Update README for the initial release of cloud-tpu-diagnostics package to production

PiperOrigin-RevId: 538581626

Author: SurbhiJainUSC

pip_package/CHANGELOG.md

@@ -38,9 +38,10 @@ To release a new version (e.g. from `1.0.0` -> `2.0.0`):
 
 ## [Unreleased]
 
-## [0.1.0] - YYYY-MM-DD
+## [0.1.0] - 2023-06-08
 
-* Initial release
+* Initial release of cloud-tpu-diagnostics PyPI package
+* FEATURE: Contains debug module to collect stack traces on faults
 
 [Unreleased]: https://github.com/google/cloud-tpu-monitoring-debugging/compare/v0.1.0...HEAD
 [0.1.0]: https://github.com/google/cloud-tpu-monitoring-debugging/releases/tag/v0.1.0

pip_package/README.md

@@ -17,3 +17,84 @@
 
 This is a comprehensive library to monitor, debug and profile the jobs running on Cloud TPU.
 To learn about Cloud TPU, refer to the [full documentation](https://cloud.google.com/tpu/docs/intro-to-tpu).
+
+## Features
+### 1. Debugging
+#### 1.1 Collect Stack Traces
+This module will dump the python traces when a fault such as Segmentation fault, Floating-point exception, Illegal operation exception occurs in the program. Additionally, it will also periodically collect stack traces to help debug when a program running on Cloud TPU is stuck or hung somewhere.
+
+## Installation
+To install the package, run the following command on TPU VM:
+
+```
+pip install cloud-tpu-diagnostics
+```
+
+## Usage
+To use this package, first import the module:
+
+```
+from cloud_tpu_diagnostics import diagnostic
+from cloud_tpu_diagnostics.configuration import debug_configuration
+from cloud_tpu_diagnostics.configuration import diagnostic_configuration
+from cloud_tpu_diagnostics.configuration import stack_trace_configuration
+```
+
+Then, create configuration object for stack traces. The module will only collect stack traces when `collect_stack_trace` parameter is set to `True`. There are following scenarios supported currently:
+
+##### Scenario 1: Do not collect stack traces on faults
+
+```
+stack_trace_config = stack_trace_configuration.StackTraceConfig(
+                      collect_stack_trace=False)
+```
+This configuration will prevent you from collecting stack traces in the event of a fault or process hang.
+
+##### Scenario 2: Collect stack traces on faults and display on console
+
+```
+stack_trace_config = stack_trace_configuration.StackTraceConfig(
+                      collect_stack_trace=True,
+                      stack_trace_to_cloud=False)
+```
+If there is a fault or process hang, this configuration will show the stack traces on the console (stderr).
+
+##### Scenario 3: Collect stack traces on faults and upload on cloud
+
+```
+stack_trace_config = stack_trace_configuration.StackTraceConfig(
+                      collect_stack_trace=True,
+                      stack_trace_to_cloud=True)
+```
+This configuration will temporary collect stack traces inside `/tmp/debugging` directory on TPU host if there is a fault or process hang. Additionally, the traces collected in TPU host memory will be uploaded to Google Cloud Logging, which will make it easier to troubleshoot and fix the problems.
+
+By default, stack traces will be collected every 10 minutes. In order to change the duration between two stack trace collection events, add the following configuration:
+
+```
+stack_trace_config = stack_trace_configuration.StackTraceConfig(
+                      collect_stack_trace=True,
+                      stack_trace_to_cloud=True,
+                      stack_trace_interval_seconds=300)
+```
+This configuration will collect the stack traces on cloud after every 5 minutes.
+
+Then, create configuration object for debug.
+
+```
+debug_config = debug_configuration.DebugConfig(
+                stack_trace_config=stack_trace_config)
+```
+
+Then, create configuration object for diagnostic.
+
+```
+diagnostic_config = diagnostic_configuration.DiagnosticConfig(
+                      debug_config=debug_config)
+```
+
+Finally, call the `diagnose()` method using `with` and wrap the statements inside the context manager for which you want to collect the stack traces.
+
+```
+with diagnostic.diagnose(diagnostic_config):
+    run_job(...)
+```