-
Type:
New Feature
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: Diagnostics
-
None
-
Fully Compatible
-
Dev Tools 2019-08-12, Dev Tools 2019-09-09, Dev Tools 2019-09-23, Dev Tools 2019-10-07, Dev Tools 2019-10-21, Dev Tools 2019-11-04, Dev Tools 2019-11-18, Dev Tools 2019-12-02
We have two 'NIX versions of the crash backtracer, one based on libunwind and one based on the 'backtrace' function from execinfo, and then a separate Windows implementation.
However, we have other places in the code that explicitly call 'backtrace' or interact with or construct the object load info per dl_iterate_phdr. One notable example is the heap profiler in src/mongo/util/heap_profiler.cpp, and another is the DiagnosticInfo support soon arriving per SERVER-41364.
In both cases, what is required is an interface into the backtracing system that only collects the backtrace, but does not symbolize it, along with mechanism to cache the necessary information to do symbolization and an entry point to do it at some later time.
Rather than re-inventing this logic several times, we should enhance the API exposed by our abstract backtrace machinery and migrate these instances making direct use of other APIs to use our new abstract API.
- depends on
-
SERVER-41364 Allow generation of compressed stack of addresses
-
- Closed
-
- is duplicated by
-
-
- Closed
-
- is related to
-
SERVER-44378 Add utilities for backtrace support API
-
- Closed
-
- related to
-
-
- Closed
-
-
-
- Closed
-