From e10942aa5aaf416994d8a6dddb3bc5b36dac9521 Mon Sep 17 00:00:00 2001 From: azhou-determined <83614683+azhou-determined@users.noreply.github.com> Date: Tue, 11 Apr 2023 16:31:09 -0700 Subject: [PATCH] docs: add release notes for 0.21.1 (#6530) --- docs/release-notes.rst | 53 ++++++++++++++++++- docs/release-notes/6460-hdfs.txt | 6 --- .../checkpoint-size-speedup-speedup.rst | 6 --- ...x-default-pool-task-container-defaults.rst | 7 --- docs/release-notes/fix-zero-slot-limits.rst | 6 --- docs/release-notes/launcher-jvm-args.rst | 9 ---- docs/release-notes/metric-stream.rst | 11 ---- docs/release-notes/remove-det-deploy.rst | 6 --- docs/release-notes/remove-old-master-logs.rst | 5 -- 9 files changed, 51 insertions(+), 58 deletions(-) delete mode 100644 docs/release-notes/6460-hdfs.txt delete mode 100644 docs/release-notes/checkpoint-size-speedup-speedup.rst delete mode 100644 docs/release-notes/fix-default-pool-task-container-defaults.rst delete mode 100644 docs/release-notes/fix-zero-slot-limits.rst delete mode 100644 docs/release-notes/launcher-jvm-args.rst delete mode 100644 docs/release-notes/metric-stream.rst delete mode 100644 docs/release-notes/remove-det-deploy.rst delete mode 100644 docs/release-notes/remove-old-master-logs.rst diff --git a/docs/release-notes.rst b/docs/release-notes.rst index 0787e935163..32da6158c8f 100644 --- a/docs/release-notes.rst +++ b/docs/release-notes.rst @@ -10,6 +10,55 @@ Version 0.21 ************** +Version 0.21.1 +============== + +**Release Date:** April 11, 2023 + +**Breaking Change** + +- Remove old master logs ``/logs`` endpoint. Users should use ``/api/v1/master/logs`` instead. + +**Bug Fixes** + +- Fix an issue introduced in 0.19.9 where ``task_container_defaults`` for the default resource + pools were not respected for experiments and tasks unless they specified the resource pool name + explicitly. + +- Checkpoints: Fix an issue where checkpoint insertion on a cluster with a lot of checkpoints and + reported metrics could take a long time. + +- Kubernetes: Fix a crash affecting zero-slot workloads when ``resources.limits`` and + ``resources.requests`` overrides were explicitly specified in the pod spec. + +**Deprecated Features** + +- HDFS checkpoint storage support has been deprecated and will be removed in a future version. + Please contact Determined if you still need it, or else migrate to a different storage backend. + +**Improvement** + +- Cluster: Add HPC Launcher support for JVM resource configuration. + + - The master configuration option ``resource_manager.launcher_jvm_args`` can be used to override + the default HPC Launcher JVM heap configuration. This support requires HPC Launcher version + 3.2.6 or greater. + +**New Features** + +- Python SDK: Add methods for efficient export of training and validation metrics to the Python + SDK. The methods are listed below. + + - :meth:`~determined.experimental.client.stream_trials_training_metrics` + - :meth:`~determined.experimental.client.stream_trials_validation_metrics` + - :meth:`~determined.experimental.client.TrialReference.stream_training_metrics` + - :meth:`~determined.experimental.client.TrialReference.stream_validation_metrics` + +**Removed Features** + +- The separate ``det-deploy`` executable was deprecated in 0.15.0 (April 2021) and is now removed. + Use the ``det deploy`` subcommand instead. + Version 0.21.0 ============== @@ -44,8 +93,8 @@ Version 0.21.0 **Improvements** - CLI: Command-line deployments will now default to provisioning Nvidia T4 GPU instances instead of - K80 instances. This change is intended to improve the performance/cost and driver support of - the default deployment. + K80 instances. This change is intended to improve the performance/cost and driver support of the + default deployment. - Kubernetes: Ease permission requirements in Kubernetes so master no longer requires access to all Kubernetes namespaces. This only affects custom modified Helm chart configurations. diff --git a/docs/release-notes/6460-hdfs.txt b/docs/release-notes/6460-hdfs.txt deleted file mode 100644 index 35ed402367c..00000000000 --- a/docs/release-notes/6460-hdfs.txt +++ /dev/null @@ -1,6 +0,0 @@ -:orphan: - -**Deprecated Features** - -- HDFS checkpoint storage support has been deprecated and will be removed in a future version. - Please contact Determined if you still need it, or else migrate to an different storage backend. diff --git a/docs/release-notes/checkpoint-size-speedup-speedup.rst b/docs/release-notes/checkpoint-size-speedup-speedup.rst deleted file mode 100644 index 567645f46b2..00000000000 --- a/docs/release-notes/checkpoint-size-speedup-speedup.rst +++ /dev/null @@ -1,6 +0,0 @@ -:orphan: - -**Bug Fixes** - -- Checkpoints: Fix an issue where checkpoint insertion on a cluster with a lot of checkpoints and - reported metrics could take a long time. diff --git a/docs/release-notes/fix-default-pool-task-container-defaults.rst b/docs/release-notes/fix-default-pool-task-container-defaults.rst deleted file mode 100644 index ee68308eeb4..00000000000 --- a/docs/release-notes/fix-default-pool-task-container-defaults.rst +++ /dev/null @@ -1,7 +0,0 @@ -:orphan: - -**Bug Fixes** - -- Fix an issue that caused ``task_container_defaults`` for the default resource pools to be ignored - for experiments and tasks, unless the resource pool name was explicitly specified. This issue - first appeared in version 0.19.9. diff --git a/docs/release-notes/fix-zero-slot-limits.rst b/docs/release-notes/fix-zero-slot-limits.rst deleted file mode 100644 index 50382a2f9b4..00000000000 --- a/docs/release-notes/fix-zero-slot-limits.rst +++ /dev/null @@ -1,6 +0,0 @@ -:orphan: - -**Bug Fixes** - -- Kubernetes: Fix a crash affecting zero-slot workloads when ``resources.limits`` and - ``resources.requests`` overrides were explicitly specified in the pod spec. diff --git a/docs/release-notes/launcher-jvm-args.rst b/docs/release-notes/launcher-jvm-args.rst deleted file mode 100644 index 769c5c32e63..00000000000 --- a/docs/release-notes/launcher-jvm-args.rst +++ /dev/null @@ -1,9 +0,0 @@ -:orphan: - -**Improvement** - -- Cluster: HPC Launcher support for JVM resource configuration. - - - The master configuration ``resource_manager.launcher_jvm_args`` can be used to override the - default HPC launcher JVM heap configuration. This support requires HPC Launcher version 3.2.6 - or greater. diff --git a/docs/release-notes/metric-stream.rst b/docs/release-notes/metric-stream.rst deleted file mode 100644 index fce0930875f..00000000000 --- a/docs/release-notes/metric-stream.rst +++ /dev/null @@ -1,11 +0,0 @@ -:orphan: - -**New Features** - -- Python SDK: Adds methods for efficient export of training and validation metrics to the Python - SDK. The methods are listed below. - - - :meth:`~determined.experimental.client.stream_trials_training_metrics` - - :meth:`~determined.experimental.client.stream_trials_validation_metrics` - - :meth:`~determined.experimental.client.TrialReference.stream_training_metrics` - - :meth:`~determined.experimental.client.TrialReference.stream_validation_metrics` diff --git a/docs/release-notes/remove-det-deploy.rst b/docs/release-notes/remove-det-deploy.rst deleted file mode 100644 index cbaf8cf1d06..00000000000 --- a/docs/release-notes/remove-det-deploy.rst +++ /dev/null @@ -1,6 +0,0 @@ -:orphan: - -**Removed Features** - -- Separate ``det-deploy`` executable was deprecated in 0.15.0 (April 2021) and is now removed. Use - ``det deploy`` subcommand instead. diff --git a/docs/release-notes/remove-old-master-logs.rst b/docs/release-notes/remove-old-master-logs.rst deleted file mode 100644 index 1c75e870a82..00000000000 --- a/docs/release-notes/remove-old-master-logs.rst +++ /dev/null @@ -1,5 +0,0 @@ -:orphan: - -**Breaking Change** - -- Remove old master logs `/logs` endpoint. Users should use `/api/v1/master/logs` instead.