Concurrency Control

RavenDB

Optimistic

Optimistic, using Vector Clock as "timestamps".

MongoDB

Pessimistic

Pessimistic, multiple granularity locks (multiple readers, single writer).

PostgreSQL

Pessimistic

Pessimistic, optionally explicit (manual) locking.

CosmosDB

Optimistic

Optimistic, using ETAGs as "timestamps".

Single-node transactions

RavenDB

Supported

Supported.

MongoDB

Supported

Supported.

PostgreSQL

Supported

Supported.

CosmosDB

Supported*

Supported.

*In CosmosDB, multi-document transaction requires running it in server-side procedure - client API support only single document changes.

Distributed Transactions

RavenDB

Ensures writes on majority of nodes

Ensures writes (assurance by consensus protocol) on majority of nodes.

MongoDB

Ensures writes on majority of nodes

Ensures writes (wait for replication) on majority of nodes.

PostgreSQL

Two-phase commit

Two-phase commit.

CosmosDB

Multiple options

Flexible, can ensure writes on majority or several more options.

Notes: RavenDB ensures majority of writes using Raft, MongoDB ensures majority of writes by waiting on async replication (for MongoDB assuming the setting writeConcern=majority).

Querying

RavenDB

Fully fledged search-engine capabilities

Allows simple and Javascript projections, facets, dynamic index declarations, loading related documents and integrated spatial and full-text search syntax.

MongoDB

Basic functionality

Allows simple (field level) projections, spatial and full-text searches.

PostgreSQL

SQL, JSON column support

Allows SQL queries, supports JSON as data columns.

CosmosDB

Multiple querying APIs

Support multiple querying APIs, such as MongoDB, SQL, Gremlin. Its own API can do queries with simple projections and filtering.

Indexing

RavenDB

Lock-free, uses DSL definitions for flexibility

Async, reacts to data changes, doesn't lock on data, can run user-code, computations and projections on field values while indexing in background.

MongoDB

Sync, locks, basic functionality

Sync, updates index as part of CRUD while locking data, indexes only field values.

PostgreSQL

Sync, locks, basic functionality, some customization

Sync, updates index as part of CRUD while locking data, indexes only field values. Can define different underlying data structures as optimization.

CosmosDB

Basic functionality, index on all fields

Reverse index on all fields in documents.

Notes: As of version 4.2, MongoDB does exclusive locking on collection at beginning and ending of indexing creation process.

Spatial Indexing

RavenDB

Highly configurable, supports WKT

Supports linear geometry, configurable indexing methods and distance units (miles/km). Supports raw longitude, latitude coordinates and standard WKT data.

MongoDB

Supports GeoJSON

Supports linear and spherical geometry, data stored as GeoJSON.

PostgreSQL

3rd-party

Supported using 3rd-party extension.

CosmosDB

Supports GeoJSON

Supports linear geometry, data stored as GeoJSON.

Notes: In PostgreSQL spatial queries are supported with PostGIS extension.

Full-text search

RavenDB

Offers all features of Lucene.Net to power rich text search

Highly configurable, leverages Lucene.Net to power rich text search, can use custom analyzers (value tokenizers). For example, custom analyzers can be used to implement efficient (and fast!) wildcard search (n-gram analyzers) or handling language specific characters.

MongoDB

Simple tokenization

Tokenizes by whitespace and punctuation, stemming, stopwords. Optional tokenization by wildcard.

PostgreSQL

Simple tokenization with configurable dictionaries

Configurable value tokenizers and "thesaurus" dictionaries for term mapping.

CosmosDB

Azure Search or 3rd-party

Either Azure Search/Cognitive Search or 3rd-party search engines.

Notes: With full-text indexes, MongoDB tokenization and text processing cannot be customized. PostgreSQL full-text search is not index-based, needs to run UPDATE statements to keep full-text search up-to-date.

Aggregation

RavenDB

Declarative definition, reacts to data changes

Async map/reduce over Lucene indexes, reacts to data changes, has all features of RavenDB indexes such as computations and projections.

MongoDB

Introduces read-locks and needs to be run manually

Aggregation/Map-reduce, both sync, introduce read-locks so they may stall because of long writes.

PostgreSQL

SQL aggregations

SQL aggregation (standard GROUP BY queries).

CosmosDB

Limited SQL-like aggregation functions

Limited SQL-like aggregation functions from DocumentDB SQL API.

Notes: For MongoDB, aggregation pipeline is more optimized but limited to one collection, map-reduce requires providing Javascript code for map and reduce stages. For CosmosDB, note that if you execute aggregation as part of stored procedure, there is a limit on execution time of stored procedures - your stored procedure may run into it.

Replication

RavenDB

Multiple replication strategies, high availability and load-balancing

Out of the box high availability (active-active topology), allow for multiple replication strategies (cross cluster replication, star topology replication) and load balancing between replica sets.

MongoDB

High availability, needs special maintenance in some use cases

Out of the box high availability (active-active topology), in case of even number of nodes, need "special" node.

PostgreSQL

Active-passive only

Active-passive topology only, supports either log-shipping or streaming to replicate data.

CosmosDB

Multiple replication strategies, high availability and load-balancing

Out of the box high availability and load balancing, support for custom load balancing strategies, supports active-active topology.

Notes: MongoDB and RavenDB implement Raft Consensus Protocol for maintaining replicated clusters and provide failover functionality.

Sharding

RavenDB

Not supported*

Not supported.

MongoDB

Multiple sharding strategies, needs specialized services

Multiple sharding strategies, background shard balancing, needs specialized services as part of sharding topology (query router, configuration servers).

PostgreSQL

Partitioning-based sharding*

Built-in sharding with declarative partitioning (using Foreign Data Wrapper).

CosmosDB

Sharding on user-defined key

Out of the box sharding (on user-defined key).

*For RavenDB, sharding support is scheduled for release in Q4 of 2020. For PostgreSQL, the sharding will work only if there is no brain split.

ETL

RavenDB

Real-time, projections and transformations

Supported, async transfer of data between either RavenDB -> RavenDB or RavenDB -> SQL, use Javascript to define transformation scripts.

MongoDB

3rd-party

3rd-party vendors.

PostgreSQL

3rd-party

3rd-party vendors.

CosmosDB

Real-time, projections, transformations, multiple types of destinations

Supported, transfer of data to/from CosmosDB and multiple sources. Can use Data Factory UX and properietary dataflow language to define transformation scripts.

Notes: For MongoDB and PostgreSQL ETL can be implemented with 3rd-party tools.

Database Event Notifications

RavenDB

Batch processing, projections and transformations

Supported, batch processing, declarative filtering, loading related documents and complex projections with optional calculations.

MongoDB

Basic filtering and field projection

Supported, allow filtering and simple field projection.

PostgreSQL

Projections and transformations

Supported using LISTEN/NOTIFY (event based processing).

CosmosDB

Projections and transformations, multiple types of destinations

Supported, allows sending events to client API and other destinations (like queues).

Notes: For RavenDB, filtering and projections of changed documents have the same flexible capabilities as queries, uses the same query syntax. Also, RavenDB's Data Subscription (the relevant feature) is persistant - if an offline client misses change events, they will be streamed in-order when the client comes back online.

Storing binary blobs

RavenDB

Supported, transactional

Supported, also supports multi-attachment transactions - documents and binary blobs can be changed as part of single transaction.

MongoDB

Supported, non-transactional

Supported, but doesn't support transactions.

PostgreSQL

Supported, transactional

Supported by using special column type or using Large Object feature.

CosmosDB

Supported

Supported, binary attachments can have indexable properties.

Management Tools

RavenDB

Integrated

Built-in administration web-app, fully-featured solution for database administration and devops.

MongoDB

3rd-party

3rd-party vendors.

PostgreSQL

3rd-party

3rd-party vendors.

CosmosDB

Integrated

Azure UI contains management section for CosmosDB.

DBaaS Costs

RavenDB

Flat-rate

Easy to understand flat-rate that depends on provider (AWS, Azure, GCP).

MongoDB

Flat-rate with additional costs

Flat-rate that depends on provider (AWS, Azure, GCP). Some of services like full-text search and backups have additional costs.).

PostgreSQL

3rd-party provider pricing

Depends on 3rd-party provider of PostgreSQL-as-a-service.

CosmosDB

Resource Units costs

Resource Units costs - complex to figure out.

Databases Comparison