CMDB / CSDM
CMDB trust is a decision problem, not a cleanup problem
Bulk cleanups remove duplicate, stale and orphaned CIs, and within months many of them are back. The cause is rarely the data itself. It is the missing decisions underneath: which attributes identify a CI, which source may update which field, which class a CI belongs to, and who owns it until it retires. Make those decisions and cleanup starts to hold.
Why cleanup does not hold
Most CMDB cleanups follow the same pattern. Someone exports the duplicates, a team spends a few weeks merging and retiring records, the CMDB Health dashboard improves, and everyone moves on. A quarter later the duplicates are back, often the same servers under slightly different names.
A cleanup removes the symptoms that existed on the day it ran. It does nothing about the integrations, discovery schedules and manual habits that created them, which keep running. If the rules that let a duplicate in are still in place, the duplicate is recreated the next time the same data arrives.
Trust in a CMDB comes from a small set of decisions: what makes a CI unique, which source is allowed to say what, which class a thing belongs in, and who is accountable for it from creation to retirement. Cleanup is worth doing once those decisions are made. Before that, it is rework scheduled for later.
Identification rules decide what counts as the same CI
In ServiceNow, identification rules tell the Identification and Reconciliation Engine, usually called IRE, how to recognize a CI it has seen before. Each rule belongs to a class and holds identifier entries, each listing the attributes that must match, such as serial number, name, or a cloud object ID, and has a priority. IRE tries the entries in order and stops at the first one that finds a match.
Child classes inherit their parent's rule unless they define their own, and an entry can use a related table, so a computer can be matched on a serial number held in the serial number table or on a network adapter's MAC address. Dependent CIs, such as an application running on a server, are identified through their relationship to the host, not on their own attributes alone.
Duplicates appear when these rules do not fit the data. A rule that matches on name alone breaks the first time one source sends a short hostname and another sends the fully qualified name. A rule that leads with serial number misbehaves for virtual machines whose serials are blank or copied from a template. The decision is which attributes are reliably present and unique, per class, across every source you actually load.
Reconciliation decides which source wins
Identification answers whether two records are the same CI. Reconciliation answers who may update it. Reconciliation rules name the data sources that are authoritative for a class, optionally down to individual attributes, and data source precedence gives each source a priority. When sources disagree, the authoritative source wins for its attributes, and data refresh rules let a lower-priority source take over a value once the authoritative one has gone quiet for long enough.
Without these rules the last writer wins. Discovery sets an operating system version, an SCCM or Intune import overwrites it with the agent's view an hour later, a cloud integration renames the CI to match an instance tag, and someone edits the record by hand in between. Nobody decided which source is right about which field, so the record changes with every load.
IRE only protects data that passes through it. Discovery and the Service Graph Connectors use it, but an older import set whose transform map writes straight to the CI tables does not, and a manual edit on a form does not either. List every path that writes to the CMDB, and for each source answer a few questions:
- Which classes may it create, and which may it only update?
- Which attributes is it the authority for?
- What does it send as the CI name, and does that match the other sources?
- Does it go through IRE, and if not, can it be moved onto it or retired?
The class you choose is part of the identity
The CMDB class hierarchy is not just a filing system. Identification rules, reconciliation rules, health rules and CSDM relationships are all defined per class, so a CI in the wrong class gets the wrong behavior everywhere. A Linux server stored as a generic computer is matched by the wrong rule, measured against the wrong recommended fields.
The usual mistakes are choosing a class that is too generic, building a custom table when an existing class already fits, and letting each integration pick its own class for the same kind of thing. The result is the same machine held twice, once in the class Discovery uses and once in a custom table that the identification rules never look at.
The decision is a class map: for each kind of thing you track, the one class it belongs in, and which sources may create it there. CI Class Manager is the place to review the hierarchy and the rules attached to each class before anyone adds a new table.
Ownership and lifecycle stop the slow decay
Stale and orphaned CIs are mostly a lifecycle problem. A server is decommissioned, Discovery stops seeing it, and nothing changes its status, so it stays operational for years. An application is retired, but its application service and relationships remain. Nobody was responsible for keeping it accurate when reality changed.
A few fields carry most of this. Operational status says whether a CI is in service now. The lifecycle fields, install status and, in newer releases, life cycle stage and life cycle stage status, say where it sits between order and disposal. Ownership fields such as managed by group and support group say who answers for it. When those are empty, stale CI tasks have nobody to go to.
The decisions are concrete. Decide what retirement means for each major class and which status values mark each step. Decide how long a CI can go unreported by its authoritative source before it is flagged, and who reviews it. CMDB Data Manager can then run retire, archive and delete policies on that basis, instead of someone running a bulk update once a year.
What CMDB Health scores tell you, and what they do not
The CMDB Health dashboard scores three areas: completeness, which checks required and recommended fields; correctness, which looks for duplicates, orphans and stale CIs; and compliance, which audits CIs against certification templates. It is good at showing trends, and weak as a single number to report upward.
Every score depends on how it was configured. Completeness only counts the fields marked as recommended. Orphan and staleness rules are defined per class, and classes without rules contribute nothing. Duplicate detection follows the identification rules, so if a rule is too loose to see two records as the same CI, they are not counted as duplicates. A high score can mean good data or a narrow measurement.
Treat the scores as a check on decisions already made, not as the decisions themselves. Once identification, reconciliation and ownership are settled, health rules can be set to measure exactly those things. A falling score then means something specific: a source misbehaving, a rule gap, or a team that stopped retiring what it decommissioned.
CSDM alignment is a sequencing decision
The Common Service Data Model describes where services, applications and infrastructure belong in the CMDB and how they relate. It is organized into domains, from foundation data through design and planning to service delivery and service consumption, and ServiceNow describes adopting it in stages rather than all at once. Populating every table in one push mostly produces empty records.
A sensible order starts with foundation data, meaning users, groups, locations, companies and models, because every later relationship points at it. Next come business applications and the application services that run them in each environment, because incident, change and Service Mapping rely on that link. Technical services and their offerings follow, and business services come last, once someone owns a service portfolio.
Business services and application services get confused because both are called services and both descend from the same service table. An application service is a deployed instance of an application, such as production payroll and the servers and databases under it. A business service is what the business consumes, such as paying employees, and may depend on several application services. CSDM versions differ in naming and in which classes they recommend, so check which version your instance follows.
Before scheduling another cleanup, write down four things for your most important classes: the identifier entries and the attributes each source really sends, the reconciliation rules and precedence for each source, the class every kind of CI belongs in, and the owner and retirement steps. Then compare it with your instance's configuration. Where the two disagree is where your next duplicates will come from.
References
- Identification and Reconciliation Engine (IRE) (opens in a new tab)ServiceNow documentation
- Reconciliation rules (opens in a new tab)ServiceNow documentation
- Overview of CMDB Health (opens in a new tab)ServiceNow documentation
- Implementing the CSDM framework in stages (opens in a new tab)ServiceNow documentation
Related reading
- CMDB and CSDM at EmetrixCMDB trust first, then CSDM alignment, starting with a CMDB Trust Check.
- Duplicates that came back every quarterField note, a composite pattern rather than a client.
- Bringing Fleet device data into the ServiceNow CMDBArticle on the decisions to settle before device data reaches the CMDB.
Want this looked at on your platform?
CMDB Trust Check: Typically two to three weeks.
Ask about a CMDB Trust CheckWe reply within 24 hours on business days.