This is on lines of Jira ticket i have submitted for kafka sink connector ( 

KAFKA-380 Mongo Sink Connector Enhancement Request - MongoDB Jira), this is a separate request for a similar enhancement for Kafka source connector

Mongo Source Connector Enhancement Request:-

                When the target Mongo DB is unavailable, the mongo sink connector with “error.tolerance = all” writes all the data to dead letter topic.

                The purpose of setting “error.tolerance = all” is to ensure continuous processing of data while at the same time messages leading to data exceptions are moved to dead letter topic.

                                Ex:- duplicate / malformed messages.

This enhancement request is expected to resolve the current limitation by providing finer granularity of exception handling and to differentiate between retriable (aka transient) and non-retriable (aka non-transient) errors, and process them accordingly.

Examples for the transient errors are mongo DB unavailable, network/connection issue & typical timeout errors.

Examples for the non-transient errors are data violation errors thrown by Mongo DB, like duplicate key/primary key violations etc.,

While encountering transient errors, the connector has to provide multiple retry attempts with time interval between each attempt.

While encountering non-transient errors, the connector should continue to put the message into dead letter topic, and start to process the next message.