* Remove all __future__ imports from RLlib.
* Remove (object) again from tf_run_builder.py::TFRunBuilder.
* Fix 2xLINT warnings.
* Fix broken appo_policy import (must be appo_tf_policy)
* Remove future imports from all other ray files (not just RLlib).
* Remove future imports from all other ray files (not just RLlib).
* Remove future import blocks that contain `unicode_literals` as well.
Revert appo_tf_policy.py to appo_policy.py (belongs to another PR).
* Add two empty lines before Schedule class.
* Put back __future__ imports into determine_tests_to_run.py. Fails otherwise on a py2/print related error.
* move more unit tests to bazel
* move to avoid conflict
* fix lint
* fix deps
* seprate
* fix failing tests
* show tests
* ignore mismatch
* try combining bazel runs
* build lint
* remove tests from install
* fix test utils
* better config
* split up
* exclusive
* fix verbosity
* fix tests class
* cleanup
* remove flaky
* fix metrics test
* Update .travis.yml
* no retry flaky
* split up actor
* split basic test
* split up trial runner test
* split stress
* fix basic test
* fix tests
* switch to pytest runner for main
* make microbench not fail
* move load code to py3
* test is no longer package
* bazel to end
Running `./ci/long_running_tests/start_workloads.sh` will start several workloads running (each in their own EC2 instance).
- The workloads run forever.
- The workloads all simulate multiple nodes but use a single machine.
- You can get the tail of each workload by running `./ci/long_running_tests/check_workloads.sh`.
- You have to manually shut down the instances.
As discussed with @ericl @richardliaw, the idea here is to optimize for the debuggability of the tests. If one of them fails, you can ssh to the relevant instance and see all of the logs.
Simon Mo