Databricks Streaming Loader configuration reference

The configuration reference in this page is written for Databricks Streaming Loader 0.4.0

License

The Databricks Streaming Loader is released under the Snowplow Limited Use License (FAQ).

To accept the terms of license and run the loader, set the ACCEPT_LIMITED_USE_LICENSE=yes environment variable. Alternatively, configure the license.accept option in the config file:

json

"license": {
  "accept": true
}

Databricks configuration

Parameter	Description
output.good.host	Required, e.g. https://{workspace-id}.cloud.databricks.com. URL of the Databricks workspace.
output.good.catalog	Required. Name of the Databricks catalog containing the volume.
output.good.schema	Required. Name of the Databricks schema containing the volume.
output.good.volume	Required. Name of the Databricks volume to which this loader will upload staging files. This must be an external volume.
output.good.token	Required if using PAT authentication. A Databricks personal access token.
output.good.oauth.clientId	Required if using OAUTH authentication. The client ID for a Databricks service principal.
output.good.oauth.clientSecret	Required if using OAUTH authentication. The client secret for a Databricks service principal.
output.good.compression	Optional. Default value snappy. Compression algorithm for the uploaded staging parquet files.
output.good.httpTimeout	Optional. Default value 20 seconds. Timeout duration of Databricks SDK's HTTP client.

Streams configuration

AWS

Parameter	Description
input.streamName	Required. Name of the Kinesis stream with the enriched events
input.appName	Optional, default snowplow-databricks-loader. Name to use for the dynamodb table, used by the underlying Kinesis Consumer Library for managing leases.
input.initialPosition.type	Optional, default LATEST. Allowed values are LATEST, TRIM_HORIZON, AT_TIMESTAMP. When the loader is deployed for the first time, this controls from where in the kinesis stream it should start consuming events. On all subsequent deployments of the loader, the loader will resume from the offsets stored in the DynamoDB table.
input.initialPosition.timestamp	Required if input.initialPosition is AT_TIMESTAMP. A timestamp in ISO8601 format from where the loader should start consuming events.
input.retrievalMode	Optional, default Polling. Change to FanOut to enable the enhance fan-out feature of Kinesis.
input.retrievalMode.maxRecords	Optional. Default value 1000. How many events the Kinesis client may fetch in a single poll. Only used when input.retrievalMode is Polling.
input.workerIdentifier	Optional. Defaults to the HOSTNAME environment variable. The name of this KCL worker used in the dynamodb lease table.
input.leaseDuration	Optional. Default value 10 seconds. The duration of shard leases. KCL workers must periodically refresh leases in the dynamodb table before this duration expires.
input.maxLeasesToStealAtOneTimeFactor	Optional. Default value 2.0. Controls how to pick the max number of shard leases to steal at one time. E.g. If there are 4 available processors, and maxLeasesToStealAtOneTimeFactor = 2.0, then allow the loader to steal up to 8 leases. Allows bigger instances to more quickly acquire the shard-leases they need to combat latency.
input.checkpointThrottledBackoffPolicy.minBackoff	Optional. Default value 100 milliseconds. Initial backoff used to retry checkpointing if we exceed the DynamoDB provisioned write limits.
input.checkpointThrottledBackoffPolicy.maxBackoff	Optional. Default value 1 second. Maximum backoff used to retry checkpointing if we exceed the DynamoDB provisioned write limits.
output.bad.streamName	Required. Name of the Kinesis stream that will receive failed events.
output.bad.throttledBackoffPolicy.minBackoff	Optional. Default value 100 milliseconds. Initial backoff used to retry sending failed events if we exceed the Kinesis write throughput limits.
output.bad.throttledBackoffPolicy.maxBackoff	Optional. Default value 1 second. Maximum backoff used to retry sending failed events if we exceed the Kinesis write throughput limits.
output.bad.recordLimit	Optional. Default value 500. The maximum number of records we are allowed to send to Kinesis in 1 PutRecords request.
output.bad.byteLimit	Optional. Default value 5242880. The maximum number of bytes we are allowed to send to Kinesis in 1 PutRecords request.
output.bad.maxRecordSize	Optional. Default value 1000000. Any single event failed event sent to Kinesis should not exceed this size in bytes
output.bad.maxRetries (since 0.4.0)	Optional. Default value 10. Maximum number of retries by Kinesis Client.

GCP

Parameter	Description
input.subscription	Required, e.g. projects/myproject/subscriptions/snowplow-enriched. Name of the Pub/Sub subscription with the enriched events
input.durationPerAckExtension	Optional. Default value 15 seconds. Pub/Sub ack deadlines are extended for this duration when needed.
input.minRemainingAckDeadline	Optional. Default value 0.1. Controls when ack deadlines are re-extended, for a message that is close to exceeding its ack deadline. For example, if durationPerAckExtension is 60 seconds and minRemainingAckDeadline is 0.1 then the loader will wait until there is 6 seconds left of the remining deadline, before re-extending the message deadline.
input.maxMessagesPerPull	Optional. Default value 1000. How many Pub/Sub messages to pull from the server in a single request.
input.debounceRequests	Optional. Default value 100 millis. Adds an artifical delay between consecutive requests to Pub/Sub for more messages. Under some circumstances, this was found to slightly alleviate a problem in which Pub/Sub might re-deliver the same messages multiple times.
input.retries.transientErrors.delay	Optional. Default value 100 millis. Backoff delay for follow-up attempts
input.retries.transientErrors.attempts	Optional. Default value 10. Max number of attempts, after which Enrich will crash and exit
output.bad.topic	Required, e.g. projects/myproject/topics/snowplow-bad. Name of the Pub/Sub topic that will receive failed events.
output.bad.batchSize	Optional. Default value 1000. Bad events are sent to Pub/Sub in batches not exceeding this count.
output.bad.requestByteThreshold	Optional. Default value 1000000. Bad events are sent to Pub/Sub in batches with a total size not exceeding this byte threshold
output.bad.maxRecordSize	Optional. Default value 9000000. Any single failed event sent to Pub/Sub should not exceed this size in bytes
output.retries.transientErrors.delay	Optional. Default value 100 millis. Backoff delay for follow-up attempts
output.retries.transientErrors.attempts	Optional. Default value 10. Max number of attempts, after which Enrich will crash and exit

Azure

Parameter	Description
input.topicName	Required. Name of the Kafka topic for the source of enriched events.
input.bootstrapServers	Required. Hostname and port of Kafka bootstrap servers hosting the source of enriched events.
input.consumerConf.*	Optional. A map of key/value pairs for any standard Kafka consumer configuration option.
input.debounceCommitOffsets	Optional. Default value 10 seconds. How frequently to commit our progress back to kafka. By increasing this value, we decrease the number of requests made to the kafka broker.
input.commitTimeout (since 0.4.0)	Optional. Default value 15 seconds. The time to wait for offset commits to complete. If an offset commit doesn't complete within this time, a CommitTimeoutException will be raised instead.
output.bad.topicName	Required. Name of the Kafka topic that will receive failed events.
output.bad.bootstrapServers	Required. Hostname and port of Kafka bootstrap servers hosting the bad topic
output.bad.producerConf.*	Optional. A map of key/value pairs for any standard Kafka producer configuration option.
output.bad.maxRecordSize	Optional. Default value 1000000. Any single failed event sent to Kafka should not exceed this size in bytes

Event Hubs Authentication

You can use the input.consumerConf and output.bad.producerConf options to configure authentication to Azure event hubs using SASL. For example:

json

"input.consumerConf": {
    "security.protocol": "SASL_SSL"
    "sasl.mechanism": "PLAIN"
    "sasl.jaas.config": "org.apache.kafka.common.security.plain.PlainLoginModule required username=\"\$ConnectionString\" password=<PASSWORD>;"
}

Other configuration options

Parameter	Description
batching.maxBytes	Optional. Default value 16000000. Events are uploaded to the Databricks volume when the batch reaches this size in bytes
batching.maxDelay	Optional. Default value 1 second. Events are uploaded to the Databricks volume after a maximum of this duration, even if the maxBytes size has not been reached
batching.uploadParallelismFactor	Optional. Default value 3. Controls how many batches can we send simultaneously over the network to Databricks. E.g. If there are 4 available processors, and uploadParallelismFactor is 3.5, then the loader sends up to 14 batches in parallel. Adjusting this value can cause the app to use more or less of the available CPU.
cpuParallelismFactor	Optional. Default value 0.75. Controls how the app splits the workload into concurrent batches which can be run in parallel. E.g. If there are 4 available processors, and cpuParallelismFactor is 0.75, then the loader processes 3 batches concurrently. Adjusting this value can cause the app to use more or less of the available CPU.
retries.setupErrors.delay	Optional. Default value 30 seconds. Configures exponential backoff on errors related to how Databricks is set up for this loader. Examples include authentication errors and permissions errors. This class of errors are reported periodically to the monitoring webhook.
retries.transientErrors.delay	Optional. Default value 1 second. Configures exponential backoff on errors that are likely to be transient. Examples include server errors and network errors.
retries.transientErrors.attempts	Optional. Default value 5. Maximum number of attempts to make before giving up on a transient error.
skipSchemas	Optional, e.g. ["iglu:com.example/skipped1/jsonschema/1-0-0"] or with wildcards ["iglu:com.example/skipped2/jsonschema/1--"]. A list of schemas that won't be loaded to Databricks. This feature could be helpful when recovering from edge-case schemas which for some reason cannot be loaded to the table.
exitOnMissingIgluSchema	Optional. Default value true. Whether the loader should crash and exit if it fails to resolve an Iglu Schema. We recommend true because Snowplow enriched events have already passed validation, so a missing schema normally indicates an error that needs addressing. Change to false so events go the failed events stream instead of crashing the loader.
monitoring.metrics.statsd.hostname	Optional. If set, the loader sends statsd metrics over UDP to a server on this host name.
monitoring.metrics.statsd.port	Optional. Default value 8125. If the statsd server is configured, this UDP port is used for sending metrics.
monitoring.metrics.statsd.tags.*	Optional. A map of key/value pairs to be sent along with the statsd metric.
monitoring.metrics.statsd.period	Optional. Default 1 minute. How often to report metrics to statsd.
monitoring.metrics.statsd.prefix	Optional. Default snowplow.databricks-loader. Prefix used for the metric name when sending to statsd.
monitoring.webhook.endpoint	Optional, e.g. https://webhook.example.com. The loader will send to the webhook a payload containing details of any error related to how Databricks is set up for this loader.
monitoring.webhook.tags.*	Optional. A map of key/value strings to be included in the payload content sent to the webhook.
monitoring.webhook.heartbeat.*	Optional. Default value 5.minutes. How often to send a heartbeat event to the webhook when healthy.
monitoring.sentry.dsn	Optional. Set to a Sentry URI to report unexpected runtime exceptions.
monitoring.sentry.tags.*	Optional. A map of key/value strings which are passed as tags when reporting exceptions to Sentry.
telemetry.disable	Optional. Set to true to disable telemetry.
telemetry.userProvidedId	Optional. See here for more information.
http.client.maxConnectionsPerServer	Optional. Default value 4. Configures the internal HTTP client used for iglu resolver, alerts and telemetry. The maximum number of open HTTP requests to any single server at any one time.