[action] [PR:4699] [orchagent]: _exit() on graceful shutdown to avoid destructor-chain memory corruption#4788
Open
mssonicbld wants to merge 1 commit into
Open
Conversation
…emory corruption **What I did** Made orchagent exit through _exit(0) when a graceful shutdown is requested, instead of falling through to return 0 and running the full ~OrchDaemon destructor chain. Before exiting it drains the async swss recorder so any pending records are flushed first. **Why I did it** Fixes sonic-net#4600. PR sonic-net#4400 added graceful shutdown handling. On SIGTERM or SIGINT the signal handler sets gOrchShutdownRequested, OrchDaemon::start() returns, and main() falls through to return 0. That runs the full ~OrchDaemon destructor chain, which is not safe to run on shutdown. During that teardown, FlexCounterManager destruction makes blocking SAI calls (stopFlexCounterPolling, which calls set_switch_attribute). Those calls go through sairedis's ZMQ channel to syncd and park the main thread in zmq_poll while several libzmq IO threads are still running. Orchs that were deleted earlier in the reverse order delete loop have freed buffers that those IO threads still reference, so the process corrupts its own heap. We saw orchagent crash in a libzmq IO thread when METADATA configuration was changed and the containers were restarted. **How I verified it** Configure the route state async publish and ZMQ options in METADATA, then restart the containers so orchagent receives SIGTERM. Before this change orchagent crashes fairly consistently during teardown. With this change orchagent shuts down cleanly and the crash no longer reproduces. **Details if related** Draining the async swss recorder through setAsync(false) joins the recorder worker so pending records are flushed, which was the reason PR sonic-net#4400 wanted start() to return cleanly. Signed-off-by: Sonic Build Admin <sonicbld@microsoft.com>
Collaborator
Author
|
Original PR: #4699 |
|
Azure Pipelines: There may be pipelines that require an authorized user to comment /azp run to run. |
Collaborator
Author
|
/azp run |
|
Azure Pipelines: Successfully started running 1 pipeline(s). |
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
What I did
Made orchagent exit through _exit(0) when a graceful shutdown is requested, instead of falling through to return 0 and running the full ~OrchDaemon destructor chain. Before exiting it drains the async swss recorder so any pending records are flushed first.
Why I did it
Fixes #4600.
PR #4400 added graceful shutdown handling. On SIGTERM or SIGINT the signal handler sets gOrchShutdownRequested, OrchDaemon::start() returns, and main() falls through to return 0. That runs the full ~OrchDaemon destructor chain, which is not safe to run on shutdown.
During that teardown, FlexCounterManager destruction makes blocking SAI calls (stopFlexCounterPolling, which calls set_switch_attribute). Those calls go through sairedis's ZMQ channel to syncd and park the main thread in zmq_poll while several libzmq IO threads are still running. Orchs that were deleted earlier in the reverse order delete loop have freed buffers that those IO threads still reference, so the process corrupts its own heap. We saw orchagent crash in a libzmq IO thread when METADATA configuration was changed and the containers were restarted.
How I verified it
Configure the route state async publish and ZMQ options in METADATA, then restart the containers so orchagent receives SIGTERM. Before this change orchagent crashes fairly consistently during teardown. With this change orchagent shuts down cleanly and the crash no longer reproduces.
Details if related
Draining the async swss recorder through setAsync(false) joins the recorder worker so pending records are flushed, which was the reason PR #4400 wanted start() to return cleanly.
Signed-off-by: Sonic Build Admin sonicbld@microsoft.com