Connect Cdc Salesforce

Redpanda Connect CDC: Salesforce

The salesforce_cdc input in Redpanda Connect streams change data capture (CDC) events and Platform Events from Salesforce into Redpanda or any Kafka-compatible topic. It uses the Salesforce Pub/Sub gRPC API (api.pubsub.salesforce.com:443) for real-time streaming and the Salesforce REST API for optional initial snapshots of existing sObject records. Authentication uses the OAuth 2.0 Client Credentials flow via a Salesforce Connected App.

This is an Enterprise feature — a Redpanda Enterprise license is required at runtime. The connector manages per-topic replay-ID state in a durable cache resource (Redis, PostgreSQL, DynamoDB, etc.) so pipelines survive restarts and resume from where they left off.

Quickstart

1. Salesforce prerequisites (5 steps)

# Install the Salesforce CLI (macOS example)
brew install salesforce-cli

# Log in to your org (opens a browser)
sf org login web --set-default

In Salesforce Setup, enable Change Data Capture for the objects you want to capture:

Setup → Integrations → Change Data Capture → select Account, Contact, Opportunity → Save

Create a Connected App for OAuth Client Credentials:

Setup → App Manager → New Connected App
  → Enable OAuth Settings
  → Selected OAuth Scopes: api, full, RefreshToken
  → Enable Client Credentials Flow
  → Save → Manage Consumer Details → copy Consumer Key + Consumer Secret

Assign the Run-As user to the Connected App:

Setup → App Manager → find app → Manage → Edit Policies
  → Client Credentials Flow: set Run As User → Save

Store credentials as environment variables:

export SALESFORCE_ORG_URL="https://acme.my.salesforce.com"
export SALESFORCE_CLIENT_ID="3MVG9..."          # Consumer Key
export SALESFORCE_CLIENT_SECRET="abc123..."     # Consumer Secret

2. Full pipeline YAML (snapshot + CDC, three core objects)

# salesforce-cdc-pipeline.yaml
input:
  label: "sf_cdc"
  salesforce_cdc:
    org_url: ${SALESFORCE_ORG_URL}
    client_id: ${SALESFORCE_CLIENT_ID}
    client_secret: ${SALESFORCE_CLIENT_SECRET}
    topics:
      - Account       # shorthand for /data/AccountChangeEvent
      - Contact       # shorthand for /data/ContactChangeEvent
      - Opportunity   # shorthand for /data/OpportunityChangeEvent
    stream_snapshot: true          # back-fill existing records first
    replay_preset: latest          # for live changes after snapshot; use earliest to recover missed events
    snapshot_max_batch_size: 2000  # records per REST query page (200–2000)
    stream_batch_size: 100         # events per gRPC Fetch call
    max_parallel_snapshot_objects: 1
    checkpoint_cache: persistent_cache
    checkpoint_cache_key: salesforce_cdc  # default
    checkpoint_limit: 1024
    batching:
      count: 100
      period: 1s

pipeline:
  processors:
    # Route each event to a topic named after the sObject
    - mapping: |
        meta kafka_topic = "sf.cdc." + metadata("sobject").lowercase()

output:
  kafka_franz:
    seed_brokers:
      - localhost:9092
    topic: ${! metadata("kafka_topic") }

cache_resources:
  - label: persistent_cache
    redis:
      url: redis://localhost:6379

3. Run the pipeline

# Via rpk
rpk connect run salesforce-cdc-pipeline.yaml

# Via the redpanda-connect binary
redpanda-connect run salesforce-cdc-pipeline.yaml

# Via Docker
docker run --rm \
  -e SALESFORCE_ORG_URL \
  -e SALESFORCE_CLIENT_ID \
  -e SALESFORCE_CLIENT_SECRET \
  -v $(pwd)/salesforce-cdc-pipeline.yaml:/pipeline.yaml \
  docker.redpanda.com/redpandadata/connect:latest \
  run /pipeline.yaml

4. Inspect emitted messages

Every message has these metadata fields accessible via metadata("key") in Bloblang:

Snapshot rows (operation: read) carry exactly sobject and operation — they have no topic, replay_id, record_ids, or event_uuid.

Example snapshot payload (operation=read):

{
  "Id": "001Dp000008KFIXIA4",
  "Name": "Acme Corp",
  "BillingCity": "San Francisco",
  "CreatedDate": "2024-01-15T10:30:00.000+0000"
}

Example CDC event payload (operation=create):

{
  "ChangeEventHeader": {
    "entityName": "Account",
    "recordIds": ["001Dp000008KFIXIA4"],
    "changeType": "CREATE",
    "changeOrigin": "com/salesforce/api/rest/65.0"
  },
  "Name": "Acme Corp",
  "BillingCity": "San Francisco"
}

Topic Syntax

The topics list accepts four forms. Each entry maps to exactly one gRPC Pub/Sub subscription with its own independent replay cursor:

The firehose (/data/ChangeEvents) and per-sObject channels are mutually exclusive — do not mix them. Platform Event topics are always skipped during the REST snapshot phase (no REST equivalent exists).

Snapshot + Stream Lifecycle

When stream_snapshot: true:

1. The connector opens the checkpoint cache and checks for a persisted snapshot_complete flag.
2. If not complete, it pages through all sObjects in topics via the Salesforce REST Query API, emitting rows with operation: read.
3. Snapshot cursor is checkpointed after each page — restarts resume mid-snapshot.
4. When all pages are exhausted, snapshot_complete: true is written to the cache.
5. The gRPC Pub/Sub subscriptions are then opened (one per topic) and stream live events.

Platform Event topics are always skipped in the snapshot phase. The firehose (/data/ChangeEvents) triggers snapshotting of all queryable sObjects.

Checkpoint Cache and Restart Semantics

The connector persists a JSON document to the cache under checkpoint_cache_key (default salesforce_cdc):

{
  "snapshot_complete": true,
  "rest_cursor": {},
  "topics": {
    "/data/AccountChangeEvent": "<hex-replay-id>",
    "/data/ContactChangeEvent": "<hex-replay-id>"
  }
}

On restart:

• If snapshot_complete is false and a rest_cursor is present, the snapshot resumes mid-page.
• Each topic's replay_id is used to resume that subscription from the exact event where it left off.
• If a replay_id is rejected by Salesforce (gRPC INVALID_ARGUMENT — stale beyond retention), it is cleared from state and the subscription restarts using replay_preset.

Use a durable cache backend (Redis, PostgreSQL, DynamoDB) for production. An in-memory cache loses all state on restart.

Operational Notes

• Event retention: Standard retention is 24 hours; Enhanced Event Retention (Salesforce add-on) extends to 72 hours. After the retention window, replaying from replay_preset: earliest starts at the oldest available event.
• API quota: The REST snapshot uses Salesforce API call quota. Each page of snapshot_max_batch_size records is one API call. Large orgs with many records can consume significant quota during the initial snapshot.
• Connected App permissions: The Run-As user must have read access to all sObjects being captured and Pub/Sub API access (requires API-enabled profile or permission set).
• License errors: salesforce_cdc is an Enterprise-gated component. A startup error stating "this feature requires a valid Redpanda Enterprise Edition license that includes the Connect product" means the binary is not licensed. See https://docs.redpanda.com/redpanda-connect/get-started/licensing/ to obtain and configure an Enterprise license.
• Stale replay recovery: When replay_id is rejected (event outside retention window), the connector automatically clears the stale ID and reconnects using replay_preset. Log line: topic /data/... replay_id rejected; clearing and reconnecting via configured preset.
• Schema and decode failures are bounded (Connect 4.106.0+): grpc.reconnect_max_attempts also governs transient schema-fetch failures — an event whose schema fetch keeps failing at the same replay position is retried that many times before the topic fails permanently. Deterministic failures (a schema that is missing, inaccessible, or fails to compile) and payloads that repeatedly fail to decode give up after a small fixed number of attempts regardless of that setting, since retrying cannot change the outcome. Terminal failures surface clearly on the health check instead of livelocking. Note the interaction with the default reconnect_max_attempts: 0 (unlimited): before 4.106.0, a transient schema fetch under that default could stall indefinitely.
• A terminal failure is never mistaken for a stale replay ID: an undecodable payload or unfetchable schema is not treated as a stale replay_id even when it arrives wrapped in a gRPC INVALID_ARGUMENT — that status describes the schema, and clearing the checkpoint would resubscribe from replay_preset and silently skip everything between the checkpoint and now.
• Buffer backpressure: a full grpc.buffer_size buffer applies backpressure rather than dropping events (4.106.0+).
• Multiple pipelines: Run multiple salesforce_cdc inputs against the same cache by setting a distinct checkpoint_cache_key per input.

Enterprise Features (License Required)

The salesforce_cdc input is a Redpanda Connect enterprise connector — it requires a valid Enterprise license that includes the Connect product. Beyond the connector, the Redpanda topics that receive the CDC stream can use broker-side enterprise features that require a Redpanda Enterprise license on the destination cluster:

• Iceberg Topics — materialize the CDC topic as an Apache Iceberg table for analytics (redpanda.iceberg.mode = key_value | value_schema_id_prefix | value_schema_latest; plus redpanda.iceberg.delete, redpanda.iceberg.invalid.record.action, redpanda.iceberg.partition.spec, redpanda.iceberg.target.lag.ms). Requires iceberg_enabled=true and Tiered Storage on the topic.
• Tiered Storage — long-term retention of the change stream past Salesforce's 24h/72h window (cloud_storage_enabled cluster-wide; redpanda.remote.write + redpanda.remote.read per topic; retention.local.target.ms/.bytes).
• Cloud Topics — object-storage-native ("diskless") destination topics for cost-sensitive CDC retention (cloud_topics_enabled=true cluster-wide, requires restart; create the topic with redpanda.storage.mode=cloud, or default_redpanda_storage_mode=cloud to make it the cluster default). Cloud Topic mode can be set only at topic creation time.
• Server-side Schema ID Validation — enforce registered schemas on the CDC topic (enable_schema_id_validation = redpanda|compat; redpanda.value.schema.id.validation, redpanda.value.subject.name.strategy).
• RBAC, Audit Logging, OIDC/OAUTHBEARER, Kerberos (GSSAPI), FIPS, Shadowing — cluster-wide controls that govern who can produce to and operate the CDC topics.

Check license status with rpk cluster license info. See redpanda-enterprise-sink.md for exact config keys, defaults, enable/disable commands, and license-expiry behavior.

Reference Directory

• config-reference.md: Every salesforce_cdc config field — type, default, required status, and description grounded in source.
• setup-salesforce.md: Creating a Salesforce Connected App, enabling Change Data Capture for sObjects, defining Platform Events, Pub/Sub API access, and the replay-ID/retention model.
• pipeline-and-output.md: Full runnable pipelines (CDC, firehose, Platform Events, mixed), the message/metadata shape, per-sObject topic routing, snapshot+stream behavior, and resume semantics.
• redpanda-enterprise-sink.md: Redpanda Enterprise features for the destination CDC topics — the connector's own Connect license requirement, Iceberg Topics (redpanda.iceberg.*), Tiered Storage (redpanda.remote.read/write, retention), Cloud Topics (cloud_topics_enabled, redpanda.storage.mode=cloud), Server-side Schema ID Validation, and pointers to RBAC, Audit Logging, OIDC/Kerberos, FIPS, and Shadowing, with exact config keys, defaults, enable/disable commands, and license-expiry behavior.