Data 360 tools
Every Data 360 connector we could verify
Connection method, direction, refresh mode and what actually catches people out, for each source, with the Salesforce page each fact came from and the date we read it.
Verified against Salesforce documentation on .
What this table covers
Salesforce publishes a connector directory that lists roughly 150 named connectors, and it changes often. This table is the 35 sources whose method, direction and refresh behaviour we verified individually against Salesforce documentation. A source that is not here is not a source Data 360 lacks: it is one we have not checked yet.
| Source | Method | Direction | Refresh | What to watch for |
|---|---|---|---|---|
| Salesforce CRMSalesforce | Native connector | Data in | Incremental refresh every 10 minutes, starting only after a full refresh. Periodic full refresh is disabled by default in new data streams. | The incremental runs every 10 minutes, not hourly, and it does not start until a full refresh has completed. Because periodic full refresh is off by default on new streams, anything the incremental cannot detect stays stale until you enable it or run a manual refresh. Initial ingestion is always batch; including batch-ingested formula fields, in-development custom entities, deleted-record DLOs or stream filters forces the stream back to batch mode. A standard CRM connection sends nothing back to the connected org except data actions, and big objects are not supported at all.Not documented by Salesforce: List of derived fields that break CRM streaming eligibility. |
| B2C CommerceSalesforce | Native connector | Data in | Order Bundle loads 30 days of historical data when it is created. Product and catalog streams run a full refresh daily. | The connector works against production sites only, so there is no sandbox rehearsal of the real data shape. The 30 day history on the Order Bundle is what you get at creation: anything older has to arrive another way. |
| Marketing Cloud EngagementSalesforce | Native connector | Data in | Pulls daily or hourly depending on the data type, scheduled automatically through Automation Studio. Data extension full extract daily, delta extract hourly. | Salesforce warns against changing the schedule after the stream is created. Data extension deltas are hourly and the full extract daily, so a segment built on the assumption of fresher extension data will be wrong for most of the day. |
| Marketing Cloud Account EngagementSalesforce | Native connector | Not documented | Not documented | Listed as a first-party connector, and that is the whole of what the sources this table cites say about it. The data stream schedule article does not cover it, so treat any cadence you have read elsewhere as unverified until you check it against your own org. |
| Marketing Cloud PersonalizationSalesforce | Native connector | Bidirectional | Hourly. Engagement event streams update incrementally within about 2 minutes in insert mode; user streams run a full refresh within about 15 minutes in upsert mode. | Only users who engage after the bundle is deployed are synced, so the profile set starts empty rather than backfilled. It is also an activation target, which is why the direction here is bidirectional while the other first-party sources are data in. |
| Data Cloud OneSalesforce | Native connector | Not documented | Not documented | Listed as a first-party connector for multi-org companion access. Neither its direction nor its refresh behaviour appears in the articles this table cites. |
| Omnichannel InventorySalesforce | Native connector | Not documented | Not documented | Listed as a first-party connector. Direction and refresh are not documented in the articles this table cites, so both are left unstated rather than guessed. |
| SnowflakeWarehouse and lakehouse | Zero copy federationZero Copy | Bidirectional | Live query federation. Acceleration caching is configurable from 15 minutes to 7 days, and incremental cache updates are upserts only. | Zero copy federation is the data in half and data sharing is the data out half, through data shares linked to data share targets: they are two mechanisms, not one switch. Note also that the directory marks many ordinary SaaS connectors Zero Copy, and those are not the same thing as warehouse grade federation. |
| Google BigQueryWarehouse and lakehouse | Zero copy federationZero Copy | Bidirectional | Live query federation. Acceleration caching is configurable from 15 minutes to 7 days, and incremental cache updates are upserts only. | One of the four platforms documented for query federation. Sharing back out is the separate data share path, not something federation gives you for free. |
| Amazon RedshiftWarehouse and lakehouse | Zero copy federationZero Copy | Bidirectional | Live query federation. Acceleration caching is configurable from 15 minutes to 7 days, and incremental cache updates are upserts only. | One of the four platforms documented for query federation, and marked bidirectional zero copy in the directory. |
| DatabricksWarehouse and lakehouse | Zero copy federationBatch, Zero Copy | Data in | Live query federation. Acceleration caching is configurable from 15 minutes to 7 days, and incremental cache updates are upserts only. | The directory lists Databricks as data in with both batch and zero copy, so what you get depends on how the stream is configured rather than on the platform alone. It is data in, unlike the three bidirectional warehouses above it. |
| Apache IcebergWarehouse and lakehouse | Zero copy federationZero Copy | Data in | Not documented | Listed as a zero copy connector. The documented file federation path names Amazon S3, Google Cloud Storage and Azure Data Lake Storage in Iceberg format, so treat Iceberg here as a table format that other platforms are read through rather than as a fifth live query platform. |
| AWS Glue Data CatalogWarehouse and lakehouse | Zero copy federationZero Copy | Data in | Not documented | Listed as a zero copy connector. No refresh or acceleration behaviour is documented for it in the sources this table cites, so plan a proof of concept rather than a design. |
| Microsoft Fabric OneLakeWarehouse and lakehouse | Zero copy federationZero Copy | Data in | Not documented | A recent addition to the zero copy set, so re-check it against the directory before you design around it. No refresh or acceleration behaviour is documented for it in the sources this table cites. |
| IBM watsonx.dataWarehouse and lakehouse | Zero copy federationZero Copy | Data in | Not documented | A recent addition to the zero copy set. No refresh or acceleration behaviour is documented for it in the sources this table cites. |
| Amazon S3Object storage and file | Native connectorBatch | Bidirectional | A predefined schedule, or manually from the data stream record home. | Carries both structured and unstructured data, and it is one of the three platforms named for file federation in Iceberg format. Federating a bucket and running a batch data stream against it are different paths with different costs, so decide which one you are on before you build. |
| Azure StorageObject storage and file | Native connectorBatch | Bidirectional | Not documented | Structured and unstructured, bidirectional and batch in the directory. Its refresh trigger is not documented alongside the Amazon S3 and Google Cloud Storage schedule, so do not assume it behaves the same way they do. Azure Data Lake Storage is named separately as a file federation target, in Iceberg format, which is a different path from a batch data stream. |
| Google Cloud StorageObject storage and file | Native connectorBatch | Bidirectional | A predefined schedule, or manually from the data stream record home. | Structured and unstructured, and one of the three platforms named for file federation in Iceberg format. Same choice as Amazon S3: federate the bucket or stream from it, not both by accident. |
| SFTP StructuredObject storage and file | Native connectorBatch | Bidirectional | Not documented | Structured SFTP is bidirectional batch. The unstructured SFTP connector is a separate row with a different direction and a different method, so check which of the two you are actually configuring. |
| SFTP UnstructuredObject storage and file | Native connectorPush | Data in | Not documented | Push rather than batch, and data in only. It is the one SFTP row that is not bidirectional, which catches people who have read about the structured connector and assumed the pair behave alike. |
| Amazon KinesisStreaming | Native connectorStreaming | Data in | Streaming. | A directory connector that reads a stream, not a transport you write to. If you are reaching for the Ingestion API to move Kinesis data, check this row first. |
| Amazon MSKStreaming | Native connectorStreaming | Data in | Streaming. | Same shape as the Kinesis connector: data in, streaming, read by Data 360 rather than pushed to it. |
| Website and Mobile AppStreaming | Native connector | Data in | Streaming, through the Salesforce Interactions SDK and the Engagement SDK. | Mobile events queue on the device, so a burst of replayed events after a period offline is normal rather than a fault. Wire consent capture at the same time as the SDK, because retrofitting it across a live event stream is the expensive version of this work. |
| Ingestion API, streamingCustom | Ingestion API | Data in | Asynchronous, processed roughly every 3 minutes. A 202 response means queued, not stored. | Payloads must use the schema field names, not the Salesforce API names: a payload with API names produces blank rows rather than an error, which is the slowest possible way to find the mistake. The body is capped at 200 KB per request and 250 requests per second across all object endpoints, and a delete call takes at most 200 ids. Allow at least 30 seconds after ingestion before the data is queryable.Not documented by Salesforce: Streaming records-per-second limit. |
| Ingestion API, bulkCustom | Ingestion API | Data in | Job based. Create a job, upload up to 100 CSV files one at a time, then close it to enqueue processing. | CSV only, UTF-8, comma delimited, and the header row must match the data source object field names exactly including case. Updates are a full replace rather than a patch, and empty cells become null. Files are capped at 150 MB, concurrency at 5, and open jobs older than 7 days are deleted from the queue. A job can finish with rejected rows, so read numberRecordsFailed rather than trusting the state.Not documented by Salesforce: Exact meaning of the bulk 20-per-hour limit. |
| MuleSoft DirectIntegration platform | MuleSoftBatch | Data in | Not documented | The directory lists the MuleSoft Direct connector as data in, unstructured and batch, which is narrower than the way MuleSoft is usually pitched for Data 360. Salesforce's wider connector count folds in Anypoint Exchange connectors, and that count is a marketing figure we could not verify against the directory itself. |
| Amazon AdsActivation | Native connectorIncremental, Direct API | Data out | Not documented | An activation target, not a source. The directory method is incremental plus direct API, so an activation sends what changed rather than the whole audience on every run. |
| Google AdsActivation | Native connectorIncremental, Direct API | Data out | Not documented | An activation target, not a source. The directory method is incremental plus direct API, so an activation sends what changed rather than the whole audience on every run. |
| Google DV360Activation | Native connectorIncremental, Direct API | Data out | Not documented | An activation target, not a source. The directory method is incremental plus direct API, so an activation sends what changed rather than the whole audience on every run. |
| LinkedInActivation | Native connectorIncremental, Direct API | Data out | Not documented | An activation target, not a source. The separate LinkedIn Conversions API connector is the streaming path and behaves differently, so check which one your use case needs. |
| MetaActivation | Native connectorIncremental, Direct API | Data out | Not documented | An activation target, not a source. The directory method is incremental plus direct API, so an activation sends what changed rather than the whole audience on every run. |
| PinterestActivation | Native connectorIncremental, Direct API | Data out | Not documented | An activation target, not a source. The directory method is incremental plus direct API, so an activation sends what changed rather than the whole audience on every run. |
| SnapchatActivation | Native connectorIncremental, Direct API | Data out | Not documented | An activation target, not a source. The directory method is incremental plus direct API, so an activation sends what changed rather than the whole audience on every run. |
| TikTokActivation | Native connectorIncremental, Direct API | Data out | Not documented | An activation target, not a source. The directory method is incremental plus direct API, so an activation sends what changed rather than the whole audience on every run. |
| LinkedIn Conversions APIActivation | Native connectorStreaming, Direct API | Data out | Streaming. | The streaming data out path, and a separate connector from the LinkedIn activation row above it. Streaming and direct API rather than incremental and direct API. |
What we could not source
List of derived fields that break CRM streaming eligibility
Tried both article id namespaces and web search; the page appears removed or renamed.
Streaming records-per-second limit
Any records-per-second number would be a derivation, not a documented limit; label it as such if shown.
Exact meaning of the bulk 20-per-hour limit
Quote the limit verbatim in the tools; do not restate as jobs per day or uploads per hour.
Sources
- Salesforce Connectors for Data Cloudmixed: title Data Cloud, body Data 360Read on .
- First-party Salesforce connectors
- Salesforce CRM ConnectorData 360Read on .
- Salesforce CRM connector direction
- Data Stream Schedule in Data 360Data 360Read on .
- Salesforce CRM connector refresh modes
- B2C Commerce refresh
- Marketing Cloud Engagement refresh
- Marketing Cloud Personalization refresh
- CRM Connector StreamingData 360Read on .
- Salesforce CRM connector streaming mode
- Data 360 Connectors and IntegrationsData 360Read on .
- Warehouse zero-copy connectors and direction
- Cloud storage and SFTP connectors
- Streaming source connectors
- MuleSoft's role
- Activation (data out) connectors
- Step-by-step Guide for Zero Copy and SharingData CloudRead on .
- Zero-copy federation partner set and direction semantics
- Data 360 Interoperability (architect decision guide)Data 360Read on .
- Federation modes and acceleration cache
- Streaming Ingestion WalkthroughData 360Read on .
- Streaming upsert endpoint
- Delete Records (Ingestion API reference)Data 360Read on .
- Streaming delete endpoint
- Get Started with Ingestion APIData 360Read on . Carries one open question, listed above.
- Streaming limits
- Eventual consistency window
- Bulk limits
- Exact meaning of the bulk 20-per-hour limit
- Bulk Ingestion WalkthroughData 360Read on .
- Bulk job endpoints
- Bulk Ingestion (Ingestion API reference)Data 360Read on .
- Bulk job lifecycle states
- Bulk CSV requirements
Salesforce ships three releases a year and its pricing artifacts move faster than that. Every fact on this page carries the date it was last checked against Salesforce documentation, and the sources are listed in full at the bottom.