gdb-mongodb-server

About

The gdbmongo package contains GDB pretty printers and commands for debugging the MongoDB Server. Its primary target audience is MongoDB employees.

Motivation

The gdbmongo package is mostly born out of joy from tinkering with low-level constructs while writing GDB pretty printers. There are some explicit areas for what it aims to achieve:

1. GDB pretty printers and commands which only work against live MongoDB processes are of limited value. This is because the hang analyzer is only granted enough time in Evergreen to save a core file for each of the lingering processes. Further analysis is deferred until later by making use of these saved core files. GDB pretty printers and commands which are implemented by walking in-memory data structures and not by executing C++ code can run against core dumps and are therefore more widely applicable.
2. New versions of the MongoDB Server are released regularly. Each new git branch fragments the tooling for testing the server. This can cause development on older branches to feel foreign and awkward because so many new enhancements were made in the meantime. Flipping the model so there’s a single version which attempts to work with all supported MongoDB versions can potentially enable more things to “just work.” Another way to think about it is that the new GDB pretty printers and commands may not be getting built for new MongoDB Server functionality and instead may be getting built for a newly-recognized debugging need.

Installation

The gdbmongo package must be loaded into the Python installation that the GDB process is running. In particular, launching gdb from within a Python virtual environment won’t give the GDB process access to the Python packages defined within the virtual environment. This is because gdb is dynamically linked against libpython and therefore always uses the site-packages of the base installation.

Adding the following snippet to a .gdbinit file will cause gdb at launch time to attempt to install the gdbmongo package if it isn’t already installed.

# In your ~/.gdbinit:
python
try:
    import gdbmongo
except ImportError:
    import sys
    if sys.prefix.startswith("/opt/mongodbtoolchain/"):
        import subprocess
        subprocess.run([sys.prefix + "/bin/python3", "-m", "pip", "install", "gdbmongo"], check=True)
        import gdbmongo
    else:
        import warnings
        warnings.warn("Not attempting to install gdbmongo into non MongoDB toolchain Python")

if "gdbmongo" in dir():
    gdbmongo.register_printers()
end

If you don’t plan to use the GDB pretty printers defined in the mongodb/mongo repository then you may want to consider enabling some of the other printers defined by the gdbmongo package by default.

register_printers(*, essentials=True, stdlib=False, abseil=False, boost=False, mongo_extras=False)

Register the pretty printers defined by the gdbmongo package with GDB itself.

The pretty printer collections other than gdbmongo-essentials are defaulted to off to avoid conflicting with the pretty printers defined in the mongodb/mongo repository.

Regardless of whether you choose to enable these other pretty printers by default, each of the gdbmongo-* pretty printer collections can still be enabled later on within GDB. For example, the gdbmongo-mongo-extras pretty printer collection can be enabled with the following command:

(gdb) enable pretty-printer global gdbmongo-mongo-extras

Tip: Use info pretty-printer, enable pretty-printer, disable pretty-printer, and print /r to inspect and toggle the state of any GDB pretty printers.

Usage

The gdbmongo package is a nascent GDB extension and quite limited in what it can do right now. But, if you’re looking to dump the contents of the global LockManager in a core dump, then you can run the following commands:

(gdb) python lock_mgr = gdbmongo.LockManagerPrinter.from_global()
(gdb) python print(lock_mgr.val)