Cosmos DB Partitioning Notes

Cloud/Azure/CosmosDB/README.md

@@ -0,0 +1,66 @@
+# Data Modeling and Partitioning Patterns in Azure Cosmos DB
+
+## Definition
+
+NoSQL Database that runs on Azure
+
+* Global Distribution
+* Horizontally Partitioned
+* Provisioned Throughput
+* Schema-Free
+
+## SQL
+
+| Customers | Products | Sales Orders |
+|----------------|----------|--------------|
+| Addresses (1:many) | Categories (1:many) | Details (1:many) |
+| Password (1:1) | Tags (many:many) |  |
+
+## Cosmos DB
+
+| Structure | Thoughts | Performance |
+|-|-|-|
+| **Container per Table** | Should you? worst design| Perform poorly and difficult to maintain. |
+|  |  |  |
+
+### Partitioning
+
+Logical Partitions based on the `partitionKey` groups documents together. Guaranteed to be phyically stored on the same server.
+
+Example:
+
+Partition key: `/username`
+
+All documents related to a user are stored in logical partitions.
+
+#### Constraints
+
+* Maximum document size: 2 MB.
+* Maximum logical partition size: 20 GB.
+
+#### Hot Partitions
+
+Some large partitions versus some smaller partitions. This generates "hot" partitons. Try for good distributions and access patterns (read and write).
+
+#### Single-partition versus Cross-partition Queries
+
+1. **Single-partition Query**: Retrieving all records for a single user.
+2. **Cross-partition Query**: Run on all servers in the cluster (fan-out queries). Generally ***undesireable*** (avoid most of the time).
+
+#### Choosing a Partition Key
+
+Know the operations:
+
+1. Creating a record.
+2. Query that record by `id`.
+
+### Rework
+
+1. Convert to JSON formatted structure.
+2. Convert ID fields to `id`.
+3. Embed linked information into parent object; combining tables into the structure.
+
+#### Embedding versus Reference
+
+* **Embed**: (1:1) or (1:few) relationship. Related items are queried or updated together.
+* **Reference**: (1:many) or (many:many) relationships. Related items are queried or updated seperately.