DRIVERS-3416 test overload retry when retryReads/Writes=false

[kevinAlbs]

Summary

• Test overload retry when retryReads/retryWrites is disabled.
• Test overload retry of aggregate with a write stage.

Background & Motivation

Review of the C driver discovered operations were retried on overload errors when retryReads=false or retryWrites=false, contradicting this spec requirement:

A retry attempt will only be permitted if: [...] The command is a write and retryWrites is enabled or the command is a read and retryReads is enabled.

• To retry runCommand, both retryWrites and retryReads MUST be enabled.

A test case for aggregate with ($out / $merge) is added as a possible edge case. Unlike other aggregate read operations, Retryable Writes suggests it would be a write operation:

Write commands other than insert, update, delete, or findAndModify will not be initially supported by MongoDB 3.6, although this may change in the future. This includes, but is not limited to, an aggregate command using a write stage (e.g. $out, $merge).

Please complete the following before merging:

• Is the relevant DRIVERS ticket in the PR title?
• [ ] Update changelog. Test changes only.
• Test changes in at least one language driver. (Tested locally in C)
• Test these changes against all server versions and topologies (including standalone, replica set, and sharded
  clusters). (Not-yet done)

[kevinAlbs]

Changes have been extended to all tested operations. And a test for aggregate with a write-stage is added as an edge case.

@connorsmacd here is a rough draft of changes in the C driver to validate the tests: kevinAlbs/mongo-c-driver@eb46d92. This was intended as "just enough" to get tests passing, but that might help as a starting point.

[kevinAlbs]

fed5e1f intends to address this C driver test failure on sharded clusters:

sending distinct to each mongos
error: database backpressure-db not found

[NoahStapp]

Python implementation for reference.

[connorsmacd]

All tests are passing in the C Driver PR as of 4087eb0.

[connorsmacd]

LGTM aside from questions about the empty lines in the changelog.

[connorsmacd]

Is this empty line a placeholder for an eventual changelog update?

[kevinAlbs]

That was a drive-by fix to address failing lint tasks on master. That has since been fixed by #1908 so I will git merge master before merging.

[connorsmacd]

See: https://github.com/mongodb/specifications/pull/1909/changes#r2913545499