We use `trial.checkpoint` to restore a perturbed trial. Currently trial.checkpoint is looking at both in-memory and persistent checkpoints to find the most recent one. The definition of "the most recent one" is based on iteration. This may no longer be a valid assumption in PBT case, considering `trial_low_quantile` may have an iter=2_persistent_checkpoint as well as a iter=1_in_memory_checkpoint (perturbed from `trial_upper_quantile`).
This will start repro docker containers with SYS_PTRACE capabilities to enable debugging e.g. via py-spy.
Additionally, default instance name tags for instance re-use will be generated using the buildkite build id and job id.
This PR refactors several components to support switching to GCS address bootstrapping later:
- Treat address from `ray.init()` and `ray` CLI as bootstrap address instead of assuming it is Redis address.
- Ray client servers support `--address` flag instead of `--redis-address`.
- A few other miscellaneous cleanup.
Also, add a test for starting non-head node with `ray start`.
Inheriting from `abc.ABC` is more readable than setting the meta class to `abc.ABCMeta`.
Relevant snippet from the Python 3.4 release notes:
> New class ABC has ABCMeta as its meta class. Using ABC as a base class has essentially the same effect as specifying metaclass=abc.ABCMeta, but is simpler to type and easier to read. (Contributed by Bruno Dupuis in bpo-16049.)
Co-authored-by: Richard Liaw <rliaw@berkeley.edu>
Co-authored-by: Matthew Deng <matthew.j.deng@gmail.com>
If we use `os.environ` to set environment variables in tests, then our tests become coupled. By using `monkeypatch`, we can safely set environment variables while ensuring our tests remain decoupled.
For more information, see the [monkeypatching documentation](https://docs.pytest.org/en/6.2.x/monkeypatch.html#monkeypatching-environment-variables).
Expands the `to_torch` method for Datasets with:
* An ability to choose to output a list/dict of feature tensors instead of just one (through setting `feature_columns` to be a list of lists or a dict of lists)
* An ability to choose whether the label should be unsqueezed or not
* An ability to pass `None` as the label (for prediction).
Furthermore, this changes how the `feature_column_dtypes` argument works. Previously, it took a list of dtypes for each feature. However, as the tensor was concatenated in the end, only one dtype mattered (the biggest one). Now, this argument expects a single dtype which will be applied to the features tensor (or a list/dict if `feature_columns` is a list of list/dict of lists).
Unit tests for all cases are included.
Co-authored-by: matthewdeng <matthew.j.deng@gmail.com>
When a list with mixed types is passed to tune.choice, they will be coerced to a single dtype during sampling (due to numpy.choice converting to an array internally). This behaviour is unintentional and surprising. This PR fixes this issue.
These changes are needed for two reasons.
**`__grouping_doc_end__` is in the wrong place**
If you look at the part of the Ray documentation where the tag is referenced, you'll read
> You can use the MultiAgentEnv.with_agent_groups() method to define these groups:
However, if you look at the code snippet below, you'll see the implementation of `to_base_env` in addition to the implementation of `with_agent_groups`.
To remove `to_base_env` from the code snippet, we need to move `__grouping_doc__end__`.
**Black cannot format `multi_agent_env.py`**
For some reason, Black errors while formatting `multi_agent_env.py`. However, if we move `__grouping_doc_end__` up, the issue is resolved.
Create an AWS instance to reproduce Buildkite CI builds.
This script will take a Buildkite build URL as an argument and create
an AWS instance with the same properties running the same Docker container
as the original Buildkite runner. The user is then attached to this instance
and can reproduce any builds commands as if they were executed within the
runner.
This utility can be used to reproduce and debug build failures that come up
on the Buildkite runner instances but not on a local machine.
There's a redis connection in gcs client, but most time the gcs client is never used in worker. We can make the initialization lazy to reduce redis connections.
After that, the number of redis connections reduces from 2 to 1 in one core worker.
This PR povided universal native memory access support in java worker mentioned in #21234, which will also be the foundation for later zero-copy and serialization.
The main changes include:
* Native memory operations based on `sun.misc.Unsafe`
* Little-Endian based Native memory buffer.
* Native memory based IO operations:
* InputStream/OutputStream
* ReadChannel/WriteChannel
* MockReadChannel/MockWriteChannel
This PR contains most of the fixes @iycheng made in #21232, to make tests pass with GCS bootstrapping by supporting both Redis and GCS address as the bootstrap address. The main change is to use address_info["address"] to obtain the bootstrap address to pass to ray.init(), instead of using address_info["redis_address"]. In a subsequent PR, address_info["address"] will return the Redis or GCS address depending on whether using GCS to bootstrap.
It seems that the `StealTasks` rpc has no different from other common rpc methods, should be implemented by `VOID_RPC_CLIENT_METHOD` macro. We find this when merge code into our internal codebase.
Please review **e2e.py and test_suite belonging to your team**!
This is the first part of https://docs.google.com/document/d/16IrwerYi2oJugnRf5hvzukgpJ6FAVEpB6stH_CiNMjY/edit#
This PR adds a team name to each test suite.
If the name is not specified, it will be reported as unspecified.
If you are running a local test, and if the new test suite doesn't have a team name specified, it will raise an exception (in this way, we can avoid missing team names in the future).
Note that we will aggregate all of test config into a single file, nightly_test.yaml.
This PR is introducing a canonical impl for stopping trials by collecting scattered logic from process_trial_result back into stop_trial. This way, we know what is expected (e.g. what callbacks are invoked and when they are invoked).
This PR will correct the current wrong logic that on_trial_complete callback is invoked before on_trial_checkpoint, which is the source of Syncer clean up issues.
