Inherited a server: what to check first when you have no ops hire

Author: OpsMate team
Updated
Based on hands-on troubleshooting with OpsMate. Run commands only on servers you are authorized to access; replace paths and service names for your environment.

Inherited a server someone else built? First find out what is running and write a one-page inventory—do not start changing config. The contractor is gone, the handover was an IP and a password, and a full-time ops hire is not in the budget. This guide is for founders, PMs, and the developer who "also looks after the server": one-hour inventory → 10-minute weekly read-only check → decide who you call when things break.

How long the inventory takes depends on how many machines you have, what the handover actually included, and whether you can log in. The one hour and the ten minutes a week above are a starting point for a small team, not a promise that every environment finishes on that clock.

This is about surviving a server you inherited. If you want to know which metrics to monitor after launch, see the small-team VPS monitoring checklist. Every command below is read-only and works on a typical Linux VPS. If the output makes no sense yet, that's fine: save it as-is and read it alongside the explanations here.

Name the environment and the stage before you inventory the host

This page is for someone who is allowed to take over the servers. The commands below apply to a typical Linux host. They do not manage switches, storage arrays, or the virtualization control plane. Decide which environment you have, and which stage you are in, before you decide how far the checks should go.

EnvironmentBesides the services on the host, get this firstA handover is useful whenDo not conclude that
Cloud serverWho administers the company cloud account, the instance list and region, and who pays the bill; where the public entry point, security group, snapshots, and managed services liveYou can match the instance to the business in a console you are allowed to use, and you know who can renew it or open an emergency sessionSSH access means you control the cloud account, or that a snapshot means the business can already be restored
Your own rack or colocationWhere the machine and rack are, the asset tag, the facility contact, and who owns out-of-band management, the network gear, and the storageYou can tie the address the business uses to the physical host and its uplink, and you know who to call if the system will not bootA healthy Linux check means the power, disks, switch, or redundant link is healthy
Cloud and facility togetherWho controls each side, plus the VPN or private circuit, DNS, identity, and backups that cross the boundaryYou can walk one real request and mark the entry point, app, database, and outside dependency, and the person who owns each segment agreesOne healthy server means the path across environments is healthy. Do not change production routes or firewalls just to test that path

Before the handover: collect the records, and do not change config

Ask for the host and service list, how deploys happen, who owns each account, and where backups live. Move passwords through the company's password manager, not inside the inventory. For each item, write who received it and whether it was checked. If you did not get it, write "not received". A spoken promise is not a check.

An IP address and an SSH password are not control of the cloud account or of the facility contract. If those are still with the person who is leaving, talk to whoever owns the company's accounts. The host commands later on this page cannot get that ownership back.

Just after you take over: write a baseline, then sort out access

Use the checks below to connect domain names, ports, processes, and databases. Mark which lines you verified and which are still a guess. For access that belonged to someone who left, follow the order for taking server access back, and keep a login you have already tested. Do not change the only password that still works.

If the site is already down, or you suspect someone broke in, deal with that event first. Use the outage order and the read-only checks for suspicious logins. A handover inventory is not incident response.

Once the service is steady: turn the inventory into a routine

For each service, name an owner, how often someone looks, and who gets the call. Use the small-team VPS monitoring checklist for alerts. Practice a restore in an isolated environment. Do not restore over production data. Update this inventory after a deploy, an account change, or a network change.

What you can check yourself, and what needs someone else

Inherited a server? What to check first

On an inherited server, run these read-only checks before you change anything, and paste the output into one document. This is the inherited server checklist we use ourselves:

If a line makes no sense, write it down as "unknown" instead of deleting or restarting anything. The sections below explain each check.

1. Find out who owns what before you touch the server

Small teams rarely get stuck on a mistyped command. They get stuck because nobody can log in to the account that matters. Fill in this table first. Record where things are and who owns them, never the passwords or keys themselves:

ItemWhereWhose loginWho can change itExpiry / renewal
Cloud account (including billing)
Domain registrar and DNS
HTTPS certificate
Code repository and deploy method
Database and where backups live
Third-party API keys (email, payments, SMS…)
User long-lived access keys (separate from role session credentials)
Vendor trust, and your own production role (keep them separate)
SaaS signed up with a personal email, including the domain renewal mailbox

2. Take inventory: what is actually running here?

System and resources:

cat /etc/os-release
uname -r
uptime
nproc
free -h
df -h

This gives you the OS version, kernel, uptime, CPU core count, memory and disk. It does not tell you whether those resources are enough; that takes trends over time.

Running services and listening ports:

systemctl list-units --type=service --state=running --no-pager
sudo ss -ltnp

Containers and other process managers:

docker ps --format 'table {{.Names}}\t{{.Image}}\t{{.Status}}\t{{.Ports}}'
docker compose ls
pm2 list

The front door (Nginx):

ls -l /etc/nginx/sites-enabled/ /etc/nginx/conf.d/ 2>/dev/null
sudo nginx -T 2>/dev/null | grep -E '^\s*(server_name|listen|proxy_pass|root)\s'

nginx -T prints the full active configuration. The filter keeps only domain names, ports, proxy targets and static roots, so you can connect "this domain → this port → this service". Check a full config for internal addresses or credentials before sharing it with anyone.

Scheduled jobs and backups:

crontab -l
sudo crontab -l
ls -l /etc/cron.d/
systemctl list-timers --no-pager
sudo grep -rIl -E 'backup|dump|rsync|restic|borg' /etc/cron* /var/spool/cron 2>/dev/null

A successful job only means that command exited 0. This page has no restore command and no completed drill record, so that exit code is not a successful restore. Rehearse only on a non-production test host, container, or cloud instance whose account and network are separate from production, and restore into that machine's own database and disk. An empty directory is only a path inside that isolated machine. It is not the isolation. Do not overwrite production paths, and do not modify the original backup.

On the test copy, before you start the restored app or its jobs, remove or replace the production database password, .env values, mail and payment API keys, and cloud credentials in that copy's config, and use a test identity instead. Do not let the test workload read production credentials from the production instance identity or from a mount. Keep the network block as its own control: block the test environment from production and from real external services, and do not open that path because the credentials were replaced. Change only the test copy. Do not modify the original backup, and do not rotate production keys that are still in use. Then turn off the copy's tasks that send mail, call a payment API, or connect back to production. Do not stop production jobs for the drill. Check one result you wrote down beforehand. Keep the time, the backup identifier, the isolated environment, and the check result. When you finish, delete only the drill copy you have confirmed. Keep the original backup. OpsMate does not take backups and does not restore them.

Certificates:

sudo certbot certificates

From your own computer you can also check the live certificate's expiry date (replace example.com with your domain):

echo | openssl s_client -connect example.com:443 -servername example.com 2>/dev/null | openssl x509 -noout -enddate -issuer

certbot certificates only returns something if the server uses certbot. Your certificate may be managed by a CDN or load balancer instead, in which case check that provider's console.

Renewed, and the browser still says expired

The command above only shows the expiry of the leaf the public endpoint sent. A newer notAfter in a file does not, by itself, prove the process is still serving an old certificate. A CDN and an origin, or two nodes, may use different certificates on purpose. Suspect a file that has not been loaded only when you have the same domain and SNI, the same termination layer, the same node, and the ssl_certificate path from the server { } block for that name. Then compare the leaf serial and SHA-256 fingerprint, and compare notBefore and notAfter with the current time. If one of those is missing, write unknown. Do not force a renewal or reload because of it. *.example.com covers one label only. -checkhost checks the name, not the full chain. The s_client print does not prove trust or revocation. A timeout, an error, or a blank result is unknown. In an ordinary shell, pipe_status 0 is only the exit status of x509 at the end of the pipe. It does not mean the TLS handshake, the trust chain, or the timeout check succeeded.

On the machine that terminates TLS, sudo nginx -T is the config on disk. Put the path from the matching block into PATH_FROM_THE_MATCHING_SERVER_BLOCK. Do not keep a sample path as if it were the real one. The timer's enabled state, exit result, and renewal log are clues. A missing certbot.timer can be snap, cron, or another client. Redact the output. The step-by-step commands are on the site is suddenly down.

sudo nginx -T
sudo certbot certificates
systemctl is-enabled certbot.timer
systemctl show certbot.timer -p ActiveState -p UnitFileState -p Result --no-pager
systemctl show certbot.service -p ActiveState -p Result -p ExecMainStatus --no-pager
sudo journalctl -u certbot.service -n 40 --no-pager
sudo ls -l /etc/letsencrypt/renewal-hooks/deploy /etc/letsencrypt/renewal-hooks/post
sudo openssl x509 -in PATH_FROM_THE_MATCHING_SERVER_BLOCK -noout -subject -issuer -serial -fingerprint -sha256 -dates -ext subjectAltName
sudo openssl x509 -in PATH_FROM_THE_MATCHING_SERVER_BLOCK -noout -checkhost example.com

If the domain renewal email still goes to someone who left, write that on the registrar and certificate rows above. OpsMate does not renew certificates and cannot sign in to the registrar.

Who can log in:

awk -F: '$3 >= 1000 && $3 < 65534 {print $1}' /etc/passwd
getent group sudo wheel
sudo find /root /home -name authorized_keys -exec wc -l {} \;

This lists regular users, who has sudo, and roughly how many public keys each authorized_keys file holds (usually one per line). Write down accounts and keys belonging to people who have left; they are the first things to decide about revoking. For the order that takes access back without locking yourself out, see Contractor or ex-employee left? How to revoke their server access .

3. Write it down on one page

One row per service is enough:

ServiceRuns via (systemd / Docker / PM2)PortConfig locationLog locationHow to restartWho knows it best

The cells you can't fill in are where you'll get stuck during the next incident. Fill those first, or ask whoever built the setup.

4. A 10-minute weekly read-only check

uptime
free -h
df -h
df -i
systemctl --failed --no-pager
docker ps -a --format 'table {{.Names}}\t{{.Status}}'

Also glance at when the last backup ran and how long until the certificate expires. For a fuller list of checks and how to route alerts, see the small-team VPS monitoring checklist. For the times you're away from your desk, patrol checks and Telegram alerts can notify you first. Receiving Telegram alerts doesn't require storing credentials in the cloud; only running SSH on the server remotely from Telegram does. See Server alerts in Telegram.

What the numbers in an alert mean

NumberWhat it isWhen it mattersFirst command to run
LoadTasks running or waiting for CPU/disk. The three numbers in uptime are 1-, 5- and 15-minute averagesStaying well above the core count while users notice slowness. Short spikes are usually fineuptime, nproc
Memory %Share of memory in use. Linux uses spare memory for cache, so this number often looks highWhen available stays very low, swap keeps growing, or the kernel log shows processes being killedfree -h
Disk %Used capacity on a filesystemNear 100%, writes fail and databases and logs break first. The growth rate matters more than the number itselfdf -h
Inode %The budget for the *number* of files, separate from capacityAt 100% you can't create files even with free space, typically from huge numbers of small filesdf -i

There is no universal threshold. Compare against what's normal for this machine. If a disk or inode alert fires, follow Linux disk usage suddenly spikes before deleting anything.

5. Your escalation path

  1. Gather evidence first. Run the read-only checks in the order from Site down at 2 a.m. and save the output. Helpers move much faster when you hand them evidence.
  2. Then call someone. Keep contact details ready for whoever built the setup, a contractor or consultant you can pay per incident, and your cloud provider's support portal.
  3. Make temporary access revocable. Give outside help their own account and key, never a shared root password. When the work is done, remove their key, disable the account and rotate any secrets they handled. When someone leaves, see Contractor or ex-employee left? How to revoke their server access .
  4. If you suspect a break-in, don't start deleting things yourself; see Was my Linux server hacked?

Where OpsMate fits

OpsMate puts an SSH terminal and an AI assistant on the same page. For the inventory, you can type each command above into the terminal yourself. If you can't remember them, ask the AI in plain language, for example "check which services are running on this machine, which ports are listening and what scheduled jobs exist, without changing anything". The AI proposes troubleshooting commands and can use ps, df, ss, docker and journalctl for checks. The commands and output appear in the terminal for you to read and verify. Each time you ask the AI for help, recent terminal output is sent to the cloud AI along with your message, so remove sensitive information first. The inventory page is still yours to write and keep up to date; the AI saves you from memorizing commands and helps you make sense of the output.

The limits:

Which handover jobs the product can help with

Job while you take overWhat you can useWhat you or the owner still have to do
Inventory processes, ports, and resources on a Linux hostRun the commands above in the SSH terminal on the same page, or ask the AI to help read the outputCheck that you were allowed to run them, and match the host facts to the real business
Explain a redacted error or logGive the AI the question and the context, then check that the answer is grounded in this machineDecide whether the evidence is enough. If it is not, collect more. Do not treat a guess as the cause
Hand over a cloud account, a facility, or a path that crosses bothUse this checklist to ask the person who has the access for the missing recordsConsole changes, ownership, and cross-environment dependencies stay with someone who is allowed to confirm them. The product does not discover every cloud asset, out-of-band port, or device on the network
Hear about a problem when you are away from the deskA patrol can push an exception to a linked Telegram private chat. Receiving that alert does not require the SSH key to be stored in the cloudAn alert is not a root cause and not a handover record. Remote SSH from Telegram does need a cloud-stored key for that server
Keep a record of a terminal command that was checkedA command that is run or blocked after the safety check can be written to the operation audit: who ran it, which server, the risk tier, the command text, the status, and a short output summaryThat row is not a handover file. The summary is not the full output, and it does not prove who owns the business, that a backup restores, or that access was removed

If you ask the AI only to look, or to wait before acting, that is a request for this pass. It is not a separate access control. Whether a command runs is still the behavior under "Where OpsMate fits" and the FAQ.

What this page and the product do not do for you

How to keep handover evidence someone else can recheck

For each conclusion, keep the time, the environment, who ran the check, the command or console it came from, a redacted result, and what is still unconfirmed. Store that record where the company allows. Do not publish account names or full configs. The environment table and this table are for the person taking over to fill in. They are not a finished product feature. The product does not generate a handover case file, and it does not fill cloud, facility, or restore checks into these tables.

What you want to confirmEvidence you can keepWhat does not replace it
Which service a domain actually reachesThe entry config, listening port, and process from the same check, reviewed by the person who owns the businessAn architecture diagram with no time on it
Whether a backup can be restoredThe isolated restore steps, when they finished, and a check that the business data worksA backup job that only says it succeeded
Whether a departed person's access is goneThe cloud and host access changes, plus a login you verified yourself, with the systems you did not cover written downChanging the root password once

The illustrative example below is not a customer case and not a public record of a real handover. This page has no measured production drill and no customer result. There is no separate English depth guide for each environment. When you use the page, keep your own check record with the fields above.

Illustrative example

Illustrative example (not a real customer case): a founder inherits a VPS with 4 GB of memory. The inventory shows Nginx forwarding two domains to ports 3000 and 3001; docker compose ls shows one Compose project with web, worker and postgres containers; ss -ltnp shows postgres listening on 0.0.0.0:5432; root's crontab runs pg_dump at 3 a.m. every day and writes the file to the same disk, with no copy anywhere else; certbot's systemd timer renews the certificate, which has 40 days left; and authorized_keys holds three keys, two of which nobody can identify. The priorities: a backup with no off-site copy that has never been test-restored, a database port that may be exposed to the internet, and unexplained keys. All three fixes (copying backups elsewhere, tightening the security group or database listen address, revoking keys) are changes to plan: assess the impact, prepare a rollback, then act, with professional help if needed.

AI diagnostic prompt

I've just inherited this server and don't know what runs on it. Without changing anything, check the OS version and resource overview, running services, listening ports and the processes behind them, Docker containers and Compose projects, Nginx domains and proxy targets, cron jobs and systemd timers, likely backup jobs, certificate expiry, and which users and SSH keys exist. Turn this into a draft inventory, marking what is confirmed, what is a guess and what is missing. Do not modify, delete or restart anything. If you find a risk that needs action, explain the impact and a rollback plan first and wait for my approval.

FAQ

My website suddenly went down and I'm not technical. What should I do?

First check from your own phone or laptop whether the site is down for everyone, and note the time it started. Don't reboot the server yet: check whether the machine is up, whether the disk is full and whether the app is running, and save that output for whoever helps you, because a full reboot can wipe the evidence. Site down at 2 a.m. gives the 15-minute order. If you cannot use SSH, start with the site is suddenly down and you don't know servers.

The freelancer who set up our server left. How do I take over?

Start with accounts, not the server: make sure the company owns the cloud account, the domain registrar and the code repository, and turn on two-factor authentication. Then record what runs on the machine and who can log in, and only after that remove the freelancer's keys and rotate any secrets they handled. Contractor or ex-employee left? How to revoke their server access covers the order that won't lock you out.

I run a small business and can't afford a sysadmin. How do I keep my server running?

Most small teams get by with three habits: a 10-minute weekly check of disk, memory, failed services and backups; an outside uptime alert that reaches your phone; and someone you can pay per incident for the hard cases. Restore a backup on another machine at least once, because a backup nobody has restored is only a guess. The small-team VPS monitoring checklist lists what to watch.

Try OpsMate

500 free AI calls per month and unlimited servers. The desktop app keeps your SSH credentials on your own machine by default.

References

Need help interpreting the evidence?

OpsMate (itops.sh) helps developers and operators investigate with AI, and you review the evidence yourself. Each time you ask the AI for help, recent terminal output is sent to the cloud AI along with your message, so remove sensitive information first.

Start free Desktop with local credentials

Troubleshooting guides