Skip to content

fix multiprocess_utilization_watcher #34

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Jan 24, 2025
Merged

fix multiprocess_utilization_watcher #34

merged 1 commit into from
Jan 24, 2025

Conversation

AllenXu93
Copy link
Contributor

Fix multiprocess_utilization_watcher;
In pre version, summonitor is sum of all process's memory in one container, that will cause the all process's memory except the first one value wrong.

chaunceyjiang
chaunceyjiang reviewed Nov 6, 2024
View reviewed changes
Copy link
Contributor

@chaunceyjiang chaunceyjiang left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

In pre version, summonitor is sum of all process's memory in one container, that will cause the all process's memory except the first one value wrong.

Could you remind me how I can reproduce this? To facilitate local testing of this PR.

Or can you provide some comparisons before and after this PR is fixed?

@AllenXu93
Copy link
Contributor Author

In pre version, summonitor is sum of all process's memory in one container, that will cause the all process's memory except the first one value wrong.

Could you remind me how I can reproduce this? To facilitate local testing of this PR.

Or can you provide some comparisons before and after this PR is fixed?

There are some problems when run mutiple process in one container.
For example , I run one process that allocate 100Mi memory(with context about 324Mi totally) , in debug log, we can see monitor record memory:

[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->339738624
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->339738624
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->339738624
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->339738624
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->339738624
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->339738624
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->339738624

Then run another process in same container, the log is

[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141488 0 339738624->339738624
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->679477248
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141488 0 339738624->339738624
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->679477248
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141488 0 339738624->339738624
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->679477248
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141488 0 339738624->339738624
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->679477248
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141488 0 339738624->339738624
[HAMI-core Info(2204:140316929261568:multiprocess_memory_limit.c:311)]: set_gpu_device_memory_monitor:3141423 0 339738624->679477248

The problem is that the second process in nvml result , will record both 2 process's memory as it's allocated memory, because of it use summonitor which is the sum memory of all process in this container:

HAMi-core/src/multiprocess/multiprocess_utilization_watcher.c

Lines 174 to 181 in a9ab3b1

sum += processes_sample[i].smUtil;
summonitor += infos[i].usedGpuMemory;
//LOG_WARN("monitorused=%lld %d %d %d",infos[i].usedGpuMemory,proc->hostpid,proc->pid,pidfound);
//LOG_WARN("smutil=%d %d %lu %u %u %u\n",virtual_map[devi],devi,summonitor,processes_sample[i].smUtil,processes_sample[i].encUtil,processes_sample[i].decUtil);
//}
}
set_gpu_device_memory_monitor(processes_sample[i].pid,cudadev,summonitor);
set_gpu_device_sm_utilization(processes_sample[i].pid,cudadev,processes_sample[i].smUtil);

Secondly, if nvmlDeviceGetProcessUtilization and nvmlDeviceGetComputeRunningProcesses return process not in the same order, the monitor will be wrong, because is use same index to get the memory and smUtil:
```

@archlitchi archlitchi merged commit f40e94f into Project-HAMi:main Jan 24, 2025
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

1 more reviewer

@chaunceyjiang chaunceyjiang chaunceyjiang left review comments

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@AllenXu93 @chaunceyjiang @archlitchi @xiaozhouX