When the underlying Hadoop infrastructure is slow, CDAP often fails with timeout. For example, transactions time out, HTTP requests, Thrift connections, HBase clients, and explore requests time out. Most of these timeouts are configurable, but it is not well documented:
- what configuration affects what calls
- how the different timeouts relate to each other (e.g., the tx timeout must be greater than the thrift client timeout)
All timeouts must be made configurable, and complete documentation needs to be added.