ITOM / Discovery
Does Discovery find what is really running? Five ways to check
A Discovery schedule that completes every night proves only that the schedule ran. Whether the CMDB reflects what is actually running depends on scope, credentials, MID Servers, schedule design, and classification. These five checks show where the gaps are, and the order to fix them in so the fixes hold.
Running is not the same as reflecting
Most platform owners can say whether Discovery is running. Fewer can say how much of the environment it reaches, or how many of the CIs it touched were classified correctly. Those are different questions. A schedule that ends with a status of Completed tells you the probes went out and something came back. It does not tell you the ranges were complete, the credentials worked, or the resulting records landed in the right classes.
The CMDB inherits every gap Discovery leaves. Change risk, incident assignment, Service Mapping, and asset reconciliation all read from it, so a quiet coverage gap becomes a wrong answer somewhere else. The checks below are meant to be run in order, because each one narrows what the next can tell you.
Check one: compare scope with the real environment
Start outside ServiceNow. Ask the network team for the subnets they route, the cloud team for the accounts and subscriptions they own, and the data center team for anything managed out of band. Then compare that list with the IP ranges and range sets attached to your Discovery schedules. The gaps are rarely mysterious: a subnet added after go-live, a VLAN for a new site, a cloud account created by a project team, a DMZ nobody wanted to open firewall ports into.
Cloud needs its own comparison. On-premises Discovery works by scanning IP ranges and probing whatever answers. Cloud discovery works by calling the provider's APIs with a service account, so its scope is defined by which accounts, subscriptions, or projects are registered, not by address ranges. A whole cloud environment can be missing from the CMDB while every IP-based schedule reports success. Short-lived resources such as autoscaled instances and containers also need a decision about what the CMDB should hold at all.
Write the result down as a coverage statement: what is in scope, what is deliberately out, and what is missing. That statement is the baseline every later fix is measured against.
Check two: read the errors, starting with credentials
Every Discovery status records what happened to each device it found, and the Discovery log and ECC queue hold the detail. In many instances nobody reviews them. A device that answers the port scan but cannot be logged into is either not created or created with too little detail to be useful, and the schedule still finishes normally.
Credential failures are usually the largest category and the most fixable. Typical causes are a service account whose password was rotated without an update in ServiceNow, a Windows account that lacks the WMI or remote access rights the probes need, an SSH account without the elevated rights some commands require, and SNMP community strings that differ by site. Credential affinity can hide the problem for a while, because a device keeps using the credential that worked last time until it stops working.
Group the errors by cause rather than by device. A long list of failed devices is discouraging and tends to be ignored. Three or four root causes behind that list is a plan someone can own.
Check three: MID Server placement and health
The MID Server is where Discovery actually runs. If it sits in the wrong network segment, its probes are blocked by firewalls before any credential is tried, and those failures can look like empty ranges rather than errors. Check that each MID Server can reach the ranges it is assigned, on the ports the probes need, and that MID Server selection by IP range, capability, and application sends work where you expect.
Then check health. A MID Server that is down, not validated, behind the instance version, short on memory, or working through a backed-up ECC queue will slow or drop work without much noise. The MID Server record, its dashboard, and the agent log tell you most of what you need. A MID Server cluster for load balancing or failover reduces the chance that one host quietly takes a whole location out of Discovery.
Drift between environments matters here too. If development, test, and production MID Servers have different capabilities, credentials, or properties, a fix that worked in test proves little about production.
Check four: schedules, duration, and stale CIs
Look at how long each schedule actually takes, not how long it was designed to take. A schedule that once finished in a couple of hours may now run into the business day, overlap the next schedule, or hit its maximum run time before it reaches the last ranges. Overlapping schedules that cover the same ranges waste MID Server capacity and can produce conflicting updates to the same CIs.
Then look at staleness. Every CI that Discovery updates carries a last discovered date. CIs that have not been discovered for longer than your expected cycle are retired, unreachable, or no longer in scope. CMDB Health's staleness metric and CMDB Data Manager policies can help you find and retire them, but only once you know which explanation is true. Retiring records that are stale because of a credential failure removes real infrastructure from the CMDB.
Design schedules around how often each kind of data changes. Servers in a busy environment may need daily runs; network gear and storage may need less. Frequency should follow need, not habit.
Check five: discovered is not the same as classified
A device that Discovery reaches still has to be classified and identified. Classification decides what kind of thing it is: a Linux server, a Windows server, a network switch, a storage array. Identification, through the Identification and Reconciliation Engine, decides whether it matches an existing CI or needs a new one. A mistake at either step produces records that exist but mislead.
Signs to look for include CIs piling up in generic classes rather than specific ones, network devices with no model or manufacturer because no SNMP classifier matched, duplicate CIs for the same server created by Discovery and another data source, and identification rules that rely on attributes that change, such as name or IP address alone. Duplicates and misclassified CIs break relationships, which matters as soon as Service Mapping or impact analysis depends on them.
Check the patterns as well. An outdated Discovery and Service Mapping Patterns application can miss newer operating system versions and application components. Keeping it current is cheap compared with explaining the gaps later.
What to fix first
Work root causes before symptoms. Coverage and access come first: ranges, cloud accounts, MID Server placement, firewall rules, and credentials. Until those are right, cleaning up records only cleans up the output of a process that will produce the same problems again on the next run.
Settle ownership before tuning. Each range, credential, MID Server, and CI class needs a named owner who acts when it fails, and the network, server, and cloud teams need to know that some Discovery errors are theirs to fix. Without that, schedule tuning and pattern fixes decay as soon as the people who made them move on to other work.
Only then move to the visible symptoms: duplicates, stale records, misclassified CIs, and gaps in service maps. With the causes fixed, those clean-ups stay done, and you can measure progress against the coverage statement from check one rather than against how busy the Discovery status list looks.
A useful first step needs no project. Pick one schedule, open its most recent Discovery status, and group its errors by cause. Then compare its ranges with the subnets the network team says exist. If those two exercises tell you something you did not know, the rest of these checks are worth the time, and the order above is a reasonable place to start.
References
- Configure a MID Server cluster (opens in a new tab)ServiceNow documentation
- MID Server system requirements (opens in a new tab)ServiceNow documentation
- Overview of CMDB Health (opens in a new tab)ServiceNow documentation
- Identification and Reconciliation Engine (IRE) (opens in a new tab)ServiceNow documentation
Related reading
- ITOM at EmetrixDiscovery, Service Mapping, Event Management and cloud discovery, with a fixed-scope health check to start.
- A Discovery environment that ran every night and told nobody anythingField note, a composite pattern rather than a client.
- CMDB trust is a decision problem, not a cleanup problemArticle on identification, reconciliation, ownership and CSDM sequencing.
Want this looked at on your platform?
Discovery and Service Mapping Health Check: Typically two to three weeks.
Ask about a Discovery and Service Mapping Health CheckWe reply within 24 hours on business days.