[DocDB] google tcmalloc can crash with a SEGV on stack trace dump #17875
Assignees
Labels
area/docdb
YugabyteDB core features
kind/bug
This issue is a bug
priority/medium
Medium priority issue
Comments
mbautin
added
area/docdb
yugabyte-ci
added
kind/bug
mbautin
added a commit
to yugabyte/yugabyte-db-thirdparty
that referenced
this issue
Jun 21, 2023
This allows the application to know in a signal handler whether a memory allocation/deallocation operation was interrupted by the signal. Needed to fix the issue yugabyte/yugabyte-db#17875. Also update LLVM version used to 16.0.5.
mbautin
added a commit
to yugabyte/yugabyte-db-thirdparty
that referenced
this issue
Jun 21, 2023
Upgrade tcmalloc to pick up IsCurThreadInAllocDealloc. This allows the application to know in a signal handler whether a memory allocation/deallocation operation was interrupted by the signal. Needed to fix the issue yugabyte/yugabyte-db#17875. The previous commit a93cc9d that attempted to do this had an incorrect tcmalloc tag. The tag/commit used here is https://github.com/yugabyte/tcmalloc/releases/tag/e116a66-yb-4 ( yugabyte/tcmalloc@677ba2d ). Also update LLVM to 16.0.6.
mbautin
added a commit
that referenced
this issue
Jun 23, 2023
Summary: When trying to capture a stack trace with a signal handler, if a memory allocation/deallocation is happening in the thread receiving the signal, the process could crash. Google TCMalloc issue: google/tcmalloc#189. In this diff, we are using the IsCurThreadInAllocDealloc malloc extension API we added in yugabyte/tcmalloc@677ba2d to skip capturing the stack trace in case the signal interrupted a thread that is currently allocating or deallocating memory. In such cases, we produce an empty stack trace which is later omitted from the overall threads dump. #17889 is a follow-up issue for retrying obtaining stack traces in such cases. Another change contained in the TCMalloc version that we are upgrading to is yugabyte/tcmalloc@d1b0e69 (adding an option to not seed lifetime profiler with live allocations). We are now setting seed_with_live_allocs to false when capturing an allocation profile. Test Plan: Jenkins Reviewers: asrivastava Reviewed By: asrivastava Subscribers: ybase, bogdan Differential Revision: https://phorge.dev.yugabyte.com/D26349
1 task
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Jira Link: DB-6957
Description
When trying to capture a stack trace with a signal handler, if a memory allocation/deallocation is happening in the thread receiving the signal, the process could crash.
Filed this issue in google/tcmalloc: google/tcmalloc#189
Warning: Please confirm that this issue does not contain any sensitive information
The text was updated successfully, but these errors were encountered: