907312 - Report: stackwalking success/failure

Status: RESOLVED INCOMPLETE

(Socorro :: General, task)

Description

[Benjamin Smedberg]

I'd like a report to help identify the success rate of stackwalking and reasons why stackwalking might fail. jMDSW can be expanded to produce output on how it performed stackwalking:

* exact frame walk via symbols
* exact frame walk via frame pointer
* guessed frame walk via frame pointer
* guessed frame walk via scanning - extra data = scan size
(there may be other reasons)

It would be good to report on the total rate of successful stackwalk (back to main or threadstart).  It is also important to know which DLLs are missing symbols, so that we can identify which external parties we need to push on to receive symbols, or worse whether we have symbol issues with Mozilla binaries.

Comment 1

[(not currently active) Ted Mielczarek]

We'll need to expose the frame trust values first.

There's also a "modules missing symbols" list available nowadays as a result of processing, we could surface that.

Comment 2

[Will Kahn-Greene [:willkg] ET needinfo? me]

There's a bunch of stuff here:

1. Which DLLs are missing symbols. Tecken is tracking missing symbols, so I think we're good here.

2. Metrics on stackwalking. In Datadog we can see success/failure for mdsw, but it's success/failure from a process perspective and not a stackwalking perspective. I'm not really sure what metrics we're looking for here or how to figure them out or what we can do knowing those metrics. Ted mentions that we need to expose the frame trust values--those are in the report view now.

Ted: Since Benjamin is gone, do you want to weigh in here with what you want out of this? What would help you and Gabriele debug breakpad/mdsw problems?

Alternatively, if these things are covered elsewhere, we can mark this INCOMPLETE.

Comment 3

[(not currently active) Ted Mielczarek]

(In reply to Will Kahn-Greene [:willkg] ET needinfo? me from comment #2)

There's a bunch of stuff here:

1. Which DLLs are missing symbols. Tecken is tracking missing symbols, so I think we're good here.

2. Metrics on stackwalking. In Datadog we can see success/failure for mdsw, but it's success/failure from a process perspective and not a stackwalking perspective. I'm not really sure what metrics we're looking for here or how to figure them out or what we can do knowing those metrics. Ted mentions that we need to expose the frame trust values--those are in the report view now.

Ted: Since Benjamin is gone, do you want to weigh in here with what you want out of this? What would help you and Gabriele debug breakpad/mdsw problems?

It's an interesting idea but I think actually making this work in practice might be more difficult than it's worth. A "successful" stack walk is one that accurately recovers all the original frames down to the system entry point, but asserting that correctness isn't easy. Given that nobody has pushed on this in the past 6 years I say we let this go.