次の方法で共有

Tell Mode vs. Ask Mode

  • [アーティクル]

The project I am currently working on is nearing the end. Last week we reached our "Feature Complete" milestone and now we have formally transitioned into the "Stabilizing" phase.

In a couple of team meetings this week, I mentioned the concepts of "Tell Mode" and "Ask Mode" -- a couple of terms I've been using for more years than I can remember. However, not everyone on the team was aware of these concepts. Consequently, I "recycled some bits" (meaning I searched through my e-mail archive and found a message I sent awhile back) in order to provide some more background and details on Tell Mode and Ask Mode.

I thought this worth sharing with a broader audience.

I'll warn you...this is a rather long post and not one I'm expecting many people to read in its entirety. Rather, I think a lot of people would benefit from reading just enough of my e-mail to understand the key concepts and perhaps skim the example bugs to get a feel for the level of detail expected during the "investigation" phase after switching to Ask Mode.

From: JAMESON,JEREMY (Non-A-BPI-AM,unix1)
Sent: Thursday, October 18, 2007 8:03 AM
Subject: v2.0 Transition from "Tell Mode" to "Ask Mode"

On Sunday night, October 21st, RC4 of the [Project] v2.0 solution will be deployed to TEST. This will mark the official transition from "Tell Mode" to "Ask Mode."

What does this mean?

Microsoft uses Tell Mode and Ask Mode to refer to different time periods after the Feature Complete milestone but prior to a release.

During Tell Mode the Development team tells the rest of the team which issues are actively being worked on (resolved). In other words, Development is primarily driving the stabilization process based on what they think should be fixed -- as determined by priority, severity, complexity (i.e. risk), and effort required. This works for a while, but at some point you have to reign in Development in order to ship, er, release.

Thus we transition into Ask Mode whereby Development must ask permission before resolving any issues. It is okay to investigate issues, but no code changes can be made to address an issue until approval by the “War Team” (er, Triage Team) is granted.

During Ask Mode we look first at the scenario that is being fixed (i.e. the motivation for fixing it), we look at why the bug happens (regression, test hole, coding error, etc.), and then we investigate the necessary source code changes and associated risk. Every potential change is heavily scrutinized to evaluate whether it is worth the risk to fix it, because at this point in the schedule, every change -- even one that seems trivial -- has the potential to destabilize the solution and slip the schedule.

While the overall process of the daily triage meetings stays the same as we transition from Tell Mode to Ask Mode, there is a noticeable shift in focus.

During Tell Mode, work items are often immediately triaged as Approved, or work items triaged as Investigate are often just fixed without further review. This is no longer the case once we enter Ask Mode.

In Ask Mode every work item is first assigned a triage of Investigate. The person responsible for investigating the item must thoroughly document details about the item -- in the Description field of the work item, unless a more formal DCR (Design Change Request) is deemed necessary.

The investigator (typically a member of the Development team) documents the:

  • Motivation -- "Why are we making this change?"
  • Proposal -- "How should we resolve the issue?"
  • Risks -- "What is the worst case scenario if we make this change?"
  • Solution -- "What specifically needs to change?"
  • Teams Impacted -- "How much work is required by Development, Test, Release Management, and (potentially) Product Management teams?" (note that the estimates need to come from the respective team members -- the investigator should not "guess")

Note that both Solution and Teams Impacted are not needed, one or the other is fine. If the necessary change (and associated risk) is very small -- such as a configuration change -- then just documenting the Solution should suffice. The goal is not to overburden ourselves with process -- but we also need to ensure that no change is made unless it has been thoroughly evaluated. If the change (or risk) is substantial, then document the impact on each of the various teams.

You can see some example bugs from v1.0 for more details: 1592, 1505, 1533, 1582, and 1549.

After documenting these details, the investigator then changes the Triage field to Recommend Approve (if the value in making the change is greater than the associated risk) or Recommend Reject (if the risk of making the change outweighs the value). The Triage Team then reviews the detailed information about the work item and either changes the Triage field to Approved or Not Approved (i.e. "punt to v.Next").

Note that the Triage Team can certainly “overrule” the investigator, which is why it is critical that no work be done on actually resolving the work item until approval is received from the Triage Team (i.e. the Triage field is set to Approved). Also note that, unless you are a member of the Triage Team – and even then, only during a formal triage meeting – you should not change the Triage field to anything except Recommend Approve or Recommend Reject. Otherwise, you should fully expect a thorough hazing from virtually all team members.

Everyone who was involved in v1.0 knows that we didn't strictly follow this process in the first release of [Project] , but we also have the "scars" to remind us that we need to improve our process for this release.

It is also important to note that Microsoft fully acknowledges that we had many more QFEs for v1.0 than we intended, and we never did deploy v1.1 as a separate release. We need to be committed to ensuring that we don't repeat this in v2.

Yes, we will release v2.0 with known issues. By now, everyone has heard Jeremy's anecdote about the 64,000 "issues" in the RTM version of Windows 2000 -- so no sense repeating that. Rather we need to focus on making sure that most of the "issues" are known, we understand the impact of those issues on the user experience, and -- where necessary -- we find creative ways to circumvent the issues until the underlying changes can be implemented, tested, and deployed.

Note that I have replaced the project code name with [Project] in order to protect the innocent. I also "spoofed" the links to the referenced bugs, since these referred to a "Work Items" list on a secured team site we used for the project (which obviously you wouldn't have access to -- unless you just happen to be one of my cohorts on the project).

I have included slightly "scrubbed" versions of a couple of the bugs below, in case you are interested in seeing examples. If you don't understand all of the acronyms, terms, and technical details...don't worry, you're not supposed to -- unless you actually were a member of this project ;-)

Bug 1592 - LiteratureResults.asp is not working
Motivation:

There appear to be (at least) two problems when following the repro steps:

  1. The Literature Summary list (rendered by LiteratureResults.asp when isortorder=9) should show the number of publications matching the specified criteria, grouped by Literature Type (a.k.a. publication type, a.k.a. Content Type in SharePoint, a.k.a. the ContentType2 managed property). The list should be ordered by Literature Type.
  2. Clicking on a publication type in the Literature Summary list should show the search results for publications of the selected type. According to the original description for this bug, an error occurred at this point.
Proposal:

Correct the Literature Summary view to properly group and order by publication type. Increase the maximum number of search results in this scenario from 200 to 1000.

Ensure that clicking a publication type in the list shows the matching publications of the specified type.

Risk:

The current implementation of LiteratureResults.asp uses a hard-coded maximum of 200 search results (this limit was chosen for performance reasons since there is significant cost in sending large result sets from the SharePoint farm to the legacy ASP farm). Therefore, simply changing the sort order to publication type could potentially truncate the Literature Summary view such that it only showed one publication type (for example, if more than 200 Applications matched the specified criteria).

While it is simple to increase the maximum number of results (for example, from 200 to 1000) when rendering the Literature Summary view, this will put considerable load on SharePoint Search for two reasons:

  1. We need to order first by ContentType2 in order to return the result set in the order expected by the legacy ASP code that generates the Literature Summary view.
  2. SharePoint Search is optimized to sort by rank first (since it is typically desired to show the "best" results at the top of the list). In order to return a result set of, say, 1000 items sorted by something other than rank, SharePoint must do a lot more work before it can trim the result set.

Therefore it is quite possible that SharePoint Search will require > 10 seconds to generate these large result sets (although it is expected that duplicate searches will be cached and therefore return the result set much faster). As noted in bug 1151, this may cause "sporadic errors" in the legacy ASP pages.

Teams Impacted

Development
Modify LiteratureSearchResults.asp to change the max number of results to be a variable with a default value of 200.

Modify SetSortParameters in SearchModule.inc to set the sort expression used in ESI (strSortExpression) and to override the default max results value to allow 1000 results. (1 hour).
 
Release ManagementMerge updated ASP files into legacy VSS and deploy to WCOSLSD and CAGCHEM (1 hour)
 
Test
Retest ESI Chromatogram searches through legacy General Site (2 hours)

Bug 1505 - There are no results displayed in WCOSLSCD if a search is performed on few publication types
Repro steps:
  1. Log on to https://wcoslscd.cos.fabrikam.com/scripts/library.asp
  2. Select the publication type as “Certificate of Analysis” or “Material Safety Data Sheet”  from the drop down and click on search
  3. There are no results displayed for these publication types even though there are lot of documents under these publication types in TEST
Motivation:

LiteratureResults.aspx currently excludes the following content types (since the number of publications of these types is disproportionately higher than other publication types and also because these publication types have separate search forms):

  • Chromatogram
  • Certificate of Analysis
  • Material Safety Data Sheet

However, the current user experience is not desirable since the search form allows users to narrow their search results to just Certificate of Analysis or MSDS publications (which will never return results).

Proposal:

Remove Certificate of Analysis and MSDS from the search criteria. This would allow users to search specifically by these publication types and would also include these results by default.

Risk:

Of the roughly 33,000 publications in the EPI Warehouse, there are approximately 1,300 Certificate of Analysis and 7,100 MSDS publications.

Since the new SharePoint search includes full-text indexing as well as metadata, including these two publication types by default may dramatically change the search results.

Teams Impacted

Development
Modify LiteratureResults.asp to no longer exclude Certificate of Analysis and MSDS publications by default (0.5 hours).

Release Management
Merge updated ASP file into legacy VSS and deploy to WCOSLSCD and CAGCHEM (0.5 hours)

Test
Retest ESI Library searches through legacy General Site (2 hours)

Product Management
Review new Library search results on legacy General Site to determine if the large number of Certificate of Analysis and MSDS publications has a negative impact on search results. (2 hours)

Comments