Inventorying What You Monitor Before You Move It
Ask an admin what their monitoring system watches and you get a confident answer: the servers, the switches, the firewalls, the usual. Ask the monitoring system the same question and the answer is longer, stranger, and full of things the admin forgot. A temperature sensor somebody added during a heatwave. A check on a vendor's portal that mattered for one project. A UPS in a closet that is only a closet because a wall was moved. I have given the confident answer myself, and I was wrong by about four hundred rows.
Migrations get built from the confident answer, and that is why so many of them miss things. The inventory phase replaces memory with an export. Before a single thing is built in the new platform, you extract everything the old one knows. That includes devices, checks, thresholds, alert rules, contacts, quiet periods, dependencies, dashboards, reports, and integrations. Before that build, put it into one spreadsheet with one row per check.
That spreadsheet becomes the single source of truth for the whole migration. It is what you build from. It is what you compare against during parallel running (both systems watching the same devices while only the old one is trusted). It is what you test against for alert parity (proof that every alert that matters still fires and still reaches a person). And it is what you sign off before cutover, the day the new system becomes the one people act on.
By the end of this guide you will know exactly what to export and from where. You will know how to lay the spreadsheet out so it stays useful for months. You will know how to find the checks nobody remembers. And you will know how to prune dead weight so you do not spend weeks recreating checks that should have been deleted years ago.
In this guide
- Why the inventory comes first
- What to export from the old system
- The migration spreadsheet: one row per check
- Column-by-column template
- An illustrative export excerpt
- Finding the checks nobody remembers
- Pruning during inventory
- Counting twice: verifying the inventory
- Where to go next
Why the inventory comes first
Everything later in a migration is a comparison against something. The build phase asks "is everything recreated?" Parallel running asks "does the new system see what the old one sees?" Parity testing asks "does every alert that matters fire?" None of those questions can be answered without a list to compare against. The list has to come from the old system's actual configuration, not from anyone's recollection of it.
There is a second reason. The inventory is the cheapest phase to do well and the most expensive to do badly. A missing row costs an hour to add now. The same missing row costs an outage to discover after cutover, when the old system is no longer there to catch it. Spend the time here.
This applies whichever way you are moving. The export side of a migration between two commercial tools, from open source to commercial, or from commercial to open source is the same job. Only the export buttons differ.
What to export from the old system
Work through these in order. Each is a different export in most tools. And some will be a screen you have to read rather than a file you can download. Write down where each came from; you will need to go back.
The device list
Record every device the old system polls, with its name, address, and how it is reached. Record the protocol (SNMP, WMI, SSH, an agent, or a plain ping) and the settings that protocol needs. For SNMP that means the version and, for version 3, the user name and the authentication and privacy methods. For WMI it means which account is used. Record the settings, not the secrets. The spreadsheet should say "SNMP v3, user monitor, auth SHA, priv AES" and never contain a passphrase. The secrets stay in your password manager; the inventory just needs enough that whoever builds the new system knows what to ask for.
Also record each device's group, site, or location in the old system, and any custom fields such as owner or contract number. Those groupings are usually how dashboards and alert rules are organized, and losing them makes the rest of the inventory much harder to interpret.
Every check and its interval
A check is one question asked of one device on a schedule. Some tools say sensor, monitor, service, or item; the word varies, the idea is the same. The question might be: is port 24 up, how full is drive D, is the web service answering, what is the CPU load. This is the row unit of the spreadsheet. For each check, record the device, what it measures, and its polling interval — how often the question is asked. Intervals matter because the new system will default to its own. A check that was tuned to run every thirty seconds because a particular link flaps will silently become a five-minute check unless someone notices.
Expect the count to be larger than you think. A single switch can carry fifty interface checks. A single server can have a dozen. A few hundred devices can easily mean several thousand rows, which is why this lives in a spreadsheet and not a document. A document would have given up somewhere around the second switch stack.
Thresholds, and which ones were tuned by hand
A threshold is the value at which a check's answer becomes an alert: disk above ninety percent, CPU above eighty for ten minutes, response time above two seconds. Most checks use the tool's default. Some have been changed by a human, usually after the default caused a false alarm or missed a real problem. Those hand-tuned thresholds are the single most valuable thing in the inventory. They encode months or years of learning about your specific environment. They encode that the backup server is supposed to hit full CPU at two in the morning. They encode that the file server's disk has hovered at ninety-four percent since forever. They encode that the branch link's latency is fine up to a number far higher than the default.
Many tools can show you which thresholds differ from the default; if yours cannot, compare the export against a freshly added device's defaults. Mark the hand-tuned ones clearly. When the new system is built, they must be reproduced exactly. And when the two systems disagree during parallel running, the hand-tuned thresholds are the first place to look.
Alert rules and escalation contacts
An alert rule decides what happens when a threshold is crossed. It decides who is told, how (email, text message, a chat channel, a ticket), how urgently, and what happens if nobody responds. Export every rule. Separately export the contact list and the escalation chain — the on-call person first, the team lead after twenty minutes, the manager after an hour. Check the contact details as you go. Alert contacts are the most out-of-date data in any monitoring system. A migration is the one time everyone will tolerate being asked to confirm their phone number.
Note which rules are actually enabled. Old systems accumulate disabled rules, rules that point at people who left, and rules whose schedule means they never fire. Record all of them, but mark them, because you will prune most. One or two will exist for a reason nobody can remember; keep those until someone can, because the reason is usually an outage.
Maintenance windows and suppressions
Export every scheduled quiet period. These include the nightly backup window when storage alerts are muted, and the monthly patch Sunday when server reboots are expected. Also export every long-standing manual suppression that someone set and never cleared. The scheduled ones are configuration and must be recreated. The manual ones are usually an unspoken decision that a check is not worth alerting on, which is worth knowing. The scheduled versus ad-hoc suppression guide covers the difference and why the ad-hoc kind tends to outlive its reason.
Dependencies and parent relationships
A dependency tells the system that if the branch router is down, the forty devices behind it are unreachable because of the router, not because of forty separate faults. In that case, the system should send one alert rather than forty-one. (Without one, you get forty-one, at three in the morning, in alphabetical order.) Dependencies are what keep alert noise down. They are among the hardest things to export cleanly, because they are often implicit in a map or a group structure rather than stored as a list. Do the work anyway: for every device, record what it depends on. If the old tool cannot export this, draw it from the network diagram and confirm it against how the old tool behaved during the last big outage.
Dashboards and reports people actually open
Not every dashboard — every dashboard people use. Most tools can show when a dashboard was last viewed or who has it bookmarked; failing that, ask each team what they look at. For each, record who uses it, what it shows, and how often. Do the same for scheduled reports: the monthly availability summary that goes to a manager, the weekly bandwidth report, the certificate-expiry list. Reports are especially important because they are the deliverables the rest of the organization actually sees, and their absence is noticed immediately after cutover.
Integrations
Record everything that talks to or is triggered by the monitoring system. Include outgoing webhooks that post to a chat channel, and the connection that opens tickets in the help-desk system. Include scripts that run on alert (restart the service, then tell someone), and inbound feeds such as a trap receiver or a syslog listener. Also include anything that pulls data out, such as a status page that queries its database. Integrations are the part of the old system most likely to have been set up by someone who left, and the most likely to break silently on cutover.
The migration spreadsheet: one row per check
All of the above lands in one workbook. One tab per export type is fine for devices, contacts, dashboards, and integrations. But the main tab — the one everyone works from — has one row per check. Every row carries the same columns. Those columns follow the check through the whole migration. They show where it came from, where it is going, whether it has been built, and who is responsible. They also show whether it has been verified in the new system.
Keep it somewhere shared and versioned, so the whole team edits the same copy. The spreadsheet is the migration. If two people have two copies, you have two migrations, and neither is finished.
Tip: Keep the old system's raw exports — the CSV files, the configuration backup, screenshots of screens that would not export — in the project folder too, untouched. The spreadsheet is your working copy; the raw exports are the evidence you go back to when a row in the spreadsheet is questioned.
Column-by-column template
| Column | What goes in it | Why it matters |
|---|---|---|
| ID | A short unique reference, such as S-0412 | Rows get referred to in emails, tickets, and findings logs |
| Device | The old system's device name and address | Joins the row to the device tab |
| Check | What it measures, in plain words: "Interface Gi1/0/24 status," "Drive D: free space" | Plain words survive the change of vocabulary between tools |
| Protocol | SNMP, WMI, SSH, ping, HTTP, agent | Tells the builder which collector to use in the new system |
| Interval | How often it polls in the old system | Must be reproduced, not defaulted |
| Thresholds | Warning and critical values, with units | Copy exactly |
| Hand-tuned? | Yes or no | The yes rows are checked first when the systems disagree |
| Alert rule | Which rule it triggers, and at what severity | Feeds the parity list |
| Dependency | Parent device, if any | Rebuilds the dependency tree |
| Window | Maintenance window it belongs to, if any | Recreates quiet periods |
| Source | Where in the old system this came from: export file, screen name | You will need to go back |
| Target | Where it lives in the new system: device and check name | The link between old and new |
| Status | To build, Built, Verified, or Pruned | The progress bar for the whole migration |
| Owner | The person who builds and verifies it | No row without an owner |
| Verified | Date and by whom, during parallel running | The gate before cutover |
| Notes | Anything odd, and the reason for any pruning | The "why" someone will need in a year |
An illustrative export excerpt
Here is what a few rows of the main tab look like once the columns are filled in, shown as comma-separated text. The device names and addresses are made up; the shape is the point.
id,device,address,check,protocol,interval_s,warn,crit,hand_tuned,alert_rule,parent,window,status,owner S-0412,sw-core-01,10.0.0.2,Interface Gi1/0/24 oper status,snmp,60,,down,no,NET-CRIT,,,to build,jsmith S-0413,sw-core-01,10.0.0.2,Interface Gi1/0/24 in utilization,snmp,60,70%,90%,no,NET-WARN,,,to build,jsmith S-0980,fs-01,10.0.1.15,Drive D: free space,wmi,300,10%,5%,YES,SRV-CRIT,sw-core-01,backup-nightly,to build,akhan S-1207,ups-closet-b,10.0.3.40,Battery runtime remaining,snmp,120,15 min,5 min,YES,FAC-CRIT,sw-closet-b,,to build,jsmith S-1208,ups-closet-b,10.0.3.40,On battery,snmp,120,,yes,no,FAC-CRIT,sw-closet-b,,to build,jsmith S-1550,vendor-portal,portal.example,HTTPS response time,http,600,3 s,10 s,no,APP-WARN,,,pruned,akhan
Row S-0980 is hand-tuned. The file server's disk has always run close to full, so the thresholds were lowered from the defaults. That row also belongs to the nightly backup window, so both facts must travel with it. Rows S-1207 and S-1208 are the closet UPS nobody remembered. The UPS was found by the alert-history method described below. It has a hand-tuned runtime threshold because the closet has no generator. Row S-1550 is a portal check from a finished project, marked pruned with a note explaining why, and it will not be built. Every row has an owner. Every row has a status.
Finding the checks nobody remembers
The export tells you what is configured. It does not tell you what matters, or whether the export is complete. Three sources fill that gap.
The alert history. Export every alert the old system raised in the last year and count them by check. The checks that alerted are, by definition, the ones that have been doing work. Cross-reference against the spreadsheet. Any check that alerted in the last year and is not in the inventory means the export missed something. In that case, you need to find out what else it missed. Any check that alerted repeatedly and was acknowledged each time without action is a candidate for a threshold fix or pruning. But that is a separate conversation.
The people. Ask each team one question: "What has monitoring told you about in the last year that you were glad to know?" The answers are the checks that must not be lost. And they are surprisingly often ones the migration team would never have prioritized. Examples are the temperature sensor, the license-expiry check, and the ping to the vendor's VPN endpoint that explains why the payroll system is slow.
The building. Walk it. Every rack, every closet, every device with a network cable and a management port. Check each one against the device tab. The forgotten UPS is found by walking past it. So is the warehouse access point that someone monitors from their laptop because it was never added to the real system.
Harbourview Clinic built its inventory from the device export alone. The alert-history cross-check was in the plan as "if time permits", and time did not permit. Three months after cutover, the gateway that sends appointment-reminder text messages stopped authenticating on a Friday evening. Nobody heard about it until Monday's waiting room was twice the size the roster said it should be. The old system had watched that gateway for six years, in a group called "Misc", on behalf of a practice manager who had long since retired. Its alert history showed the check firing about twice a year, each time caught and fixed within the hour. The new system had never been told the gateway existed.
Pruning during inventory
Do not migrate dead checks. These include a check whose device was decommissioned, and a check on a service that no longer exists. They include a check that has sat in an unknown state for a year. They also include a check that alerts weekly and is acknowledged without anyone doing anything. Mark these pruned in the status column, with a reason, and do not build them. Keep the rows, though. A pruned row with a reason is evidence that a decision was made. That is far better than an absent row that looks like an oversight when someone asks about it later.
Be more careful pruning alerts than pruning checks. An alert that fires constantly and is ignored may be badly tuned rather than useless. Deciding which is a review exercise in its own right. The alert review process describes how to do it properly. During the inventory, be aggressive about pruning checks on things that no longer exist and conservative about pruning checks on things that do.
The inventory is also the one chance to fix naming. If half the switches are named by asset tag and half by location, decide the convention now. In that case, put the new name in the Target column, and build the new system consistently. You will not get another chance, because nobody renames four hundred switches for fun.
Counting twice: verifying the inventory
Before the inventory phase is declared done, a second person verifies it. Not re-does it: verifies it. That means four things, in order.
- 1. Count. The number of devices on the device tab equals the number in the old system. The number of check rows equals the old system's check count. If either count differs, the difference is written down and explained: pruned rows, disabled checks, template-generated checks the export flattened.
- 2. Sample forward. Pick twenty random rows and look them up in the old system. Interval, thresholds, alert rule, dependency: do they match what the old system actually has?
- 3. Sample backward. Pick twenty random checks in the old system's console and find them in the spreadsheet. This catches export gaps the forward sample cannot.
- 4. Ask. Each stakeholder confirms that the dashboards and reports they use are on the list, by name.
When all four pass, the gate is open and the build phase can begin. When any fails, the fix is cheap now and expensive later, which is the whole argument for doing this first. Write the result down — who verified, what matched, what was added — and file it with the exports.
Where to go next
The inventory is the least exciting phase of a migration and the one that decides whether the rest goes well. Export the device list, every check with its interval, and every threshold with the hand-tuned ones marked. Export alert rules and contacts, maintenance windows, dependencies, the dashboards and reports people open, and the integrations. Put it all in one spreadsheet, one row per check, with a source, a target, a status, an owner, and a verified column. Use the alert history, the people, and a walk around the building to find what the export missed. Prune what is dead, keep the row and the reason, and have a second person count twice. Then, the next time someone asks what the monitoring watches, you can give the long answer. It is the only one that survives a cutover.
With the spreadsheet in hand, the next phase is standing the new system up beside the old one: Parallel Running: Two Monitors, One Truth. The alert-rule column feeds The Alert Parity Checklist, and the reports and dashboards you listed here are the starting point for Data Export Realities. For the plan the inventory belongs to, and the failures it prevents, see Why Monitoring Migrations Fail. For the whole topic, see Migrating Between Monitoring Platforms Without Losing Your History.
Put this into practice with SysPeep Monitor
When you reach the build phase, a new system's own discovery makes a useful cross-check on the inventory. SysPeep Monitor auto-discovers devices and interfaces on the network. So comparing what it finds against your device tab is a quick way to catch a closet you did not walk. The free trial runs on a Windows machine and is a reasonable candidate for the parallel run, if your shortlist has room for one.
Download Free TrialLearn About SysPeep