DP-420 flashcards
DP-420 · Build recall with a focused review session.
32 cards. Read the term and the definition here. The deck at the top of this page is optional.
- Partition key immutability
- A container's partition key path is fixed at creation; you cannot change it later. Migrating to a new key requires creating a new container and copying data.
- Hierarchical partition key
- A container can define up to three partition key levels, such as tenantId, userId, and sessionId. Queries that supply the full prefix are pruned to a single logical partition.
- Transactional batch scope
- A transactional batch commits atomically only if every operation targets the same logical partition key. Limits are 100 operations and a 2 MB total request payload.
- Session consistency default
- Session is the default consistency level. It guarantees monotonic reads, monotonic writes, read-your-writes, and writes-follow-reads for one client session using a session token.
- Strong consistency constraint
- Strong consistency guarantees linearizability but limits the account to a single write region and cannot be combined with multi-region writes.
- All-versions-and-deletes change feed
- This change feed mode returns creates, updates, and deletes along with intermediate versions. It requires the account to have continuous backup, meaning point-in-time restore, enabled.
- Stored procedure execution limit
- Stored procedures run inside a single logical partition key and must finish within five seconds on the server; exceeding that time aborts the transaction.
- ETag optimistic concurrency
- Every item carries an _etag value. Send it in an If-Match header through ItemRequestOptions; a mismatch returns HTTP 412 Precondition Failed and prevents lost updates.
- RU cost of point read
- A point read of a 1 KB item by id and partition key costs 1 request unit. Larger items cost more, and queries cost more than equivalent point reads.
- Indexing root path exclusion
- Only the root paths /* and /? can be excluded from indexing. Excluding /* and re-including specific paths lowers write request-unit cost on write-heavy containers.
- TTL precedence rules
- Item-level ttl overrides the container default. A value of -1 means the item never expires; if the container DefaultTimeToLive is null, TTL is disabled entirely.
- UDF restrictions
- User-defined functions are side-effect free, cannot perform I/O, and run within a limited CPU and memory budget. They are usable only inside queries, not as standalone calls.
- Cross-partition query fan-out
- A query without the partition key in its filter fans out to every physical partition, raising latency and request-unit charge. Prefer filters that include the partition key.
- Cosmos DB built-in data roles
- Cosmos DB Built-in Data Reader and Cosmos DB Built-in Data Contributor are data-plane RBAC roles assignable to Microsoft Entra identities at account, database, or container scope.
- disableLocalAuth setting
- Setting disableLocalAuth to true on the account rejects requests authenticated with account keys or resource tokens, forcing all clients to use Microsoft Entra ID tokens.
- Customer-managed key prerequisite
- Customer-managed keys are stored in Azure Key Vault and must be configured when the account is created; the account uses a managed identity to unwrap the key.
- Dedicated gateway requirement
- The integrated cache works only through a dedicated gateway provisioned on the account; cache hits on point reads and queries are served without consuming provisioned request units.
- Autoscale billing behavior
- Autoscale adjusts between 10 percent of the configured maximum RU/s and that maximum, and billing uses the highest RU/s reached during each hour.
- Indexing mode none
- Setting indexingMode to none stops all index maintenance and minimizes write cost. Reads then require scans, so use it only with analytical store or change feed consumers.
- Request charge header
- The x-ms-request-charge response header, exposed as RequestCharge in the SDK, reports the request units consumed by one operation; query metrics break down fan-out cost.
- Bulk execution mode
- Setting AllowBulkExecution to true lets the SDK group many concurrent item operations into batches that saturate throughput. It does not apply to queries or stored procedures.
- Serverless account limits
- Serverless accounts bill per request unit consumed and support only a single write region with no provisioned throughput; they suit intermittent or development workloads.
- Failover priority list
- Each replicated region has a failover priority. Automatic failover promotes the next region in the list when the write region is unavailable; manual failover switches immediately.
- Vector embedding policy
- A container policy declaring each vector path, its data type (float32, float16, int8, uint8), the number of dimensions, and the distance function: cosine, dotproduct, or euclidean.
- DiskANN vector index
- A vectorIndexes entry of type diskANN provides low-latency approximate nearest-neighbor search at scale and is the recommended index for large vector collections.
- quantizedFlat vector index
- A vectorIndexes entry of type quantizedFlat stores compressed vectors to reduce memory use, trading some recall and suited to smaller vector collections.
- VectorDistance ordering
- ORDER BY VectorDistance(property, @embedding) combined with a TOP clause returns the nearest neighbors ranked by the container's configured distance function.
- Analytical store columnar format
- Synapse Link stores data in a columnar analytical store that syncs automatically from the transactional store; analytical queries do not consume transactional request units.
- Analytical TTL values
- analyticalStoreTimeToLiveInSeconds of -1 retains analytical data indefinitely, while 0 disables the analytical store; the setting is independent of transactional TTL.
- Change feed embedding pipeline
- An Azure Functions change feed trigger sends new or updated items to Azure OpenAI, then writes the returned embedding back into the same container for vector search.
- RAG retrieval pattern
- Embed the user question, run a top-k vector query in Cosmos DB, and inject the retrieved documents into the model prompt as grounding context before generation.
- Azure AI Search indexer
- A Cosmos DB indexer in Azure AI Search pulls items through the change feed and adds full-text search, filtering, and semantic ranking on top of the same documents.