diff --git a/docs/release-notes.txt b/docs/release-notes.txt index a8741462e6f..c635cd11824 100644 --- a/docs/release-notes.txt +++ b/docs/release-notes.txt @@ -8,6 +8,57 @@ Version 0.18 ************** +Version 0.18.3 +============== + +**Release Date:** July 07, 2022 + +**Breaking Changes** + +- WebUI: Remove previously unlisted cluster page. This page has been replaced by a new version + available through the navigation bar. + +**New Features** + +- Workspaces & Projects: Teams can now organize related experiments into projects and workspaces. + See `video `_ for a walkthrough. + +- Logging: Master configuration now supports ``logging.additional_fluent_outputs`` allowing + advanced users to specify custom integrations for task logs. + +- Kubernetes: Task init containers no longer require root privileges. + +- API: Trial API now uploads profiling data to the checkpoint storage from all workers. Core API + users can now pass a new optional argument, ``tensorboard_mode``, to ``core.init()``. The default + value is ``AUTO``. In ``AUTO`` mode, TensorBoard metrics are written on the chief, and metrics as + well as profiling data are uploaded to checkpoint storage from the chief only. In ``MANUAL`` + mode, the user is responsible for writing TensorBoard metrics and uploading profiling data. In + order to make that possible, two new methods are introduced on + :class:`~determined.core.TrainContext`: + :meth:`~determined.core.TrainContext.get_tensorboard_path()` returns the path to the directory + where metrics can be written and :meth:`~determined.core.TrainContext.upload_tensorboard_files()` + uploads metrics and other files, such as profiling data, to checkpoint storage. + +- Add support for recovering live commands, notebooks, TensorBoards, and shells on master restart. + This is an extension of live trial recovery, available since version 0.18.1. + +**Bug Fixes** + +- WebUI: Fix a bug where a previous resource pool selection would not update when a new resource + pool is selected for viewing associated jobs. +- API: Fix a bug where ``/api/v1/tasks/{taskId}`` would often return incorrect allocation states. +- Since 0.17.15, there was a bug where ``task_container_defaults.registry_auth`` was not correctly + passed to tasks, resulting in tasks being unable to pull images. + +**Improvements** + +- CLI: Add new flag ``--agent-config-path`` to ``det deploy local agent-up``` allowing custom agent + configs to be used. +- CLI: Add ``det (notebook|shell|tensorboard) list --json`` option, allowing user to get + JSON-formatted notebook, shell or tensorboard task list. +- Configuration: Experiment configuration ``resources.shm_size`` now supports passing in a unit + like ``4.5 G`` or ``128MiB``. + Version 0.18.2 ============== diff --git a/docs/release-notes/4142-tensorboard-profiler-data-from-workers.txt b/docs/release-notes/4142-tensorboard-profiler-data-from-workers.txt deleted file mode 100644 index 61e77f1671d..00000000000 --- a/docs/release-notes/4142-tensorboard-profiler-data-from-workers.txt +++ /dev/null @@ -1,14 +0,0 @@ -:orphan: - -**New Features** - -- API: - Trial API now uploads profiling data to the checkpoint storage from all workers. Core API - users can now pass a new optional argument, `tensorboard_mode`, to `core.init()`. The default - value is `AUTO`. In `AUTO` mode tensorboard metrics are written on the chief, and metrics as - well as profiling data are uploaded to checkpoint storage from the chief only. In `MANUAL` - mode the user is responsible for writing tensorboard metrics and uploading profiling data. In - order to make that possible, two new methods are introduced on `TrainContext`: - `get_tensorboard_path()` returns the path to the directory where metrics can be written - `upload_tensorboard_files()` uploads metrics and other files, such as profiling data, to - checkpoint storage. diff --git a/docs/release-notes/4391-cluster-overview.txt b/docs/release-notes/4391-cluster-overview.txt deleted file mode 100644 index cfc7fe7febd..00000000000 --- a/docs/release-notes/4391-cluster-overview.txt +++ /dev/null @@ -1,11 +0,0 @@ -:orphan: - -**Breaking Changes** - -- WebUI: Remove previously unlisted cluster page. This page has been replaced by a new version - available through the navigation bar. - -**Bug Fixes** - -- WebUI: Fix a bug where a previous resource pool selection would not update when a new - resource pool is selected for viewing associated jobs. diff --git a/docs/release-notes/4412-registry-auth.txt b/docs/release-notes/4412-registry-auth.txt deleted file mode 100644 index 8d0fbc865b3..00000000000 --- a/docs/release-notes/4412-registry-auth.txt +++ /dev/null @@ -1,6 +0,0 @@ -:orphan: - -**Fixes** - -- Since 0.17.15, there was a bug where ``task_container_defaults.registry_auth`` was not correctly - passed to tasks, resulting in tasks being unable to pull images. diff --git a/docs/release-notes/additional_fluent_outputs.txt b/docs/release-notes/additional_fluent_outputs.txt deleted file mode 100644 index ff29ad0a9ea..00000000000 --- a/docs/release-notes/additional_fluent_outputs.txt +++ /dev/null @@ -1,6 +0,0 @@ -:orphan: - -**New Features** - -- Logging: Master configuration now supports ```logging.additional_fluent_outputs``` allowing - advanced users to specify custom integrations for task logs. diff --git a/docs/release-notes/agent-config-path.txt b/docs/release-notes/agent-config-path.txt deleted file mode 100644 index 36af32534b5..00000000000 --- a/docs/release-notes/agent-config-path.txt +++ /dev/null @@ -1,7 +0,0 @@ -:orphan: - -**Improvements** - -- CLI: - Add new flag ``--agent-config-path`` to ``det deploy local agent-up``` allowing custom agent - configs to be used. diff --git a/docs/release-notes/non-root-init-k8s.txt b/docs/release-notes/non-root-init-k8s.txt deleted file mode 100644 index 815e6ff6f82..00000000000 --- a/docs/release-notes/non-root-init-k8s.txt +++ /dev/null @@ -1,5 +0,0 @@ -:orphan: - -**New Features** - -- Kubernetes: Task init containers no longer require root privileges. diff --git a/docs/release-notes/rolling-commands.txt b/docs/release-notes/rolling-commands.txt deleted file mode 100644 index 73d5e5cb5d5..00000000000 --- a/docs/release-notes/rolling-commands.txt +++ /dev/null @@ -1,13 +0,0 @@ -:orphan: - -**New Features** - -- Add support for recovering live commands, notebooks, tensorboards and shells on master restart. - This is an extension of live trial recovery, available since version 0.18.1. - -:orphan: - -**Improvements** - -- CLI: add `det (notebook|shell|tensorboard) list --json` option, allowing user to get - JSON-formatted notebook, shell or tensorboard task list. diff --git a/docs/release-notes/shm-size-str.txt b/docs/release-notes/shm-size-str.txt deleted file mode 100644 index edd48ddfce5..00000000000 --- a/docs/release-notes/shm-size-str.txt +++ /dev/null @@ -1,6 +0,0 @@ -:orphan: - -**Improvements** - -- Configuration: Experiment configuration`resources.shm_size`` now supports passing in a unit like - ``4.5 G`` or ``128MiB``. diff --git a/docs/release-notes/state-wrong-allocation-state.txt b/docs/release-notes/state-wrong-allocation-state.txt deleted file mode 100644 index 49ff7666bb6..00000000000 --- a/docs/release-notes/state-wrong-allocation-state.txt +++ /dev/null @@ -1,5 +0,0 @@ -:orphan: - -**Bug Fixes** - -- API: Fix a bug where ``/api/v1/tasks/{taskId}`` would often return incorrect allocation states.