[GOBBLIN-1025] Add retry for PK-Chunking iterator

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.15.0
• Assignee: Unassigned
• Reporter: Alex Li
• Created: 14/Jan/20 19:07
• Updated: 30/Jan/20 01:30
• Resolved: 30/Jan/20 01:30

Description

In SFDC connector, there is a class called `ResultIterator` (I will change the name to SalesforceRecordIterator).
It was using by only PK-Chunking currently. It encapsulated fetching a list of result files to a record iterator.

However, the csvReader.nextRecord() may throw out network IO exception. We should do retry in this case.

When a result file is fetched partly and one network IO exception happens, we are in a special situation - first half of the file is already fetched to our local, but another half of the file is still on datasource. 
We need to
1. reopen the file stream
2. skip all the records that we already fetched, seek the cursor to the record which we haven't fetched yet.

Issue Links

• links to: GitHub Pull Request #2868