>,mail list<text>,phone set<int>,primary key (id,address,mail,phone));
it prompts "Invalid collection type for PRIMARY KEY component phone"
One of the basics in Cassandra is that you can't modify primary keys. Always keep that in mind.
You can't use a collection as primary key unless it is frozen, meaning you can't modify it.
This will work
CREATE TABLE test (id int,address frozen<map<text, int>>,mail frozen<list<text>>,phone frozen<set<int>>,primary key (id,address,mail,phone));;
However, I think you should take a look at this document: http://www.datastax.com/dev/blog/cql-in-2-1
You can put secondary indexes on collections after cql 2.1. You may want to use that functionality.
The question is that, When we use
INSERT INTO USERS (login, email, name, login_count)
values ('jbellis', 'jbellis#datastax.com', 'Jonathan Ellis', 1)
IF NOT EXISTS
in IF NOT EXIST exactly which columns are compared together? primary key(partition-key + clustering-key)? or just partition-key?
Here is a diagram of the 4 phases of LWT: http://www.slideshare.net/doanduyhai/cassandra-introduction-nantesjug/89
The original blog post is here: http://www.datastax.com/dev/blog/lightweight-transactions-in-cassandra-2-0
exactly which columns are compare together? primary key? partition-key?
The whole primary key is checked on disk. In the example it is a simple primary key so it's also the partition key, e.g. login column
I want to bulk insert an array of data using NodeJS and RethinkDB but I don't want to insert existing records (where name & value already has a record, I don't want to dupcheck on primary key id).
[
{name:"Robert", value:"1337"},
{name:"Martin", value:"0"},
{name:"Oskar", value:"1"}
]
If any of the above values already exist, don't insert, but update "value".
My current working solution is that I loop through the array and first check if it exists using a filter, if not, i insert it. But it's very slow on 10.000 records.
I don't think we have that kind of concept in RethinkDB. I tried to read the doc more. To insert a new document, use insert, to update field, use update, to replace to a whole new document, use replace(the primary key won't change)...So I don't think it's possible in RethinkDB.
Here is some way you can make it run faster:
Create a compound index contains those two fields: name and value
Then using that index to check for existence instead of using filter
Generate your own id field, instead of letting RethinkDB generated it. Therefore, you know the primary key, and use it to look up document with get which will be very fast.
I had a similar requirement in a RethinkDB project, but in that case the primary key was being checked for duplicates, and it was also custom instead of being auto-generated.
What you could do is run an async.series or async.waterfall two-step check. First pick a single object from your array, then filter the database for the name-value pairs of your current object. If the results come up null, it is unique. If not, you have a pre-existing record with same details.
Depending on the result, you can then pass on the control to next step which will either insert the new document or update existing one. It will be simpler if you use a flag for this in async.waterfall.