Technology for stall detection

摘要

Detecting stalling of a software process in a computer system includes receiving identification of a task thread group executing in a work process executing on a computer system. The task thread group includes one or more threads and the receiving includes receiving identification of the one or more threads by a control process executing on a computer system. The detecting includes detecting whether there is a thread state change for the task thread group, marking the task as running responsive to detecting a thread state change for the task thread group, marking the task as stalled responsive to detecting an absence of a thread state change for at least a predefined amount of time, and marking the work process as stalled responsive detecting an absence of a predetermined signal from the work process for at least a predefined amount of time.

主权项

1. A method for detecting stalling comprising:
receiving identification of a work process thread group for a task, the work process thread group executing in a work process executing on a computer system, wherein the work process thread group includes one or more work process threads and the receiving includes receiving identification of the one or more work process threads by a control process executing on the computer system; detecting whether there is at least one state change for at least one of the one or more work process threads of the work process thread group; marking the task as running responsive to detecting the at least one state change for the at least one of the one or more work process threads of the work process thread group; marking the task as stalled responsive to detecting an absence of the at least one state change for the at least one of the one or more work process threads of the work process thread group for at least a first predefined amount of time; marking the work process as stalled responsive to detecting an absence of a predetermined signal from the work process for at least a second predefined amount of time; getting stack traces for the at least one of the one or more work process threads of the work process thread group via calls of an application programming interface; and calculating instances of a hash at respective times for each of the stack traces, and wherein the detecting of whether there is at least one state change for the at least one of the one or more work process threads of the work process thread group comprises: comparing respective instances of the hash calculated at least the first predetermined amount of time apart.