These are the queries used for this test:

Query A

SELECT 
	{Actual} ON 0
FROM [GL] 
CELL PROPERTIES VALUE

Query B

SELECT 
	{Actual} ON 0,
	NON EMPTY Hierarchize({DrilldownLevel({[Account].[Accounts].[All]},,,INCLUDE_CALC_MEMBERS)}) ON 1
FROM [GL] 
CELL PROPERTIES VALUE

Query C

SELECT 
	{Actual} ON 0,
	NON EMPTY Hierarchize({DrilldownLevel({[Date].[Fiscal].[All]},,,INCLUDE_CALC_MEMBERS)}) ON 1
FROM [GL] 
CELL PROPERTIES VALUE
GO

Results (Cache was cleared with ClearCache on the entire Database before each Series)

Series 1	Series 2	Series 3

So basically the Grand Total of this GL cube is a random number based on whoever sneaks the first query in after the cache is cleared (processing, sync, etc).

And for all of you that think the MDX script breaks everything everywhere - I did comment out my entire MDX script before running these tests.

The first method is XMLAParallel - in this mode, the application constructs an XMLA command to process the specified number of databases inside of a Parallel element. Only one XMLA command is issued at a time so the applicatoin will not submit another command until the longest processing job completes in each batch.

The second method is Threaded - in this mode, the application creates the specified number of threads and each thread processes databases independantly of the other threads. This allows all threads to remain busy at all times as a given thread will start processing the next unprocessed database as soon as it finishes processing the previous database.

It should be noted that because this app performs only ProcessFull at the database level, it is currently only suitable for processing a large number of small analysis services databases. The environment this was developed for has a process that generates and deploys metadata for over 250 small (< 100MB) ssas databases and we need to process them as quickly as possible.

If you have suggestions for features that would help the application be more useful to you, please post them in the Discussions area.

And yes, I probably should have called it "ParallelDatabaseProcessor." PCP was a cooler acronym.

The application and source code can be downloaded on Codeplex: Parallel Cube Processor

First things fist - the original thought for this came form Chris Webb's article, Killing sessions automatically with SSIS. Now on to my solution - which is basically identical in theory, just implemented in PowerShell with an email notification at the end.

Create a SQL Agent job with a PowerShell step that runs the following script and schedule it to run every 30 seconds

#======= START Config ========
$asInstance = "ssasinstance";
$emailServer = "1.2.3.4";
$fromAddress = "you@yourcompany.com";
$toAddress = "you@yourcompany.com";
$maxQueryElapsedMS = 30000;
$bypassUserPattern = "yourusername";
#======== END Config =========
 
[System.Reflection.Assembly]::LoadWithPartialName("Microsoft.AnalysisServices.Xmla") ;
[Microsoft.AnalysisServices.xmla.xmlaclient]$client = 
  new-object Microsoft.AnalysisServices.Xmla.XmlaClient ;
$client.Connect($asInstance) ;
$commands = "";
$sessions = "";
$client.Discover("DISCOVER_COMMANDS", "", "", [ref] $commands, 0, 0, 1) ;
$client.Discover("DISCOVER_SESSIONS", "", "", [ref] $sessions, 0, 0, 1) ;
$sessionsXml = [xml]$sessions;
$sessionsRows = $sessionsXml.return.root.row;
$commandsXml = [xml]$commands;
$commandsRows = $commandsXml.return.root.row;
 
foreach ($row in $commandsRows) 
{
  if([int]$row.COMMAND_ELAPSED_TIME_MS -gt $maxQueryElapsedMS -and $row.COMMAND_TEXT -match "SELECT") 
  {
    $spid = $row.SESSION_SPID;
    $commandText = $row.COMMAND_TEXT;
    $session = ($sessionsRows | ? { [int]$_.SESSION_SPID -eq [int]$spid });
    $user = $session.SESSION_USER_NAME;
    $db = $session.SESSION_CURRENT_DATABASE;
    if($user -notmatch $bypassUserPattern) {
      $cancelXMLA = "<Cancel xmlns=""http://schemas.microsoft.com/analysisservices/2003/engine""><SPID>$spid</SPID></Cancel>";
      $client.Send($cancelXMLA, $null);
      $smtp = new-object Net.Mail.SmtpClient($emailServer); 
      $smtp.Send(
        "$asInstance <$fromAddress>", 
        $toAddress, 
        "Killed SPID: $spid", 
        "SPID $spid was detected as the owner of a long running query and was killed.`r`nUser: $user`r`nDB: $db`r`n`r`n$commandText"
      );
    }
  }
}
$client.Disconnect();

Great. Now sessions with active queries (aka "commands" with the word "SELECT" in them) taking over 30 seconds will be cancelled and we'll get an email w/ the query so we can debug! So you wake up the next morning and Japan called - a bunch of their pivots errored out last night and they're not happy. Hmm... Pry should have warned someone before we started cancelling queries...

Phase 2: Turn off the query/session killer, turn on the logging

Create the Log table:

CREATE TABLE [dbo].[LongRunningMDXQueryLog](
	[SPID] [INT] NOT NULL,
	[CommandStartTime] [DATETIME] NOT NULL,
	[Username] [NVARCHAR](255) NULL,
	[DATABASE] [NVARCHAR](255) NULL,
	[ElapsedTimeMS] [INT] NOT NULL,
	[CommandText] [NVARCHAR](MAX) NULL
)

Create the MERGE proc to handle log entries

CREATE PROCEDURE [dbo].[LongRunningMDXQueryLog_Merge]
	 @SPID INT
	,@CommandStartTime DATETIME
	,@Username VARCHAR(255)
	,@DATABASE VARCHAR(255)
	,@ElapsedTimeMS INT
	,@CommandText NVARCHAR(MAX)
AS
BEGIN
	;MERGE LongRunningMDXQueryLog a USING
	(
		SELECT
			 @SPID 
			,CAST(@CommandStartTime AS DATETIME)
			,@Username 
			,@DATABASE
			,@ElapsedTimeMS 
			,@CommandText 
	) b
	([SPID]
	,[CommandStartTime]
	,[Username]
	,[DATABASE]
	,[ElapsedTimeMS]
	,[CommandText])
	ON
	(a.[SPID] = b.[SPID]
	 AND a.[CommandStartTime] = b.[CommandStartTime]
	 AND a.[Username] = b.[Username]
	 AND a.[DATABASE] = b.[DATABASE]
	 AND a.[CommandText] = b.[CommandText])
	WHEN NOT MATCHED THEN
	INSERT
			   ([SPID]
			   ,[CommandStartTime]
			   ,[Username]
			   ,[DATABASE]
			   ,[ElapsedTimeMS]
			   ,[CommandText])
		 VALUES
			   ([SPID]
			   ,[CommandStartTime]
			   ,[Username]
			   ,[DATABASE]
			   ,[ElapsedTimeMS]
			   ,[CommandText])
	WHEN MATCHED THEN
	UPDATE
	SET
		[ElapsedTimeMS] = b.[ElapsedTimeMS]
	;
END

Create a job to run the following Powershell script every 30 seconds:

#======= START Config ========
$asInstance = "ssasinstance";
$logConnectionString = 
  "server=sqlserverinstance;" + 
  "database=databaseYouCreatedTheLogTableIn;" + 
  "trusted_connection=true;";
$maxQueryElapsedMS = 30000;
#======== END Config =========
 
[System.Reflection.Assembly]::LoadWithPartialName("Microsoft.AnalysisServices.Xmla") ;
[Microsoft.AnalysisServices.xmla.xmlaclient]$client = 
  new-object Microsoft.AnalysisServices.Xmla.XmlaClient ;
$client.Connect($asInstance) ;
$commands = "";
$sessions = "";
$client.Discover("DISCOVER_COMMANDS", "", "", [ref] $commands, 0, 0, 1) ;
$client.Discover("DISCOVER_SESSIONS", "", "", [ref] $sessions, 0, 0, 1) ;
$sessionsXml = [xml]$sessions;
$sessionsRows = $sessionsXml.return.root.row;
$commandsXml = [xml]$commands;
$commandsRows = $commandsXml.return.root.row;
 
foreach ($row in $commandsRows) 
{
  $elapsedTimeMS = $row.COMMAND_ELAPSED_TIME_MS;
  if([int]$elapsedTimeMS -gt $maxQueryElapsedMS -and $row.COMMAND_TEXT -match "SELECT") 
  {
    $spid = $row.SESSION_SPID;
    $commandText = $row.COMMAND_TEXT -replace "'", "''";
    $session = ($sessionsRows | ? { [int]$_.SESSION_SPID -eq [int]$spid });
    $user = $session.SESSION_USER_NAME;
    $db = $session.SESSION_CURRENT_DATABASE;
    $commandStartTime = 
      ([string]$row.COMMAND_START_TIME -replace '\.\d+$', '' -replace 'T', ' ');
    $sqlConn = New-Object System.Data.SQLClient.SQLConnection($logConnectionString);
    $sqlConn.Open();
    $sqlCommand = New-Object System.Data.SQLClient.SQLCommand;
    $sqlCommand.Connection = $sqlConn;
    $sqlCommand.CommandText = 
      "EXEC LongRunningMDXQueryLog_Merge " + 
        "$spid, " +
        "'$commandStartTime', " + 
        "'$user', " + 
        "'$db', " + 
        "$elapsedTimeMS, " + 
        "'$commandText'";
    $sqlCommand.ExecuteNonQuery()
    $sqlConn.Close();
  }
}
$client.Disconnect();

And now you have a table filled with queries that will be cancelled (read: "error out") when the query killing job is running. So now you fix it:

1. locate these users
2. figure out what crazy pivots are causing the problem
3. Refactor the pivot OR determine what dimensionality is missing from the cube

Once you go a day or so w/ a quiet log table, you can turn on the query killer and go on your merry way. It's been suggested that we should send the user an email - let them know that it wasn't an error... that we cancelled their query. Maybe - but honestly, you should probably email or call them personally to hash it out. If it happens once, chalk it up to happenstance. If it continues to occur, it's time for some user training and/or requirements gathering.

The real problem is that most end users don't know that 30 seconds is ridiculously long - they shouldn't have to wait that long. Not only does it hurt them, it hurts others using the system as well.

UPDATE: You may want to, at some point, actually view one of these monster queries to see what on earth they were querying for. Over a certain size, getting it out of SSMS can be pretty troublesome - A query similar to the following should do the trick:

DECLARE @SQLcommand VARCHAR(4000)
SET @SQLcommand = 'bcp "select CommandText 
from Database.dbo.LongRunningMDXQueryLog 
where SPID = 511519" 
queryout "c:\output.dtsx" -T -c'
 
EXEC xp_cmdshell @SQLcommand

Here's our starting point in the Adventure Works cube (note, I commented out all the scope statements in the MDX script that distribute quarterly quotas/targets to months - in AW, quotas only exist in the data warehouse at the calendar quarter level):

Let's start with something simple - Equal Allocation:


Now let's try to figure out this Weight Expression thing by throwing a 1 in there:

Now discard previous changes and enter the quota at the Semester level again:

Ok, that wasn't what we intended - but it does kind of explain what it's up to. That's the number of detail cells (leaves) that it wrote data to - and it wrote the entire quota amount to each cell, then added them all up. Note that the first leaf got the true-up balance - if your custom aggregation doesn't roll up properly, it seems that it figures out how much you were off and tosses whatever number is required to make it aggregate right in to the first leaf.

Now to get the right weight expression, we have to divide 1 (the entire quota) by the number of leaves that it's writing data to. Sales quotas are related to the Date dimension, the Sales Territory dimension, and the Employee dimension, so it's going to write out one value for every intersection of the leaves of those dimensions (at whichever level the measure group is related to the dimension, so it doesn't write out a value for every date, it writes out a value for every quarter - because that's the level that quotas are associated with the date dimension, calendar quarter). So we basically want:

1 / [number of employees] / [number of sales regions] / [number of quarters]

Here's the weight expression to accomplish that (note that we assume that we know what we're trying to do here: enter quotas at the fiscal semester level. That's why the following expression gets the Semester ancestor of the current fiscal calendar member, because we know that we want to distribute evenly across all quarters and we entered data at the semester level)

1
/Count(
	Descendants(
		 [Employee].[Employee].[All]
		,[Employee].[Employee].[Employee]
	)
 )
/Count(
	Descendants(
		 [Sales Territory].[Sales Territory Region].[All]
		,[Sales Territory].[Sales Territory Region].[Sales Territory Region]
	)
 )
/Count(
	Descendants(
		 Ancestor(
			 [Date].[Fiscal].CurrentMember
			,[Date].[Fiscal].[Fiscal Semester]
		 )
		,[Date].[Fiscal].[Fiscal Quarter]
	)
)

And, of course, we end up with our old friend Equal Allocation:

Now say you wanted to enter quotas at the intersection of the Fiscal Year and the Country. Now we don't want all the sales regions anymore, we just want the Descendants of the current sales region's Ancestor at the Country level:

We start here:

And enter the following Weight Expression:

1
/Count(
	Descendants(
		 [Employee].[Employee].[All]
		,[Employee].[Employee].[Employee]
	)
 )
/Count(
	Descendants(
		 Ancestor(
			 [Sales Territory].[Sales Territory].CurrentMember
			,[Sales Territory].[Sales Territory].[Country]
		 ) 
		,[Sales Territory].[Sales Territory].[Region]
	)
 )
/Count(
	Descendants(
		 Ancestor(
			 [Date].[Fiscal].CurrentMember
			,[Date].[Fiscal].[Fiscal Year]
		 )
		,[Date].[Fiscal].[Fiscal Quarter]
	)
)

And then we enter the quota at the intersection of Fiscal Semester and Country:

So all we've done this far is reproduce numbers similar to what Equal Allocation would get us. But what did we just do really? We distributed the sales quotas across all employees - but only Sales People have sales quotas... oops (the following is sliced by [Employee].[Sales Person Flag] on columns):

And now we finally get to why we want to use Weight Expressions in the first place. We only want to distribute quotas across sales people (note that we've went back to just distributing evenly across all sales regions again for simplicity):

iif([Employee].[Employee].CurrentMember.Properties("Sales Person Flag") = "True"
,
	1
	/Count(
		Descendants(
			 [Employee].[Employee].[All]
			,[Employee].[Employee].[Employee]
		) * [Employee].[Sales Person Flag].&[True]
	 )
	/Count(
		Descendants(
			 [Sales Territory].[Sales Territory Region].[All]
			,[Sales Territory].[Sales Territory Region].[Sales Territory Region]
		)
	 )
	/Count(
		Descendants(
			 Ancestor(
				 [Date].[Fiscal].CurrentMember
				,[Date].[Fiscal].[Fiscal Semester]
			 )
			,[Date].[Fiscal].[Fiscal Quarter]
		)
	)
,Null)

And the results:

Great, we got that particular slice all fixed up - but what about if we slice by employee? Did we distribute sales quotas for the US across sales people that don't even work in the US? What-If analysis is just that - "what happens if I do this?" If you're never going to look at more detailed slices, it's fine the way it is. If you're using writeback to publish data to the cube and let others slice on it all willy-nilly, you need to examine all the slices you just wrote data to and make sure your weight expression allocates the data the way you intended it to.

In the end, do we expect business users to be able to write their own weight expressions, publish to the masses, and understand what exactly just happened? I doubt it. That's a bit much to ask. We can, however, create accurate weight expressions for business users if we know exactly where they want to enter the new data.

"Excel freezes (or hangs) when I try to change this filter"

Most of us have seen this at one point or another and shrugged it off as a busy time or processing is going on or there are cats clogging up the tubes, etc. Tonight, I finally decided to figure out what's causing it.

When you have an OLAP cube filter in your pivot table and expand the items list to select your filter items, Excel fires a bunch of Discover commands at the cube (these are pretty light in most cases). If you don't have anything selected yet (so you just have the default member selected) Excel isn't doing much work here and the list pops right up. Here's two basic flow examples for what discover commands are fired off when you DO have an item selected:

If you have one level 1 member selected

1. MDSCHEMA_CUBES (whatever cube you're connected to)
2. MDSCHEMA_CUBES (again, same thing)
3. MDSCHEMA_MEMBERS (LEVEL_UNIQUE_NAME of the filter), no TREE_OP
4. MDSCHEMA_MEMBERS (MEMBER_UNIQUE_NAME of the selected member) TREE_OP 4, only immediate parent
5. MDSCHEMA_MEMBERS (MEMBER_UNIQUE_NAME of the selected member's parent) TREE_OP 8, returns itself
6. MDSCHEMA_MEMBERS (MEMBER_UNIQUE_NAME of the selected member's parent) TREE_OP 1, returns all children

If you have one level 2 member selected in a hierarchy

1. MDSCHEMA_CUBES (whatever cube you're connected to)
2. MDSCHEMA_CUBES (again, same thing)
3. MDSCHEMA_MEMBERS (LEVEL_UNIQUE_NAME of the filter), no TREE_OP
4. MDSCHEMA_MEMBERS (MEMBER_UNIQUE_NAME of the selected member) TREE_OP 4, only immediate parent
5. MDSCHEMA_MEMBERS (MEMBER_UNIQUE_NAME of the selected member's parent's parent) TREE_OP 4, immediate parent
6. MDSCHEMA_MEMBERS (MEMBER_UNIQUE_NAME of the selected member's parent's parent) TREE_OP 8, returns itself
7. MDSCHEMA_MEMBERS (MEMBER_UNIQUE_NAME of the selected member's parent's parent) TREE_OP 1, all children
8. MDSCHEMA_MEMBERS (MEMBER_UNIQUE_NAME of the selected member's parent) TREE_OP 8, returns itself
9. MDSCHEMA_MEMBERS (MEMBER_UNIQUE_NAME of the selected member's parent) TREE_OP 1, returns all children

So, that's six requests for a level 1 member and 9 requests for a level 2 member - both of these take milliseconds to complete the whole flow, so no big deal. When does this become a big deal? When you do something like click the "Select multiple items" checkbox, expand a hierarchy and UNCHECK something... like if you opened the Adventure Works 2008 cube, added Delivery Date.Date as a filter, expanded the "All Periods" level and deselected the first day:

Now click OK and expand the items box - we get a familiar popup:

It comes back after a few seconds with the expected item list, but why did it take so long? Well, based on our flows above, we have a whole bunch of level 1 members selected - 1187 of them to be exact. That means there will be 1187 requests for their parent member (TREE_OP 4) which will all return the [All Periods] member. To illustrate this, I traced that exact operation:

The profiler captured 1192 discover requests - 1187 of them were getting the parents of selected members, two were MDSCHEMA_CUBES requests, one was getting [All Periods], one was getting all the children of [All Periods], and one was getting the Unique Name of the root level. Not cool. Imagine what happens when your talking about tens of thousands of members at a particular level! Granted, that seems unlikely but understand that the logic behind this is "I don't want to see this", "hide just this member", and so forth - in some cases people have no idea what actually happened was they unchecked one of 6,000 members.

So what to do about this? Educate. To fix the immediate problem of a seemingly broken Excel pivot table - just wait for the filter to load, it'll come back eventually. In the future when working with Excel filters, always favor selecting vs. deselecting and filtering at higher levels in the hierarchy whenever possible.

So we went with a simple, albeit relatively crude (but effective), alternative - implementing a URL action for Cells so users can easily link out to a definition of the measure they're looking at.

Create a new action in your cube (Open up the cube definition, Actions tab) and configure similar to this:

Name: View Member Definition
 
Action Target
    Target Type: Cells
    Target object: All cells
 
Action Content
    Type: URL
    Action expression: "http://i.domain.com/doc/Defs.aspx#" 
                       + [Measures].CurrentMember.Name
 
Additional Properties
    Invocation: Interactive
    Description: View Member Definition
    "View Definition Of " + [Measures].CurrentMember.Name + "..."
    Caption is MDX: True

When you're finished, it should look something like this:

For a possible way to implement the aforementioned Definitions.aspx, check out http://timlaqua.com/2009/03/scrolling-to-and-highlighting-anchor-target-via-javascript/ - which describes an early endpoint we used for this project.

<div id="dataDictionary">
    <div id="termNotFound">Specified definition could not be found.</div>
 
    <div>
        <h1>Term goes here</h1>
        <p>Definition goes here</p>
        <h2>Additional Comments</h2>
        <ul>
            <li>comment</li>
        </ul>
    </div>
    <div>
        <h1>Other Term goes here</h1>
        <p>Other Definition goes here</p>
        <h2>Additional Comments</h2>
        <ul>
            <li>comment</li>
        </ul>
    </div>
</div>

And the following JavaScript is used to locate the appropriate heading, modify it's parent DIV (to add the selected className so it gets different styling), and scroll down to it:

ParseURI function from http://blog.stevenlevithan.com/archives/parseuri

function parseUri (str) {
	var	o   = parseUri.options,
		m   = o.parser[o.strictMode ? "strict" : "loose"].exec(str),
		uri = {},
		i   = 14;
 
	while (i--) uri[o.key[i]] = m[i] || "";
 
	uri[o.q.name] = {};
	uri[o.key[12]].replace(o.q.parser, function ($0, $1, $2) {
		if ($1) uri[o.q.name][$1] = $2;
	});
 
	return uri;
};
 
parseUri.options = {
	strictMode: false,
	key: ["source","protocol","authority","userInfo","user","password","host","port","relative","path","directory","file","query","anchor"],
	q:   {
		name:   "queryKey",
		parser: /(?:^|&)([^&=]*)=?([^&]*)/g
	},
	parser: {
		strict: /^(?:([^:\/?#]+):)?(?:\/\/((?:(([^:@]*):?([^:@]*))?@)?([^:\/?#]*)(?::(\d*))?))?((((?:[^?#\/]*\/)*)([^?#]*))(?:\?([^#]*))?(?:#(.*))?)/,
		loose:  /^(?:(?![^:@]+:[^:@\/]*@)([^:\/?#.]+):)?(?:\/\/)?((?:(([^:@]*):?([^:@]*))?@)?([^:\/?#]*)(?::(\d*))?)(((\/(?:[^?#](?![^?#\/]*\.[^?#\/.]+(?:[?#]|$)))*\/?)?([^?#\/]*))(?:\?([^#]*))?(?:#(.*))?)/
	}
};

ScrollToElement function from http://radio.javaranch.com/pascarello/2005/01/09/1105293729000.html

function ScrollToElement(theElement){
 
  var selectedPosX = 0;
  var selectedPosY = 0;
 
  while(theElement != null){
    selectedPosX += theElement.offsetLeft;
    selectedPosY += theElement.offsetTop;
    theElement = theElement.offsetParent;
  }
 
 window.scrollTo(selectedPosX,selectedPosY);
 
}

Some quick string manipulation functions form http://www.webtoolkit.info/javascript-trim.html

function trim(str, chars) {
	return ltrim(rtrim(str, chars), chars);
}
 
function ltrim(str, chars) {
	chars = chars || "\\s";
	return str.replace(new RegExp("^[" + chars + "]+", "g"), "");
}
 
function rtrim(str, chars) {
	chars = chars || "\\s";
	return str.replace(new RegExp("[" + chars + "]+$", "g"), "");
}

And finally, some original code to make it all work:

function hotlink() {
	var colP = document.getElementsByTagName('p');
	if ( colP.length > 0 ) {
		for(var i=0; i<colP.length; i++) {
			colP[i].innerHTML = colP[i].innerHTML.replace(new RegExp("(" + ddTerms + "^$)","g"),"<a href=\"javascript:findTerm('$1')\"'>$1</a>");
		}
	}
}
 
function findTerm(term) {
	var colDIV = document.getElementsByTagName('div');
	if ( colDIV.length > 0 ) {
		for(var i=0; i<colDIV.length; i++) {
			if (i % 2 == 0) {
				colDIV[i].className = 'alt';
			}
			else {
				className = '';
			}
		}
	}
 
	var colH1 = document.getElementsByTagName('h1');
	if ( colH1.length > 0 ) {
		var found = false;
		for(var i=0; i<colH1.length; i++) {
			var ddTerm = trim(colH1[i].innerHTML.toLowerCase());
			if (ddTerm == term.toLowerCase()) {
				found = true;
				colH1[i].parentNode.className = 'selected';
				ScrollToElement(colH1[i]);
			}
			ddTerms += "\\b" + ddTerm.replace(new RegExp("(\\(|\\))", "g"), "\\$1") + "\\b|";
		}
 
		if (!found && term.length > 0) {
			var tnf = document.getElementById('termNotFound');
			tnf.style.display = 'block';
			ScrollToElement(tnf);
		}
	}
}
 
var ddTerms = '';
 
window.onload = function () {
	findTerm(unescape(parseUri(window.location.href).anchor));
	hotlink();
}

Note, there's also a hotlink() function in there - since this is a page full of terms and definitions, we might as well link up the page so all terms are clickable within the document.

Finally, make it looke somewhat presentable with either an inline style block or just add these to your external style sheet:

	#dataDictionary a { font-style: italic; text-decoration: underline; }
	#dataDictionary 
	{
		font-size: 110%;
	}
 
	#dataDictionary div 
	{ 
		background-color: #ccccff; 
		margin-bottom: 4pt;
	}
 
	#dataDictionary div p
	{ 
		margin: 2pt;
	}
 
	#dataDictionary div ul
	{ 
		margin-top: 2pt;
	}
 
	#termNotFound 
	{ 
		display: none; 
		color: red; 
		font-weight: bold; 
		font-style: italic; 
		border: 2px dotted red;
		background-color: #ffff77 !important;
		padding: 2pt;
	}
 
	#dataDictionary div h1 
	{ 
		font-size: 130%; 
		border-bottom: 1px solid white; 
		background-color: #ccbbff;
		margin: 0pt, 0pt, 2pt, 0pt;
	}
	#dataDictionary div h2 { 
		font-size: 115%; 
		margin: 3pt, 0pt, 2pt, 0pt;
	}
 
	#dataDictionary div ul 
	{
		margin-bottom: 2pt;
	}
 
	#dataDictionary div.alt  h1
	{ 
		background-color: #88ff88; 
	}
	#dataDictionary div.alt 
	{ 
		background-color: #bbffaa; 
	}	
 
	#dataDictionary div.selected h1 
	{ 
		background-color: #ff7755; 
	}
	#dataDictionary div.selected 
	{ 
		background-color: #ff9988; 
		font-size: 125%;
	}

This isn't really a problem as in SSRS, you can simply define the Category Group for the Series in your chart (ya, I didn't mention that yet - this is all for a chart) as DatePart("w", Fields!DateTime) - which will group everything by the number of the given weekday and then you can just have your "This Week" and "Prevoius Week" series.

WITH 
  MEMBER [Measures].[This Week] AS 
    Aggregate
    ([Time].[Calendar Date].LastChild:[Time].[Calendar Date].LastChild.Lag(6)
     ,[Measures].[Revenue])
 
  MEMBER [Measures].[Previous Week] AS 
     Aggregate([Time].[Calendar Date].LastChild.Lag(7)
     :
     [Time].[Calendar Date].LastChild.Lag(13)
     ,[Measures].[Revenue])
SELECT 
  NON EMPTY 
    {[Measures].[Previous Week], [Measures].[This Week]} ON 0
 ,NON EMPTY 
    { 
        [Time].[Date Time].[Date Time].MEMBERS 
        * 
        [Time].[24 Hour].[24 Hour].ALLMEMBERS
    } ON 1
FROM 
(
  SELECT 
    Tail(
         Filter(
               [Time].[24 Hour].[24 Hour].MEMBERS
               ,Sum([Time].[Calendar Date].LastChild,[Measures].[Revenue]) > 0
         ), 1
    ) ON 0
  FROM 
  (SELECT 
      {
            [Time].[Calendar Date].LastChild
            :
            [Time].[Calendar Date].LastChild.Lag(13)
      } ON 0
    FROM [Cube]
  )
)

And to top it all off - after seeing the chart for a while, nobody liked it (after all, it's a relatively abstract view of revenue) and we dumped it.

And the parts that I have to keep looking up (I use my blog as a notebook for things I don't want to forget :

Sub AddCalculatedMeasure() 
	Dim pvt As PivotTable
	Dim strName As String
	Dim strFormula As String 
 
	Set pvt = Sheet1.PivotTables("PivotTable1")
	strName = "[Measures].[Internet Sales Amount 25 %]"
	strFormula = "[Measures].[Internet Sales Amount]*1.25"
	pvt.CalculatedMembers.Add Name:=strName, Formula:=strFormula, Type:=xlCalculatedMember 
End Sub 
 
Sub AddCalculatedMember() 
	Dim pvt As PivotTable
	Dim strName As String
	Dim strFormula As String 
 
	Set pvt = Sheet1.PivotTables("PivotTable1")
	strName = "[Product].[Product Categories].[Bikes].[Mountain Bikes].[Mountain-100 Silver, 38 25 %]"
	strFormula = "[Product].[Product Categories].[Bikes].[Mountain Bikes].[Mountain-100 Silver, 38]*1.25"
	pvt.CalculatedMembers.Add Name:=strName, Formula:=strFormula, Type:=xlCalculatedMember
	pvt.ViewCalculatedMembers = True 
End Sub 
 
Sub AddNamedSet() 
	Dim pvt As PivotTable
	Dim strName As String
	Dim strFormula As String
	Dim cbf As CubeField 
 
	Set pvt = Sheet1.PivotTables("PivotTable1")
	strName = "[My Mountain Bikes]"
	strFormula = "[Product].[Product Categories].[Bikes].[Mountain Bikes].children"
	pvt.CalculatedMembers.Add Name:=strName, Formula:=strFormula, Type:=xlCalculatedSet
	Set cbf = pvt.CubeFields.AddSet(Name:="[My Mountain Bikes]", Caption:="Mountain Bikes") 
End Sub

He also mentions that you can expose these members to Excel Services 2007 by creating the new objects and then removing the VBA code - very useful article.

There are two places these references exist (afaik) - The My Data Sources folder (Windows XP / Excel 2007) which holds all of the odc objects for the logged in user, and the real pain - Embedded data sources in Excel Workbooks.

For the odc objects - you can just loop through the My Data Sources folder files and do a little find/replace action.

For the Excel files - we need to first find all the files, then loop through them and hunt for embedded Connections. Once we find a connection, we can just blindly replace occurrences of the old server name with the new server name. The first thing we need here is a function we can call recursively to go hunt down all the excel files - then we loop through and play with the connections (when found).

Let me know how it goes! Here's the full version:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88

Set regEx = New RegExp
regEx.IgnoreCase = True
regEx.Global = True
Set WshShell = Wscript.CreateObject("Wscript.Shell")
strMyDocsPath = WshShell.RegRead("HKCU\Software\Microsoft\Windows\CurrentVersion\Explorer\Shell Folders\Personal")
strDataSourceDir = strMyDocsPath & "\My Data Sources"
 
'======================== Fix all the Data Sources
Set objFSO = CreateObject("Scripting.FileSystemObject")
Set Folder = objFSO.GetFolder(strDataSourceDir)
Set Files = Folder.Files
 
For Each File In Files
	If LCase(Right(File.Name, 4)) = ".odc" Then
	    Set objReadFile = objFSO.OpenTextFile(File.Path, 1)
	    strContents = ""
 
		Do While Not objReadFile.AtEndOfStream
			strContents = strContents & objReadFile.ReadLine & vbCrLf
		Loop
 
	    objReadFile.Close
 
		regEx.Pattern = "(?=.*Data Source=OldCrappyServerName.*)(<odc:ConnectionString>.+</odc:ConnectionString>)"
 
		If regEx.Test(strContents) Then
			Set colMatches = regEx.Execute(strContents)
 
			For Each objMatch in colMatches
				strOldConnectionString = objMatch.SubMatches(0)
				regEx.Pattern = "Data Source=OldCrappyServerName"
				strNewConnectionString = regEx.Replace(strOldConnectionString, "Data Source=NewAwesomeServerName")
				regEx.Pattern = strOldConnectionString
				strContents = regEx.Replace(strContents, strNewConnectionString)
			Next
 
			Set objWriteFile = objFSO.CreateTextFile(File.Path, True)
			objWriteFile.Write strContents
			objWriteFile.Close
		End If
	End If
Next
 
 
'================================ Fix all the embeded connections (yikes)

Set arrFileList = CreateObject( "System.Collections.ArrayList" )
Set objExcel = CreateObject("Excel.Application")
objExcel.DisplayAlerts = False
 
populateExcelFileList(strMyDocsPath)
 
For Each strFileName In arrFileList
	objExcel.Workbooks.Open(strFileName)
	Set objWorkbook = objExcel.Workbooks(1)
	For Each objWorkbook in objExcel.Workbooks
		If objWorkbook.Connections.Count > 0 Then
			WScript.Echo strFileName
			For Each objConnection in objWorkbook.Connections
				strOldConnection =  objConnection.OLEDBConnection.Connection
				regEx.Pattern = "OldCrappyServerName"
				strNewConnection = regEx.Replace(strOldConnection, "NewAwesomeServerName")
				objConnection.OLEDBConnection.Connection = strNewConnection
			Next
 
		End If
 
		objWorkbook.Close True
	Next
 
	objExcel.Quit
Next
 
Function populateExcelFileList(path)
	Set pathFolder = objFSO.GetFolder(path)
	Set pathFiles = pathFolder.Files
	For Each pathSubFolder in pathFolder.SubFolders
		populateExcelFileList(pathSubFolder.Path)
	Next
 
	For Each pathFile in pathFiles
		If (LCase(Right(pathFile.Name, 5)) = ".xlsx" _
			Or LCase(Right(pathFile.Name, 4)) = ".xls") _
			And Left(pathFile.Name, 1) <> "~" Then
			arrFileList.Add pathFile.Path
		End If
	Next
End Function

Update
I added in support for .xls files - doing this against .xls files can (and usually will) break backwards compatibility for SSAS Pivot Tables.