You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We have both TEMP_FAILURE_RETRY_BLOCK_SIGNALS and TEMP_FAILURE_RETRY in our sources. In the presence of the profiler we should not ever use TEMP_FAILURE_RETRY as it has the potential of spinning without making any forward progress.
The text was updated successfully, but these errors were encountered:
In the worst case what can happen when using TEMP_FAILURE_RETRY to wrap system calls is:
10 thread enters system call
20 thread receives SIGPROF signal
30 thread exits system call with EINTR
40 GOTO 10
By using TEMP_FAILURE_RETRY_BLOCK_SIGNALS the program will never get stuck in that loop. Given that we expect the profiler to always be enabled (today it is enabled on all Linux and Mac bots) it doesn't make sense to ever use TEMP_FAILURE_RETRY instead of TEMP_FAILURE_RETRY_BLOCK_SIGNALS. We should remove TEMP_FAILURE_RETRY.
We have both TEMP_FAILURE_RETRY_BLOCK_SIGNALS and TEMP_FAILURE_RETRY in our sources. In the presence of the profiler we should not ever use TEMP_FAILURE_RETRY as it has the potential of spinning without making any forward progress.
The text was updated successfully, but these errors were encountered: