Edit this page | Blame

Using PDB to Troubleshoot Python Code


Tags

Historically, debugging GeneNetwork code has been a bit of a pain where you would have `print' and `logging' statements to help view offending code chunks. This is not efficient, and we can do better! One painful side-effect wrt logging worth mentioning is that our logs grow quite fast and we need to rotate them, atm manually:

Here are examples of some logging that we do:

@app.route("/n/logout")
def logout():
    logger.debug("Logging out...")
    UserSession().delete_session()
    flash("You are now logged out. We hope you come back soon!")
    response = make_response(redirect(url_for('index_page')))
    # Delete the cookie
    response.set_cookie(UserSession.user_cookie_name, '', expires=0)
    return response
@app.route("/tmp/<img_path>") def
tmp_page(img_path): logger.info("In tmp_page")
logger.info("img_path:", img_path)
logger.info(request.url) initial_start_vars =
request.form logger.info("initial_start_vars:",
initial_start_vars) imgfile =
open(GENERATED_IMAGE_DIR + img_path, 'rb') imgdata
= imgfile.read() imgB64 =
base64.b64encode(imgdata) bytesarray =
array.array('B', imgB64) return
render_template("show_image.html",
img_base64=bytesarray)

Earlier this year, one of our members introduced us to pudb---a graphical based logging utility for Python. I have gravitated away from this because it adds yet another dependency in our toolchain; in addition to it being ncurses-based, lacking guarantees in how it behaves in different terminals. It also lacks support in different OS'es, thereby forcing end-users to SSH into one of our remote servers to troubleshoot.

Python PDB ships with Python, and as such, works well in different setups. There are multiple ways of getting into a pdb session, the easiest being to set a `breakpoint()'. Assume we are trouble-shooting this function:

from typing import List


def avg(numbers: List) -> int:
    return sum(numbers)/len(numbers)


print(avg([20, 21]))

This will fail for a list that contains non-integer value, say a list containing ["1", "2"]. The first step to troubleshoot, assuming we have no test would be to set a `breakpoint()' as such:

from typing import List


def avg(numbers: List) -> int:
    breakpoint()
    return sum(numbers)/len(numbers)


print(avg([20, "21"]))

Useful commands while you are in pdb that are useful:

When we step into our debug session, we can view all the variables in a local scope using: "locals()"; and the global scope using: "globals()". With this information, we can quickly work out where our problem is by just inspecting the variables we have at hand.

Another cool trick/pattern when debugging is to tell pdb to jump to where the error occured in a try/except block using `import pdb; pdb.post_mortem()' like so:

from typing import List


def avg(numbers: List) -> int:
    try:
        return sum(numbers)/len(numbers)
    except Exception:
        import pdb; pdb.post_mortem()


print(avg([20, "21"]))

With regards to testing, pdb is also integrated with test-runners. To use pdb with pytest, simply run:

,---- | pytest --pdb `----

Useful Tutorials


(made with skribilo)