Apache Druid vs Snowflake

Compare and contrast Apache Druid and Snowflake by architecture, ingestion, queries, performance, and scalability.

Compare Apache Druid to Rockset here

Compare Snowflake to Rockset here

Apache Druid vs Snowflake Architecture

Architecture

Apache Druid

Snowflake

Deployment model

• SaaS or self managed. • Imply’s Druid cloud offering requires customers to configure, scale, and capacity plan

SaaS - infrastructure, software and cluster ops managed by service provider

Use of storage hierarchy

• Queries are served from data on disk and an in-memory cache • Cloud storage or HDFS for deep storage

Cloud object storage for shared data accessible from any virtual warehouse

Isolation of ingest and query

• Ingestion and queries are run on the same node by default • The ingestion and querying processes can be run on separate nodes, however not for real-time data

Yes - separate virtual warehouses for batch data loading, ELT jobs and queries

Separation of compute and storage

No, although Imply’s offering separates compute and storage

Yes

Isolation for multiple applications

Yes - separate virtual warehouses for each workload

Druid’s architecture employs nodes called data servers that are used for both ingestion and queries. High ingestion or query load can cause CPU and memory contention compared with Druid alternatives. Breaking apart the pre-packaged ingestion and query server components involves planning ahead and additional complexity, and is not dynamic.

Snowflake is the data warehouse built for the cloud. Snowflake is well-known for separating storage and compute for better price performance. With Snowflake, multiple virtual warehouses can be spun up or down for batch data loading, transformations and queries all on the same shared data.

Apache Druid vs Snowflake Ingestion

Ingestion

Apache Druid

Snowflake

Data sources

Data connectors to: • Events streams (e.g. Kafka, Kinesis) • Data lakes (e.g. S3, Google Cloud Storage) • RDBMS and HDFS databases CDC events from databases require manual conversion to Druid events

• Third party ETL tool to ingest data into Snowflake including Fivetran, Hevo or Striim • Bulk loading from S3, GCS, Azure Blob Storage • Sink Connector for Apache Kafka in Confluent Cloud

Semi structured data

Druid requires flattening nested data at ingest and maintaining a flattening spec as the schema changes over time

Ingests JSON and XML as a VARIANT data type

Transformations and rollups

Yes, using ingestion specs written in JSON support rollups and simple transformations. SQL ingest transformations available for Imply

• Third party ELT/ETL tools like dbt • Simple COPY commands at data loading for column recording, omission and casts

Druid has built-in connectors that manage ingestion from common data sources. Unlike some Druid competitors, it doesn’t support nested data, so data must be flattened at ingest. Denormalization is also required at ingest, increasing operational burden for certain use cases.

Snowflake is an immutable data warehouse that is built for batch ingestion and relies heavily on the modern data stack ecosystem for data connectors and transformations. Snowflake has a number of integrations to ETL and ELT solutions including Fivetran, Hevo, Striim and dbt. While Snowflake does have support for semi-structured data in the form of a VARIANT type, it is best to structure the data for optimal query performance.

Apache Druid vs Snowflake Performance

Performance

Apache Druid

Snowflake

Updates

Druid segments become immutable once committed and published, making it more appropriate for append-only use cases

Data warehouse with immutable storage. Updates rewrite and merge entire partitions

Indexing

• Bitmap index

Query latency

Typically sub-second query latency for denormalized, flattened datasets up to 100s of TBs

Seconds to minutes on petabytes of data

Storage format

Columnar format partitioned by time

Compressed columnar format stored in cloud object storage

Streaming ingest

• Sub 15 seconds

• Ingests on a batch basis • Snowpipe typically ingests in minutes

Druid is designed to make streaming data queryable as quickly as possible. JOINs are either impossible or incur a large performance penalty. Updates are only possible via batch jobs. Druid leverages data denormalization and write-time aggregation at ingestion to reduce query latency.

Snowflake is designed for batch analytics with analysts and data scientists infrequently accessing large-scale data for trend analysis. Snowflake, like many data warehouses, is immutable and does not support frequently changing data efficiently. Snowflake uses a columnar store to return aggregations and metrics efficiently, often with query response times in the seconds to minutes on petabytes of data.

Apache Druid vs Snowflake Queries

Queries

Apache Druid

Snowflake

Joins

Yes, only for datasets that fit in memory, and with a query latency penalty

Yes

Query language

Druid native queries • Druid SQL

SQL

Developer tooling

• Druid SQL API

• SQL APIs - make SQL calls to Snowflake programmatically • UDFs for Javascript, Python, Java and SQL functions • Go, JDBC, .NET, Node.js, ODBC, PHP, Python drivers

Visualization tools

Pivot, maintained by Imply

Integrations with QuickSight, Chartio, Domo, Looker, PowerBI, Mode, Qlik, Sigma, Sisense, Tableau, ThoughtSpot and more

Druid has a native JSON-based query language and provides Druid SQL as an alternative that translates into its native queries. JOINs are not recommended.

Snowflake supports SQL as its native query language and can perform SQL joins. Snowflake for developers introduced a number of developer tools including SQL APIs, UDFs and drivers to support application development. As Snowflake was originally built for business intelligence workloads, it integrates with a number of visualization tools for trend analysis.

Apache Druid vs Snowflake Scalability

Scalability

Apache Druid

Snowflake

Vertical scaling

Users can manually increase or decrease server sizes

Resize virtual warehouses via web interface or using DDL commands for warehouses

Horizontal scaling

• Users can manually add additional nodes to a scaled-out cluster. • Imply automates some of the horizontal scaling process.

• Multi-cluster warehouses allocate additional clusters for higher concurrency workloads • Auto scaling policies can be set

Druid users are exposed to complex decisions about the number and size of servers as clusters are scaled.

Snowflake virtual warehouses can be scaled up for faster queries or scaled out using multi-cluster warehouses to support higher concurrency workloads. Snowflake has shared blob storage that scales automatically and independently.