IIQ AD Updater Technical Design

The purpose of this project is to replace the old AD Daily Updater with newer better code residing in an IIQ instance. However, the future plan to support Exchange Online routing of Email (Mail Relay Replacement) adds and additional set of requirements.

Issues

This project replaces

1. The AD Daily Updater, a Java program driven by the Oracle ACS1 database, which populated a select set of "biographical" fields in the AD for non-private people. None of these fields were important from a systems point of view, except for UPN which is an identifier to login to services.
2. The Mail Relay configuration, which maps Email Aliases (ending in "@yale.edu") to the native email addresses of Yale systems. Originally this meant "Netid@connect.yale.edu" for Exchange accounts and "Netid@pantheon.yale.edu" for IMAP. Current Exchange has moved to O365 and IMAP has been replaced by Eliapps.
3. The Exchange mail routing fields in AD (MailNickName and ProxyAddresses). When mail was sent to Exchange, it delivered mail addressed to various Email Aliases using data stored in two fields of the AD. A separate Exchange field management function in the AD Daily Updater maintained these two Exchange Mail Routing fields.

This new project continues to support the biographical fields of AD, but now gets the data from IIQ. It updates the logic of function 3) to reflect differences between the original on premise Exchange and the O365 Azure AD requirements, and merges the function 2) code in by populating the MailNickname, ProxyAddresses, and now Target Address field of an AD object for Eliapps and legacy departmental Email Aliases.

We have to plan ahead for expected changes that do not currently have a schedule:

• The on premise Exchange system serves no useful purpose. It should be decommissioned. This would involve removing certain Exchange related fields from the on premise AD (fields that begin with the "msExch" prefix. The old AD Daily Updater code depends on these fields, but the new replacement code will be written to work if they are null. The one exception is that Private users are excluded from the Outlook directory with two of these fields and that function still needs to be supported.
• The Mail Relay machines will at some point be decommissioned and the Exchanged Online function in the Azure Cloud will take over routing all mail addressed to an "@yale.edu" Alias. Since Excahnge Online is configured by Azure AD in essentially the same way that on premise Exchange was configured by on premise AD, we have to add AD Email Aliases to the ProxyAddresses list and the native Email address (MAILBOX) of the aliases to the TargetAddress field for existing User objects or new Contact objects to be created for the purpose.

Currently mail addressed to an "ALIAS_NAME@yale.edu" Email address goes to the Mail Relay machines. They look up the ALIAS_NAME in configuration data and generate a native Email address that in practice sends the mail to O365 or Eliapps (although there are a few legacy entries that may go to deparmental standalone mail servers). If a Primary Email Alias points to a native Email Address ending in "@bulldogs.yale.edu" then the mail goes to Eliapps.

However, when the Mail Relays go away and are replaced by Exchange Online, then what was once a two tiered processing becomes a single step, and Exchange gets first dibs on the mail. Because Exchange Online puts its configuration data in the Azure/Local AD, some rules about the way AD has to work can conflict with specific settings in the Email Alias table. Fortunately, the current configurations that will become invalid are also in violation of Yale policy. This will not, however, avoid the requirement to communicate the change to affected users.

There are a set of Yale and Exchange rules:

• It is an AD rule that the UserPrincipalName must be unique.
• It is a Yale rule that for Users with a mail account, the Primary Email Alias (first.last@yale.edu) also be the UserPrincipalName and its prefix is the MailNickName (first.last). Users marked "Private" and Users without mail get a UPN of "Netid@yale.edu". These are the only two possible values.
• It is a Yale rule that everyone with an ITS mail account (O365 or Eliapps) must have a Mail Alias that points to that account. (Everyone with any Mail Alias is supposed to have a Primary Alias, but that rule has been broken and the data would have to be cleaned up to enforce it.)
• It is an Exchange Online rule that the UPN has to be in the ProxyAddress list of a mail User object (that is, the UPN has to be a deliverable mail address to anyone with an O365 mail account).
• It is an Exchange Online rule that the same Email address cannot be in two ProxyAddress lists of two Azure AD objects.

Yale policy is to give users either O365 or Eliapps mail but not both, but exceptions are allowed. Yale policy is that when a user has both O365 and Eliapps mail, the O365 account is Primary. There are a small number of case where that is not true, but when you combined all the rules we will no longer be able to allow exceptions. Exchange Online will force the O365 account to be Primary even if the Email Alias table says that Eliapps is supposed to be Primary.

Except when one of these structural rules forces a change, the objective of this design is to deliver mail with the same Email address to the same mailbox it was delivered to with the old Updater and the old Mail Relays. That is, changes may be forced by structural differences but they will not happen by coding accidents.

It should generally be possible to accomplish exactly the same mail delivery for all addresses, but this may require a slightly complicated rearrangement of Aliases and the AD. For example, you can rename AD objects so that an O365 Mail account that previously belonged to a personal Netid now appears to belong to a dependent Netid, and point the old alias to it. This frees the personal Netid User object to point to Eliapps, while the O365 account, now on the dependent Netid, continues to get mail through all the old addresses. However, this is a very delicate operation and will not be done for any but the most important users.

Taking all of these requirements into consideration, data from the Alias table will be arranged by a database view to create one of four situations:

1. Unchanged User object - Netids with Primary O365 accounts and no Eliapps account (most employees, including faculty). The Mail Routing fields are populated by entries in the Alias table that have a MAILBOX value of "Netid@connect.yale.edu" (the formal address of an O365 account).
2. User object with Eliapps routing - Netids with a Primary Eliapps account and no O365 Mail (although non-mail O365 licenses) will have the mail routing data in the Alias Table turned into fields of the AD User object. The MAILBOX column value from the Primary Alias row is used to find the Secondary Aliases of the same account, and it also becomes the TargetAddress. The matching column values that share that MAILBOX will become ProxyAddresses.
3. Unchanged User object and Contact with Eliapps Routing - Users with a Primary O365 account and a Secondary Eliapps account will find their User object unchanged, but a new Contact object will be created with the same TargetAddress and ProxyAddresses values described in the previous case, but now applied to the new Contact
4. Routing "Junk" Contact objects - Entries in the Email Alias table that were not processed to generate a ProxyAddress in one of the previous three steps are grouped by unique MAILBOX value. Each MAILBOX value creates a "junk" Contact whose ProxyAddresses are the list of ALIAS_NAMEs that had that MAILBOX value. Note that this method of collection may combine Aliases owned by different Netids if they point to the same MAILBOX, but this is OK.

There is a fifth category of unprocessed aliases. Any Alias with a MAILBOX of "xxxx@connect.yale.edu" where the prefix xxxx is not a Netid must be a resource (a room, distribution list, etc.) and will not be processed. Exchange Online knows these things and handles them natively, so we don't have to create objects to inform Exchange about its own native stuff.

Any user can have Junk, so while each mail user Netid is either 1, 2, or 3, they can also have any number of Type 4 entries. Since Type 4 is grouped by MAILBOX and not owner, several Netids can share a Type 4.

Users with an O365 or Eliapps account can delete it, and users with one type of account can get the other. So you can transition between Type 1 and 3, or 2 and 3. This is a delicate process because IIQ cannot directly move a field like ProxyAddresses from a User object to a Contact object in a single transaction (as would be possible with database SQL). This is, however, a manual process and is outside the scope of the current design.

There is bad data in the Alias table which will have to be cleaned up. For example, some O365 users have a MAILBOX value that is not "Netid@connect.yale.edu" and alternate names for that MAILBOX have to be changed to the proper value in order for the grouping to work correctly. Bad data that represents aliases to mailboxes that no longer exist may or may not be cleaned out (the mail would have bounced before and will bounce after, although the exact way it bounces may change).

We synchronize the on premise Local AD to the Azure AD. IIQ must provision changes to the Local AD and wait for the synchronization to occur within the next half hour. The only field in Azure AD that can be provisioned directly is UPN.

There are 1800 Junk Email Aliases that do not resolve to O365 or Eliapps. They are spread across mailman, elilists, panlists, cs, som, aya, invest, med, physics, chem, geology, cmp, math, astro (all .yale.edu). Some of this is obsolete, but cleaning it up is future scope.

There are existing Contacts created in AD to make specific Eliapps users visible to Outlook. These "non-routing" Contacts have a biographical (non routing) "mail" attribute and not much else. We will not create or delete these objects, and we will attempt to make Routing Contact objects invisible to Outlook.

Assignment of Functions to Instances

The update of biographical fields (firstname, lastname) will occur in the Identity Management instance of IIQ and will be driven by changes to Identity variables. The update of mail routing fields will be done in the Email Provisioning instance of IIQ.

For sanity, we try to ensure that both instances of IIQ share a common set of defined Identity Variables, but we do not guarantee that these Variables are actually populated with values in an instance that does not use that particular variable in any code. The two instances will source the variable value from different places, and they MUST in general partition the Target (Application and field) where values are provisioned. We do not want to get into even temporary chase conditions where one Instance sets a field to one value and the other instance sets it back to the previous value due to different schedules in the aggregation.

There is a UserPrincipalName in both Local AD and Azure AD. It is identical to the Primary Email Alias, and the rules constrain it to interact with Mail Routing (it must be in the ProxyAddresses list of an O365 User), but used as a UPN is it a string used to log you in and is therefore like the "biographical" fields. However, it is more closely tied to the other fields used by Mail Routing (it is sourced, for example, from the Alias table, and so we regard it as a Mail Routing field and provision it along with the real Mail Routing fields.

Identity Management Instance

Additional Applications (data sources) required:

• LocalAD Users (copy existing application from Email Provisioning instance)

Email Provisioning Instance

Additional Applications (data sources) required:

• Dependent Netids (copy existing application from Identity Management instance)
• Full Email Alias view (there is an existing Application named Email but it only aggregates Primary Aliases)
• Local AD Contacts (provisioned, may become a source of identities)

Provisioning Strategy (Biographical data)

As is currently done in Email Provisioning, Identity Management will have a Role with an Entitlement to membership in the Provisioned Users group of LocalAD. Assigning an Identity to the Role will create a new AD User object for that Netid if one does not already exist. The User object is initially created with minimal information, and is then filled in with the full set of biographical data in the next cycle.

This application adds a set of Identity Variables named ADXXXX (example: ADgivenName) are filled in by a Rule run after an Identity Refresh task. Normally Identity Variables are filled in after new data is read in from a source system, but the ADXXXX variables are assigned values from other Identity Variables that were themselves populated from source systems. For example, your Job Title comes in from Workday Worker and is used to populate a previously configured Identity Variable which then used the raw title for various Yale systems. However, the ADtitle variable is constrained to be a maximum of 128 characters because that is the maximum field length in AD. So at Yale there are currently 51 people with longer Job Titles in Workday Worker and other Yale systems, but we truncate that to 128 when we create the ADtitle variable that provisions the Title property in AD.

There are also restrictions due to Privacy rules for certain people, and in a few cases there are multiple possible source fields for the same AD variable, and the RefreshRule will select the right source based on Affiliation and other attributes.

The Netid System is the source for Identities created for Dependent Netids. It does not have any biographical data, so the source fields are all null, and the ADXXXX variables are all null, and the AD gets no biographical information.

Biographical ADXXXX variables have a LocalAD Target field only in the Identity Management instance. These Identity Variables may be filled in by the RefreshRule running in the Email Provisioning instance of IIQ, but they will not provision AD User objects from that instance.

Provisioning Strategy (Email Routing fields)

The Email Routing ADXXXX variables are calculated from the Account generated by aggregating the Full Email Alias table, an application that is only defined in the Email Provisioning instance. They also only have Target definitions and only update the LocalAD in that instance. The same variables have neither a source nor a target in Identity Management.

Because ADProxyAddresses is a multivalued field, we want to avoid updating it if the only difference between the current value in AD and the calculated correct value from the Email Alias table is to list the same values in a different order.

Dependent Netids and Contacts can own or describe Email accounts, so they do have ADMailNickName, ADProxyAddresses, and ADTargetAddress fields that must be calculated and, if necessary, provisioned.

However, correlating the Email Alias data to the right Identity cannot be done on the fly in IIQ. We rely on a database view to present the data in the right order. So most of the work is done in the database, and a little work may be done in the correlation rule.

Unlike the previous AD Daily Updater Java program, IIQ will create AD User objects for Identities that need them.

Mail Relay Replacement

Currently all mail with an address ending in "@yale.edu" is processed by the Mail Relay machines at Yale. They are configured (once an hour) from the data in the Mail Alias system, where is unique MailAliasName+"@yale.edu" is resolved to a native Mail system address (the Mailbox). If the Mailbox is "netid@connect.yale.edu" then this is an O365 address to be processed by Exchange Online. If it is "@bulldogs.yale.edu" then it is an Eliapps account to be processed by Google. In all other cases, the Mailbox address substitutes for the original "@yale.edu" and is processed through SMTP.

The Mail Relay function is to be retired and all mail will now pass through Exchange online. Essentially, we don't have to do anything differently for O365 accounts because that mail already passes through Exchange and is delivered correctly. So the change is for Eliapps and other addresses.

Exchange Online is configured by data in the Azure AD. So the problem is to use the current contents of the Email Alias table (DIR_ONLINE_INFO) to configure (Azure) AD objects so the mail will be delivered the correct way. Exchange has a concept called MailNickName which is not standard and can be ignored at this point. Otherwise, Exchange Online looks at any "xxxx@yale.edu" address and tries to match it to exactly one "smtp:xxxx@yale.edu" entry in the ProxyAddresses list of an AD User or Contact object. If it matches and this is not an O365 account, then it delivers mail based on the TargetAddress field of the matched object.

So the requirement is simple:

Every AliasName in the Alias table (prefixed with "smtp:" and suffixed by "@yale.edu") must be in exactly one AD User of Contact ProxyAddress list and

The TargetAddress of the matched User or Contact must be set to the Mailbox value of that entry in the Alias table.

Today we populate the ProxyAddress list of the User object of everyone who has an O365 account with an entry for every Alias Table row that has a Mailbox value of "netid@yale.edu".

So the new step is to do essentially the same thing for Eliapps mail users. If they do not have an O365 account, then we can set the TargetAddress of their AD User object to "EliappsAccount@bulldogs.yale.edu" and then populate their ProxyAddresses with every Email Alias with a Mailbox that points to that account (exactly what we do now for O365 users). If someone has both an O365 and an Eliapps account, then the O365 account has to go in the User object, so we have to create a second Contact object for the Eliapps ProxyAddresses list and TargetAddress.

Then for any other Alias (not O365 or Eliapps) the one Contact object per row in the Alias table works fine.

We accomplish this by aggregating Alias entries by Netid, and then for each Netid

• If there is any Alias with an O365 MAILBOX, then all the Alias names with that MAILBOX value generate an smtp:AliasName@yale.edu entry in the User object (the PrimaryAlias is capitalized SMTP:), the TargetAddress of the User object is set to some O365 value, and now the User object is full and any other aliases have to go into Contacts.
• Starting with the Primary, if there is any Alias with a MAILBOX ending in "@bulldogs.yale.edu" then that Mailbox value becomes a TargetAddress and all aliases with the same MAILBOX value populate a ProxyAddresses list in the same way. If there was no O365 account, these go in the User object. If not, they populate Identity Variables that will generate a Contact object.
• Create one Contact object for EACH remaining Alias (that are not O365 or Eliapps) with one ProxyAddress equal to the AliasName and a TargetAddress equal to the MAILBOX. We do not attempt to combine entries that resolve to the same MAILBOX because these types of Aliases are almost always unique.

Unfortunately, the existing Alias table has some Netids that have a mixture of personal and departmental Eliapps accounts. We can looks for Google Accounts with a name like PrimaryAliasName, or PrimaryAliasName.eliapps, or Netid. After that, it is difficult to write a program that can reliably distinguish an alias name of "Alfred.E.Neuman" as personal but "Mad.Magazine" as departmental (and therefore an Alias entry that should have been moved to a Dependent Netid). For now we will follow the Shibboleth login rule that regards a PrimaryAlias as personal, but if the Primary is O365 then the first Secondary Eliapps alias is Personal (where it becomes the "special" Eliapps Contact where multiple Aliases are aggreated) and any subsequent Eliapps alias with a different MAILBOX is treated as Departmental and generates Contacts where we do not attempt to aggregate more than one Alias into the ProxyAddresses list. The Mail Routing will be done correctly no matter what. If we confuse personal and departmental then the number of Contacts can increase and the Outlook directory may not be optimal. Users can fix this by moving their damn departmental Eliapps aliases to Dependent Netids like they were supposed to.

Deprovisioning Departmental Contacts is not optional. Any Contact not supported by a current Email Alias entry has to be deleted. In fact, because the Azure AD will complain if any entries violate the rule that the same alias name cannot appear in the ProxyAddresses list of two objects at the same time, deprovisioning should occur before provisioning to handle the cases where Aliases are transferred from one Netid to another. Unfortunately, this is exactly the sort of thing IIQ does not handle well. So while we could continue the current model of building tools that update the Alias Table and then expect routing information to be generated in the background as is done for the current Mail Relays, future tooling should be designed to change the AD at the same time that the Alias table is changed.

It is proposed that Users are created as they are currently (CN=Netid, CN=Users, DC=yu, DC=yale, DC=edu), that Eliapps personal Contacts be created as (CN=Netid, OU=EliappsUsers,...) and that the remaining individual Aliases generate Contacts in (CN=AliasName, OU=Aliases, ...).

Inactive Accounts

There are 300K Netids, but only ~50K are active.

There is a big difference in performance between frequently aggregating all the users, or just the active users. An inactive user with no mail account does not change information frequently and there is no SLA on updating the last name in AD if an ALUM or retiree changes it. Therefore, it would be vastly better if we can agree to aggregate only the active users most of the time, and to update the inactive users (SOI=IDR and no Email Address) less frequently. Currently the TEST AD has such a scheme, but the PROD AD does not.

If inactive users can be deprovisioned from AD (which means they will need a PIN to reactivate), this would solve the problem.

It is a requirement that the code not create AD objects for inactive accounts (SOI=IDR) with no mail.

Bean Shell or SQL

The current AD Updater is driven by values generated by the YUONLINEDIR.PUBLISHED_AD_ACCOUNTS_V view in ACS1. It has logic of the form:

          DECODE (r.flag,
                  'G', ph.grad_student_phone,
                  'U', ph.campus_phone,
                  ph.office_phone
                 ) telephonenumber

Which basically says that the variable 'telephonenumber' will be taken from the grad_student_phone if the identity is a Grad Student, the campus_phone if the identity is an Undergrad, or else the office_phone for everyone else (Employees).

There are two ways this can be translated. We can create a comparable Database View only in IDR instead of ACS1 (which means translating Oracle SQL into SQL Server SQL) or we can create a Global Rule in IIQ that populates an Identity Variable named 'telephonenumber' from one of three IDR source fields (which means translating Oracle SQL into Java).

The primary difference is staffing (Java or SQL programmer) and maintenance (Change to the application or to the database).

We have to decide this before coding can begin. 

Out of Scope

For the moment, the management of the Google Directory (GADS) is out of scope. We expect this might change in a Phase 2 of the development rather than being added as initial work. Once Eliapps users have to have Contacts in the AD in order to be Email routed by Exchange Online, we have all the information needed to provision the Google Directory.

Service Accounts will not be managed. They have no interesting attributes, no source of identity, and we have nothing extra of offer.

Computer objects are currently handled by the Microsoft Domain.

Groups may be managed by Grouper.

We are interested in the Users container and the SOM OU. Anything outside those two areas will be ignored if it is well behaved. Generally speaking, if an object has a SAMAccountName outside the range of valid Netids and if it does not own ITS managed Email accounts, then we can leave it alone.

User and Contact objects not related to a Netid are out of scope. We may choose to declare that they are illegal in the Users container, but migrating them elsewhere is not a separate cleanup.

The Email Alias system will remain in its current form in ACS1. In some sense the information in AD duplicates the content of DIR_ONLINE_INFO, but AD is not able to provide the required function for generating unique Email Aliases.

Identities and Accounts

There will be an Identity object for every Netid (personal and dependent).

An Account object is created when aggregation correlates data from any of the six sources to an Identity. Correlation is by Netid, except for Azure AD where we will use the existing Email Provisioning correlation logic based on ImmutableID (because Azure doesn't store Netid).

Personal identities will have an IDR Account object.

Dependent netids will have an Netid System Account object.

At this time we believe that Contacts can be maintained as an Account under the Netid that owns the Alias that created the Contact. If that proves unworkable, then Contacts become a separate Identity.

We may choose to ignore personal Netid Accounts because they provide no useful information (the Netid is in IDR).

Each Email Alias is an Account that correlates to the Netid (personal or dependent) that owns it. Email Aliases are used to generate the proxyAddresses lists and drive Contact creation for a Netid with both O365 and Eliapps accounts.

Identity Variables

Although it is possible to provision fields from the IDR Account record directly into the AD object, the simplest approach is to create identity variables for each item that must be provisioned into the AD and Azure AD.

If we do not create Contact Identities, then when a Netid has both an O365 mail account and an Eliapps mail account, requiring the generation of a Contact to contain the routing information for the Eliapps account, then the Contact name, mail, mailNickName, proxyAddresses, and targetAddress  will be separate Identity variables from the corresponding variables that generate the same fields for the User object.

Indexed

• Netid
• UPN
• PrimaryAliasName
• ImmutableID (from AD, correlates Azure AD)
• SID (from AD, correlates Azure AD)

Needed for code selections

• YaleAffiliation (change EPA to distinguish Undergraduate, GradStudent, ...)
• O365Mail indicator
• EliappsUsername
• PrivacyIndicator

AD Fields (unindexed, from IDR)

• UPI
• Mail
• Surname
• GivenName
• DisplayName
• Title
• Description
• Department
• Company
• PhysicalDeliveryOfficeName
• Street Address
• PostOfficeBox
• Location
• ST
• PostalCode
• TelephoneNumber

Calculated

• O365Addresses (from AliasNames)
• EliappsAddresses (from AliasNames)

Mail Related Fields

Any non-mail entry (X500 or sip) in the ProxyAddresses list will be ignored by processing and will "pass through" from aggregation to provisioning.

The "SMTP/smtp" entries of the proxyAdddresses list will be generated by us and will replace all entries in the current AD. This will fix reported problems where the same value is duplicated in more than one proxyAddresses list belonging to different objects (which cannot be allowed if this information is used to do Mail Routing).

Because it is Yale Policy that an O365 account is Primary, and because the Primary Email Alias sets the UPN which is a key field of Azure AD, we do not allow an Eliapps account to be Primary if an O365 mailbox is assigned to the AD User object. Generally speaking, for an O365 user we will ignore the Mailbox (should it point to bulldogs) and associate the Primary Alias with O365.

Therefore, there are four AD configurations:

1. Primary is O365 and there is no Eliapps
2. Primary is Eliapps and there is no O365
3. Primary is O365 and there is an Eliapps secondary
4. Alias unrelated to ITS mail systems

There are some variables that need to be defined, and in a few cases the exact definition is technical:

• Netid
• PrimaryAlias is the type='person' Alias and has a PrimaryAliasName and a PrimaryMailbox.
• GoogleAccount is the name of the Eliapps account in the G Suite User Directory. If the PrimaryMailbox is of type 'bulldogs' then the Account is the part in front of "@". If the Primary is O365, then the Account is the part in front of "@" in the first "bulldogs" Mailbox we encounter in the Alias table, and if there is more than one bulldogs Mailbox the result is undefined.
• GoogleAlias is the AliasName from the Alias entry that set the GoogleAccount.
• O365Aliases are all the Secondary AliasNames that have a Mailbox of 'netid@connect.yale.edu'.
• EliappsAliases are all the Secondary AliasNames that have a Mailbox value of "GoogleAccount@bulldogs.yale.edu" except for the GoogleAlias.

The "Mail" attribute in AD is not a technical field. It is the Email Address to be "published" in the directory. At Yale it is the PrimaryAliasName@yale.edu and that happens to be the O365 account of a two-mailbox user because we require O365 to be Primary if it exists.

Primary O365

User Object

Primary Eliapps

User Object

Primary O365 with Eliapps Secondary

User Object (Same as Primary O365 above)

Contact Object

Foreign Mail Alias

Contact Object