Release Management/WER Investigation

This page will track the initial investigation of Windows Error Reporting (WER). Our current plan is to hook our bug tracking and metrics into Winqual, the site/API used to access Microsoft's crash info. Using this info will help diagnose hangs (which aren't currently tracked) on Windows XP and later, and may even uncover new crashes.

Meetings

• 2011-10-10 - Kickoff
• 2011-10-19

Associated Bugs

• Meta bug - bug 695713
• bug 429592 - discussion of what to do with process hangs
• bug 600275
• bug 692859
• bug 693315 - Add annotation for source of crashes
• bug 696207 - Add AppMap.exe to the build process

Resources

• Short description on MSDN: WER: Getting Started
• Longer guide (more targeted at hw/drivers though): Developers Guide to WER
• Full WinQual Docs
• Microsoft's WER client: Stackhash
• MSDN Webcast: Getting Started with Windows Error Reporting (video from 2006)
• Breakpad design docs - https://code.google.com/p/google-breakpad/w/list
• FAQ - WER Services

Builds

Creating File Mappings

• MS suggests using their Microsoft Product Feedback Mapping Tool (AppMap.exe) to create a file manifest - required for crash collection for a new product version
• (?) Need to explore generating this file manifest so that the build machines can create it
  • Or use the command line interface during the build on a windows machine as outlined at Microsoft Product Feedback Mapping Tool Readme

Uploading File Mappings

• MS suggests using the "Upload File Mappings option" on the Administration menu
• (?) Need to figure out if the WER web API can instead be used to perform an upload

WinQual Update Frequency

• • "By default we collect 10 cab (minidump) files per event"
• Lag times
  • "Once we receive cab files for an event you will generally be able to see these cabs within a few hours of us receiving them."
  • "For newly detected crashes it can take more than 4 days to get the crashes processed and up on the site."
• (?) Determine if there are any limitations to the number of products that can be registered (for possible use with nighties
• (?) Bandwidth limits for pulling down
• Only the first few cab files are stored for a crash. The event viewer web interface offers the ability to make a "Data Request" to collect "Processor & Memory Information", the heap, specific files (logs, etc.), additional cab files
  • (?) Can the web API be used to expose this to developers or will we need more accounts?
• (?) Need to find out what it means when no cab files are available and the web interface offers "(click here to switch to collection mode)" - aren't we always collecting?

Accessing WinQual Data

Windows Live Login

• (?) Access in winqual requires a Windows Live login. Similarly, in MS's StackHash client requires the use of Windows Live Sign-in Assistant. Need to determine if the associated library is required for logging in.
  • Windows Live ID Web Authentication
  • Using the REST API Service
• (?) Need to figure out what account we'd use for automated tools

WER Web API

• API documentation: StackHash source download > 3rdparty > WinQual API > Data Services.docx

Breakpad/Soccoro

• Crash dumps are stored in buckets
  • "For crash events the bucketing parameters are Application Name, Application Version, Application Build Date, Module Name, Module Version, Module Build Date, Exception Code, and Code Offset"
• (?) Need to understand the overlap between WinQual crash data and Breakpad crash data, map the applicable info, and decide what to do with "additional" info
  • (?) Can we make use of individual hit event info in general? (as opposed to just crash cabs)
• (?) Will hangs (heap dumps) need to be handled any differently than minidumps?
• (?) Who do we give access to minidumps?
• (?) What is our current data retention policy?
  • We may want to keep hangs around for longer since there may be a lot, and they've never been investigated
• (?) What is our access audit ability?

Cab File Contents (for collector/processor)

• FAQ - WER Services - more info here under the question "What are the different types of memory dumps?"
• WERInternalMetadata.xml - (possibly) not present if version.txt is. Includes
  • OSVersionInformation - windows version info, architecture, etc.
  • ProblemSignatures - event type (crash/hang), crashing executable name, exe version/timestamp, methodDef token of faulting method (?), and IL offset of faulting instruction (?)
  • DynamicSignatures -
  • SystemInformation - HW info. What's an MID?
• AppCompat.txt (also all lower) - not present if WERDataCollectionFailure.txt is. Includes information on all images loaded by the process.
• WERDataCollectionFailure.txt - includes error message if processing failed in MS.
• version.txt - only came across this once. Only included OS version.
• For crashes
  • ______.{m}dmp - minidump file. These types of dumps already appeared to be handled according to http://code.google.com/p/google-breakpad/wiki/ProcessorDesign
• For hangs
  • <process-name>.xml - additional hang metadata like the wait chain list
  • memory.hdmp - info about the difference between a heapdump and a minidump outlined here

Future Investigations

• Consider providing a solution (link) when the user is presented with the Windows crash dialog
  • Can even link to an exe (if part of the "Designed for Windows" logo program), which may be a good idea for last ditch effort if even Firefox's safe mode fails (application files no longer pristine, need reinstall).