Triage Best Practices

  • A process (often taking the form of a regular meeting) for making decisions about the need, priority, timing, and assignment of work requests.
  • Triage boils down to Three Questions:
    1. Do we want to do this work?
      • Pro Tip: Be definitive.  If the work isn't something that we intend on doing (or isn't something we're able to confirm is a real issue), then close the issue out and provide honest and courteous feedback.  It's always best to keep your backlog reflective of the work that you intend on doing.
    2. How important is it to us (Priority) and when do we want it done (Schedule/Milestone)?  
      • Note: In the standard workflow Priority and Schedule are very closely related concepts (please see the background on priority below).
    3. Who should own it?
      • Pro Tip: People can only do a finite amount of work, adding more work to someone's pile won't get it done faster.
      • Pro Tip: It's perfectly fine to mark something as Status:Available (i.e. no Owner), if the work doesn't need to be scheduled immediately, but please make sure that the work has a specific categorization label (e.g. Component=UI>Browser>Foo), so that it's clear which team(s) are accountable for it's priority.

How to Think about Priorities

  • P0 - Emergency
    • Requires immediate resolution
  • P1 - Needed for current milestone
    • Teams should use best discretion here, but this bucket should generally represent high user impact issues and quality issues (e.g. regressions, crashers, security issues, feature completeness, etc...).
    • The test: "Would someone notice (in a bad way) if this issue were present in the release?"
  • P2 - Wanted for current milestone
    • Low/ No user impact, can safely be punted from a release.
    • Most in-development work should have this priority.
  • P3 - Not time sensitive
    • Can be completed at any time, no release targeting required.

Recommended Processing Order for Triage Practice

  • Start by checking your team's load balance.
    • Using your standard triage query, pull up the "Grid" view. Set Rows to be "Owner," Columns to be "None," and Cells to be "Counts"
    • Get a rough sense for everyone's issue distribution.
    • Pro Tip: Keep that page opened in the background, as a reference for when you are assigning issues.
  • Assign unassigned (-has:owner) Release Blockers + P0 & P1 issues.
    • These are important to start with first, since these issues are likely time sensitive (notice Blockers, which actively hold back releases, are prioritized first).
    • Pro Tip: It's OK to change the Blocking status or Priority, but please be clear about why (e.g. "this isn't a release block stable because ....")
  • Untriaged (status=untriaged and status=unconfirmed) issues next, ordering by age (i.e. try and process the requests that have been in the queue longest)
    • The point here is really to answer the 3 questions (i.e. "Is this something we want to do?," "How important is it?," "Who should do it?")
    • Please ensure that all P1+ issues are assigned an owner.
    • Pro Tip: Ideally you'll budget enough time to get through everything, but if your list is too large to process in a single triage session, prioritize looking at Type-Bug-Regression (i.e. they are the most likely to be P1+ issues).
  • Once you've triaged the most important issues, sanity check your team's load balance (to make sure that you haven't overloaded anyone).
  • Review / Assign unassigned P2 & P3 issue, as well as any open Available issues.

More Pro Tips

  • Set aside a regular time (and place) to Triage.
  • Give yourself enough time to handle your incoming stream of issues and make progress on (or at least review) your backlog.
  • Setup a rotation w/ your team members, to share the load. Though they are a little more burdensome to administer, they tend to be the most sustainable triage efforts, help prevent people from burning out, and ensure that the process doesn't break when people take a vacation.
  • Wade into and clean-up your backlog. New team members will often start from this pool of issues, please be kind and make sure that they start by working on valid/ relevant issues.
  • Make sure that things are properly categorized, especially as Available issues, so that they are discoverable.
  • Spend most of your time on issues opened in the last year.
    • Most issues have a half-life. Issue reports are a static reflection of what was... not necessarily what is. That is... due to a variety of factors (e.g. code churn, environment changes, etc...) as issues age it becomes increasingly likely that they will become less reflective of the current state of the product.
    • People will re-file (and often duplicate) important issues. Many of the old issues that linger often have not been properly duped, because the person who fixed the issue wasn't aware of the other issue's existence.