Enable CUDA checkpointing with multiple processes #2470
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
avagin
reviewed
Aug 17, 2024
| * before freezing; otherwise, the cuda-checkpoint tool may hang. | ||
| */ | ||
| if (opts.freeze_cgroup) | ||
| ret = pause_devices_in_cgroup(); |
There was a problem hiding this comment.
Here is a race window between pause_devices_in_cgroup and freeze_processes.
avagin
reviewed
Aug 17, 2024
avagin
reviewed
Aug 17, 2024
| run_plugins(PAUSE_DEVICES, pid); | ||
| ptr = end_ptr; | ||
| } else { | ||
| /* Move to the next line if the current line was invalid */ |
There was a problem hiding this comment.
report a error if the line is "invalid"
avagin
reviewed
Aug 17, 2024
| } | ||
|
|
||
| /* Read cgroup.procs into a buffer */ | ||
| while ((bytes_read = read(procs_fd, buffer, sizeof(buffer) - 1)) > 0) { |
There was a problem hiding this comment.
you must understand the last line in the buffer can be incomplete and you need to handle it properly
rst0git
force-pushed
the
cuda-multiple-processes
branch
from
6043fc4 to
421e873
Compare
|
Any update? |
When checkpointing a container, CRIU uses the cgroup path specified by the container runtime (e.g., runc) via the `--freeze-cgroup` option to pause all running processes in the container and obtain a consistent process tree. However, if the container has multiple processes with CUDA state, we need to "lock" these processes before freezing the cgroup; otherwise the cuda-checkpoint tool may hang. To address this problem, this patch updates the collect_pstree function to run the CUDA plugin PAUSE_DEVICES hook for all processes in the container cgroup prior to freezing. In addition, this change introduces a mechanism to disable the use of freeze cgroups during process seizing, even if explicitly requested via the --freeze-cgroup option. The CUDA plugin is updated to utilize this new mechanism to ensure compatibility. Signed-off-by: Andrei Vagin <[email protected]> Signed-off-by: Radostin Stoyanov <[email protected]>
rst0git
force-pushed
the
cuda-multiple-processes
branch
from
421e873 to
76b2fa5
Compare
|
Closing in favour of #2475 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When checkpointing a container with multiple CUDA processes, CRIU currently fails with the following error:
This error occurs because CRIU uses the cgroup specified by the container runtime to freeze all processes running in the container. However, if there are multiple processes that have CUDA state, we need to "lock" all these processes before freezing the cgroup; otherwise the
cuda-checkpointtool hangs.To address this problem, this pull request updates the
collect_pstreefunction to run thePAUSE_DEVICEShook for all processes in the container cgroup prior to freezing.The error above can be replicated with the following example: