Layout of Janusgraph backend tables - cassandra

I am using JanusGraph with Cassandra as its storage backend. Its a fairly large graph with over 70 million vertices. What is the best way to get a list of all the vertex IDs. The query g.V() will take a long time to return.
Is there a way I can pull out this data (all vertex IDs and if possible all edge IDs) directly from the Cassandra tables. I couldn't figure out how the data is stored. Can anybody help me please, or is there any clear documentation on how data is laid out in the backend tables.

The way JanusGraph data is stored in Cassandra is documented at https://docs.janusgraph.org/advanced-topics/data-model/ but if you use CQL or Nodetool to look at the tables created you will see the data is stored as blobs. I think the only way to get IDs is using Gremlin.

Related

Columnar/Column-oriented database vs wide-column/column family database

I got really confused about Cassandra recently because most online material even AWS and Google describe it as columnar database. But actually it is row-based, partitioned, column-family database. Now everything makes sense to me:
Then are all characteristics such as
being optimised for foreign join
aggregating many rows and a few columns/ single column aggregation
Scanning only column by column as opposed to row-based database
still valid or true for Cassandra AKA wide-column/column-family database. If not, what are their real characteristics? Write Performance?
IN addition, I really need a true columnar database example to study on?(as opposed to Cassandra and data are store in different column blocks)
Can anyone help? It troubles me for a long time.

Columnar storage: Cassandra vs Redshift

How is columnar storage in the context of a NoSQL database like Cassandra different from that in Redshift. If Cassandra is also a columnar storage then why isn't it used for OLAP applications like Redshift?
The storage engines of Cassandra and Redshift are very different, and are created for different cases.
Cassandra's storage not really "columnar" in wide known meaning of this type of databases, like Redshift, Vertica etc, it is much more closer to key-value family in NoSQL world. The SQL syntax used in Cassandra is not any ANSI SQL, and it has very limited set of queries that can be ran there. Cassandra's engine built for fast writing and reading of records, based on key, while Redshift's engine is built for fast aggregations (MPP), and has wide support for analytical queries, and stores,encodes and compresses data on column level.
It can be easily understood with following example:
Suppose we have a table with user id and many metrics (for example weight, height, blood pressure etc...).
I we will run aggregate the query in Redshift, like average weight, it will do the following (in best scenario):
Master will send query to nodes.
Only the data for this specific column will be fetched from storage.
The query will be executed in parallel on all nodes.
Final result will be fetched to master.
Running same query in Cassandra, will result in scan of all "rows", and each "row" can have several versions, and only the latest should be used in aggregation. If you familiar with any key-value store (Redis, Riak, DynamoDB etc..) it is less effective than scanning all keys there.
Cassandra many times used for analytical workflows with Spark, acting as a storage layer, while Spark acting as actual query engine, and basically shouldn't be used for analytical queries by its own. With each version released more and more aggregation capabilities are added, but it is very far from being real analytical database.
I encountered the same question today, and found that this resource on AWS: https://aws.amazon.com/nosql/columnar/

How to export large Neo4j datasets for analysis in an automated fashion

I've run into a technical challenge around Neo4j usage that has had me stumped for a while. My organization uses Neo4j to model customer interaction patterns. The graph has grown to a size of around 2 million nodes and 7 million edges. All nodes and edges have between 5 and 10 metadata properties. Every day, we export data on all of our customers from Neo4j to a series of python processes that perform business logic.
Our original method of data export was to use paginated cypher queries to pull the data we needed. For each customer node, the cypher queries had to collect many types of surrounding nodes and edges so that the business logic could be performed with the necessary context. Unfortunately, as the size and density of the data grew, these paginated queries began to take too long to be practical.
Our current approach uses a custom Neo4j procedure to iterate over nodes, collect the necessary surrounding nodes and edges, serialize the data, and place it on a Kafka queue for downstream consumption. This method worked for some time but is now taking long enough so that it is also becoming impractical, especially considering that we expect the graph to grow an order of magnitude in size.
I have tried the cypher-for-apache-spark and neo4j-spark-connector projects, neither of which have been able to provide the query and data transfer speeds that we need.
We currently run on a single Neo4j instance with 32GB memory and 8 cores. Would a cluster help mitigate this issue?
Does anyone have any ideas or tips for how to perform this kind of data export? Any insight into the problem would be greatly appreciated!
As far as I remember Neo4j doesn't support horizontal scaling and all data is stored in a single node. To use Spark you could try to store your graph in 2+ nodes and load the parts of the dataset from these separate nodes to "simulate" the parallelization. I don't know if it's supported in both of connectors you quote.
But as told in the comments of your question, maybe you could try an alternative approach. An idea:
Find a data structure representing everything you need to train your model.
Store such "flattened" graph in some key-value store (Redis, Cassandra, DynamoDB...)
Now if something changes in the graph, push the message to your Kafka topic
Add consumers updating the data in the graph and in your key-value store directly after (= make just an update of the graph branch impacted by the change, no need to export the whole graph or change the key-value store at the same moment but it would very probably lead to duplicate the logic)
Make your model querying directly the key-value store.
It depends also on how often your data changes, how deep and breadth is your graph ?
Neo4j Enterprise supports clustering, you could use the Causal Cluster feature and launch as many read replicas as needed, run the queries in parallel on the read replicas, see this link: https://neo4j.com/docs/operations-manual/current/clustering/setup-new-cluster/#causal-clustering-add-read-replica

Big data solution for frequent queries

I need a big data storage solution for batch inserts of denormalized data which happen infrequently and queries on the inserted data which happen frequently.
I've gone through Cassandra and feel that its not that good for batch inserts, but an OK solution for querying. Also, it would be good if there was a mechanism to segregate data separately based on a data attribute.
As you mentioned Cassandra I will talk about it:
Can you insert in an unbatched way or is this impossed by the system? If you can insert unbatched, Cassandra will probably be able to handle it easily.
Batched inserts should also be handable by Cassandra nodes, but this won't distribute the load properly among all the nodes (NOTE: I'm talking about load balancing, not about data balance, which will be only depending on your partition key setup). If you are not very familiar with Cassandra you could tell us your data structure and your query types and we could suggest you how to use Cassandra's data model to fit it.
For the filtering part of the question, Cassandra has clustering keys and secondary indexes, that are basically like adding another column configuration to the clustering key so that you have both for querying.

PySpark Cassandra Connector efficiently querying across partition keys

I'm faced with the following problem using PySpark and dataframes with the cassandra-connector. My Cassandra data lake consists of metric measurements across (network) devices, and the entries are of type (device,interface,metric,time,value).
My cassandra table for the raw data has:
PRIMARY KEY ((device,interface,metric),time)
for supposedly efficient fetching of time ranges for a given measurement.
Now for reporting purposes, users can query any set of device/interface/metric combinations (ie give me a specific metric for all interfaces of a device). Now I know the list of each, so I'm not looking to do wildcard searches, but rather IN queries.
I'm using Spark 1.4, so I'm adding filters like to obtain dataframes to calculate min/max/percentile/etc... on the recorded metric values.
metrics_raw_sub = metrics_raw\
.filter(metrics_raw.device.inSet (device_list))\
.filter(metrics_raw.interface.inSet (interface_list))\
.filter(metrics_raw.metric.inSet (metric_list))
This isn't very efficient as these predicates do not get pushed down to CQL (only the last predicate can be an IN query), so I'm pulling in tons of data and filtering on the client side. (not good)
Why doesn't cassandra-connector allow multiple IN predicates across partition columns? Doing this in a native CQL shell appears to work?
Another approach to my problem above would be to (and this yields efficient individual queries as predicates are pushed down to Cassandra):
for device in device_list:
for interface in interface_list:
metrics_raw_sub = metrics_raw\
.filter(metrics_raw.device == device)\
.filter(metrics_raw.interface == interface)\
.filter(metrics_raw.metric.inSet (metric_list))
And then run the aggregation logic for each subquery, but I feel like this is largely serialising what should be a parallel computation across all requested device/interface/metric values... Can I batch the Cassandra queries so I can run my analytics on one large distributed dataframe?
Bottom line, I'm looking to do this very efficiently. If the turn-around times are short enough, we'll run these on-demand. If not, we'll need to look into pre-computing them and storing into tables (which sacrifices flexibility for doing custom time-range reporting)
Any insights would be much appreciated!!
Nik.

Resources