Closed nhamilakis closed 11 months ago
An interesting alternative is to use a timer from threading, as shown on this stack overflow thread.
Implementing 6f1223c showed that the process kill does not come from the SIGALRM, but from the Grouping computing itself.
HINT: the exit code 137 indicated the process received a SIGKILL, this could be from the OS as the computing used too many resources (see attached screenshot) ?
HINT2: should check on a correct way to debug the issue (obtain a proper exit message)
The cause of the kill was the OOM-Killer that runs on Linux and kills any process that uses too much memory, the solution was to use a stronger PC (16GB of RAM seems to be enough).
Though observation, the most used was 10 GB of RAM.
When calculating Grouping in the tde17 benchmark we use a SIGALRM to put a timeout, if the computation takes more than
grouping_max_time
to compute we skip it.The SIGALRM sometimes causes the process to be killed (maybe if the computation ends too quickly ?).