he Exchange Sustained Engineering team recently made the decision to recall

the June 22, 2011 release of Exchange 2010 SP1 Rollup 4. This was not

an action we took lightly and we understand how disruptive this was to

customers. We would like to provide you with some details that will

give you a deeper understanding of what actually happened and, more

importantly, what improvements we are making to prevent this in the

future.

• Q: What actually triggered the recall?

   

  A:

  While fixing a bug that prevented deleted public folders from being

  recovered, we exposed an untested set of conditions with the Outlook

  client. When moving or copying a folder, Outlook passes a flag on a

  remote procedure call that instructs the Information Store to open

  deleted items which haven’t been purged. Our fix inadvertently caused

  the RPC to skip all content that wasn’t marked for deletion because we

  were not expecting this flag on the call from Outlook on the copy and

  move operations.

• Q: Why didn’t you test this scenario?

   

  A:

  The short answer is we thought we did. We didn’t realize we missed a

  key interaction between Exchange and Outlook. The Exchange team has

  well over 100,000 automated tests that we use to validate our product

  before we ship it. With the richness and number of scenarios and

  behaviors that Exchange supports, automated testing is the only scalable

  solution. We execute these tests in varying scenarios and conditions

  repeatedly before we release the software to our customers. We also

  supplement these tests with manual validation where necessary. The

  downside of our tests is that they primarily exercise the interfaces we

  expose and are designed around our specifications. They do test

  positive and negative conditions to catch unexpected behavior and we did

  execute numerous folder copy and move tests against the modified code

  which all passed. What we did not realize is that our tests were not

  emulating the procedure call as executed by Outlook.

• Q: Exchange has been around a while, why did this happen now?

   

  A: In Exchange 2010 we introduced a feature called RPC Client Access. This functionality is responsible for serving as the MAPI

  endpoint for Outlook clients. It allowed us to abstract client

  connections away from the Information Store (on Mailbox servers) and

  cause all Outlook clients to connect to the RPC Client Access service.

  As

  part of our investigation, we discovered that there was some specific

  code added to the Exchange 2003 Information Store to handle the

  procedure call from Outlook using the extra flag. This code was also

  carried forward into Exchange 2007. But when the Exchange team added

  the RPC Client Access service to Exchange 2010, that code was not

  incorporated into the RPC Client Access service because it was

  mistakenly believed to be legacy Outlook behavior that was no longer

  required. That, unfortunately, turned out not to be the case. The fact

  that we were not allowing a deleted public folder to be recovered was

  masking this new bug completely.

• Q: Are there other similar issues lurking in RPC Client Access?

   

  A:

  We do not believe so. The RPC Client Access functionality has been

  well-tested at scale and proven to be reliable for the millions of

  mailboxes hosted in on-premises deployment and in our own Office 365 and

  Live@EDU services.

• Q: What are you doing to prevent similar things from happening in the future?

   

  A:

  We have conducted a top-to-bottom review of the process we use to

  triage, develop and validate changes for Rollups and Service Packs and

  are making several improvements. We have changed the way we evaluate a

  customer requested fix to ensure that we more accurately identify the

  risk and usage scenarios that must be validated for a given fix.

  Recognizing the diversity of clients used to connect to Exchange, we are

  increasing our client driven test coverage to broaden the usage

  patterns validated prior to release. Most notably, we are working even

  closer with our counterparts in Outlook to use their automated test

  coverage against each of our releases as well. We are also looking to

  increase coverage for other clients as well.