I wanted one monitoring box that raises its own alerts, so I went back to the TICK stack instead of wiring Prometheus, Alertmanager and Grafana together. A TICK stack install gives you four InfluxData packages that already know about each other: Telegraf collects, InfluxDB stores, Chronograf draws and Kapacitor decides when something’s wrong. My first attempt still broke in the most boring way possible, because the stock Telegraf config on Ubuntu ships without a single output and the service crash-loops until you give it one.
These notes cover the signed InfluxData repository, all four services, Telegraf metrics landing in InfluxDB, Chronograf wired to InfluxDB and Kapacitor, a host dashboard, a CPU alert I actually fired with a load test, and ufw rules that expose only Chronograf. You’ll also see where I think TICK still beats InfluxDB 3. I ran everything in November 2026 on an Ubuntu 26.04.1 LTS VM called box1 with InfluxDB 1.13.2 and the matching Chronograf, Kapacitor and Telegraf builds, repeated it on Ubuntu 24.04, and every output below comes from those two boxes.
Mapping what each letter of TICK does
All four TICK components run on one host in this build, and only one of them needs to face your network. Telegraf pushes line protocol into InfluxDB on 8086, Kapacitor subscribes to InfluxDB so every new point is streamed to it on 9092, and Chronograf talks to both over their HTTP APIs while serving its own UI on 8888. The subscription is the part people miss. Kapacitor doesn’t poll for data, InfluxDB pushes it, so the stream only flows once Kapacitor has registered that subscription. Kapacitor’s packaged config waits up to 5 minutes for InfluxDB at startup and re-syncs subscriptions every minute, so a slow InfluxDB start delays alerts rather than silently breaking them.

The yellow note in the diagram is the crash loop I mentioned, and it gets its own section below. Each letter has one job and, apart from Telegraf, one port:
- T, Telegraf: the agent, no listening port by default
- I, InfluxDB 1.x: the time-series database, HTTP API on 8086
- C, Chronograf: dashboards and the alert rule builder on 8888
- K, Kapacitor: stream processing and alerting on 9092
InfluxDB also binds 8088 for backup and restore RPC, but only on 127.0.0.1, so there’s nothing to firewall there. I confirmed that with ss after the install rather than trusting the docs, and the port check further down shows it.
Choosing TICK over TIG or InfluxDB 3
Is the TICK stack legacy? Not in the sense that matters, because InfluxData shipped a new InfluxDB 1.13 patch on 30 October 2026 and new Chronograf and Kapacitor builds on 4 November 2026, and the Kapacitor release notes show steady fixes rather than a freeze. It is not where InfluxData points new users, though, since that’s InfluxDB 3. So the choice comes down to which problem you’re solving, and I’d pick TICK when one of these is true:
- Alerting matters more than pretty graphs
- You already write InfluxQL and keep 1.x data
- One small box, one vendor, four apt packages
- You need stream tasks fed by 1.x subscriptions
If you’re starting fresh, want SQL, and care more about long-term storage and Grafana panels than a built-in alert engine, the InfluxDB 3, Telegraf and Grafana build is the better starting point. Anyone already on InfluxDB 2 should read the InfluxDB 2 to 3 migration notes before deciding anything, because the upgrade path matters more than the dashboard tool. I haven’t measured how much of Chronograf and Kapacitor works against InfluxDB 3’s v1 compatibility API yet. That gets its own lab later in this series, and I won’t guess at it here.
Prerequisites
Nothing exotic is needed, and a small VM is enough for a handful of hosts. My lab box had 2 vCPUs and 4 GB of RAM, and the whole stack used about 1.1 GB of disk once installed. Have these ready before you start:
- Ubuntu 26.04 LTS or 24.04 LTS, freshly installed
- A sudo user (mine is philip on box1)
- Outbound HTTPS to repos.influxdata.com
- Your admin subnet for the Chronograf firewall rule
Sizing grows with series cardinality, not with the number of hosts, because every unique tag combination is a series InfluxDB keeps an index entry for. InfluxData’s hardware sizing guide puts under 5,000 writes per second and under 100,000 series at 2-4 cores and 2-4 GB of RAM, and the AWS hosting notes on its install page ask for 8 GB per machine. Treat my 4 GB box as a floor for a lab or a small office, not a production target, and watch memory once you add more agents.
Adding the InfluxData repository with the signed key
Ubuntu’s own archive carries an influxdb package too, and that’s a trap. On 26.04 it’s 1.6.7~rc0-2build2, a years-old release candidate with no Chronograf or Kapacitor to match, so the InfluxData repository is the only sane source. I followed the official InfluxDB 1.x install page, which checks the key’s fingerprint before trusting it. If the fingerprint doesn’t match, the grep -q fails and nothing is written to the keyring, which is the whole point of that long middle line. The echo line still writes the source file in that case, so delete /etc/apt/sources.list.d/influxdata.list before running apt-get update if the check ever fails.
curl --silent --location -O https://repos.influxdata.com/influxdata-archive.key
gpg --show-keys --with-fingerprint --with-colons ./influxdata-archive.key 2>&1 | grep -q '^fpr::*24C975CBA61A024EE1B631787C3D57159FC2F927:$' && cat influxdata-archive.key | gpg --dearmor | sudo tee /etc/apt/keyrings/influxdata-archive.gpg > /dev/null
echo 'deb [signed-by=/etc/apt/keyrings/influxdata-archive.gpg] https://repos.influxdata.com/debian stable main' | sudo tee /etc/apt/sources.list.d/influxdata.list
You can eyeball the key yourself before going further. One signing subkey expired in January 2026 and a newer one runs until 2029, and apt is happy with that because the 2025 subkey signs the current repository metadata.
gpg --show-keys --with-fingerprint ./influxdata-archive.key
The fingerprint on the second line should match the one the grep checked. If it shows a different key, stop there, because whatever key you trust here can sign packages that apt then installs as root.
pub rsa4096 2023-01-18 [SC]
24C9 75CB A61A 024E E1B6 3178 7C3D 5715 9FC2 F927
uid InfluxData Package Signing Key <[email protected]>
sub rsa4096 2023-01-18 [S] [expired: 2026-01-17]
sub rsa4096 2025-07-10 [S] [expires: 2029-01-17]
Refresh the package index next, then confirm that apt picks the InfluxData builds over Ubuntu’s old one. The last command prints Ubuntu’s own candidate so you can see it lose.
sudo apt-get update
apt-cache policy influxdb chronograf kapacitor telegraf | grep -E "^[a-z]|Candidate"
apt-cache policy influxdb | grep -E "ubuntu.com" -B1 | head -2
Every candidate should be an InfluxData version. The last two lines are Ubuntu’s stale package, still sitting in resolute/universe with a lower version, so apt never chooses it while the InfluxData repo is configured.
influxdb:
Candidate: 1.13.2-1
chronograf:
Candidate: 1.11.7-1
kapacitor:
Candidate: 1.8.9-1
telegraf:
Candidate: 1.40.3-1
1.6.7~rc0-2build2 500
500 http://archive.ubuntu.com/ubuntu resolute/universe amd64 Packages
A candidate of 1.6.7~rc0 means the repository line didn’t take, usually because of a typo in the signed-by path. Fix that before installing, since removing a half-configured 1.6 database later is more work than checking now.
Running the TICK stack install
All four packages come down in one transaction. That’s about 299 MB of archives, which unpack to roughly 1.1 GB. If you only need one piece, say a search for “install Chronograf Ubuntu” or “install Kapacitor” brought you here because Grafana already draws your dashboards, the same repository works and you name just that package in the apt-get install line. Chronograf still needs an InfluxDB to point at and Kapacitor needs one to subscribe to, so a single-package install only makes sense when InfluxDB already runs somewhere.
sudo apt-get install -y influxdb chronograf kapacitor telegraf
A fifth package rides along, influxdata-archive-keyring. It drops its own keyring under /usr/share/keyrings and wanted to write /etc/apt/sources.list.d/influxdata.list as well, but because I had already created that file it saved its version as influxdata.list.ucf-dist and left mine alone. You end up with one active source either way, so there’s nothing to clean up. These are the lines worth reading from the install:
0 upgraded, 5 newly installed, 0 to remove and 2 not upgraded.
Need to get 299 MB of archives.
After this operation, 1101 MB of additional disk space will be used.
Setting up influxdb (1.13.2-1) ...
Created symlink '/etc/systemd/system/influxd.service' → '/usr/lib/systemd/system/influxdb.service'.
Created symlink '/etc/systemd/system/multi-user.target.wants/influxdb.service' → '/usr/lib/systemd/system/influxdb.service'.
Setting up influxdata-archive-keyring (2026.01.09) ...
Setting up telegraf (1.40.3-1) ...
Created symlink '/etc/systemd/system/multi-user.target.wants/telegraf.service' → '/usr/lib/systemd/system/telegraf.service'.
Setting up kapacitor (1.8.9-1) ...
Created symlink '/etc/systemd/system/multi-user.target.wants/kapacitor.service' → '/usr/lib/systemd/system/kapacitor.service'.
Setting up chronograf (1.11.7-1) ...
Created symlink '/etc/systemd/system/multi-user.target.wants/chronograf.service' → '/usr/lib/systemd/system/chronograf.service'.
Every unit is enabled, yet that does not mean every unit is running. I checked straight after the install and only Chronograf had started, which surprised me.
systemctl is-enabled influxdb telegraf kapacitor chronograf
systemctl is-active influxdb telegraf kapacitor chronograf
Four times enabled, then inactive for InfluxDB, Telegraf and Kapacitor. That’s a packaging choice: every postinst enables its unit, but only Chronograf’s also starts it, which makes sense because Chronograf needs nothing else to serve its setup wizard.
enabled
enabled
enabled
enabled
inactive
inactive
inactive
active
So you start the other three by hand, and InfluxDB goes first. Telegraf and Kapacitor both expect it to answer on 8086, and while Kapacitor would keep retrying for up to 5 minutes (its startup-timeout), starting them in order keeps Kapacitor’s log free of connection errors.
Starting InfluxDB and checking the API
InfluxDB 1.x starts with authentication off. That’s acceptable here only because ufw will keep 8086 off the network, and turning auth on is covered later in this series. Start it and hit the ping endpoint.
sudo systemctl enable --now influxdb
systemctl status influxdb --no-pager | head -12
curl -sI http://localhost:8086/ping
A 204 No Content with an X-Influxdb-Version header is the healthy answer. It also tells you which build you’re really talking to, which is the quickest way to catch Ubuntu’s old package if it slipped in.
● influxdb.service - InfluxDB is an open-source, distributed, time series database
Loaded: loaded (/usr/lib/systemd/system/influxdb.service; enabled; preset: enabled)
Active: active (running) since Sat 2026-11-07 13:39:03 UTC; 3s ago
Invocation: cb9b328a874146118af0789bb099183b
Docs: https://docs.influxdata.com/influxdb/
Process: 2982 ExecStart=/usr/lib/influxdb/scripts/influxd-systemd-start.sh (code=exited, status=0/SUCCESS)
Main PID: 2998 (influxd)
Tasks: 9 (limit: 3666)
Memory: 24.9M (peak: 45.2M)
CPU: 366ms
CGroup: /system.slice/influxdb.service
└─2998 /usr/bin/influxd -config /etc/influxdb/influxdb.conf
HTTP/1.1 204 No Content
Content-Type: application/json
Request-Id: 9a3718f5-dab0-11f1-8002-bc241102f30f
X-Influxdb-Build: OSS
X-Influxdb-Version: 1.13.2
X-Request-Id: 9a3718f5-dab0-11f1-8002-bc241102f30f
Date: Sat, 07 Nov 2026 13:39:06 GMT
InfluxDB sat at 25 MB of RAM with nothing written yet. If you’d rather run the 2.x line with its own UI and Flux, that’s a different package and a different guide, and the InfluxDB v2 install covers it. Chronograf’s InfluxQL dashboards and Kapacitor’s subscriptions assume 1.x, though, so stay on 1.x for this stack.
Pointing Telegraf at InfluxDB
This is where my first try fell over. The packaged /etc/telegraf/telegraf.conf is 15,293 lines long, and once you strip the comments it enables eight inputs (cpu, disk, diskio, kernel, mem, processes, swap, system) and zero outputs. Starting Telegraf in that state fails at once.
sudo systemctl start telegraf
sudo journalctl -u telegraf --no-pager -o cat | grep "E!" | head -2
systemd retries five times, gives up with “Start request repeated too quickly”, and the journal names the real cause.

That error is the whole story, and the fix is a single small file. The service isn’t broken and nothing needs reinstalling.
Error: “no outputs found, probably invalid config file provided”
Nothing is wrong with the file’s syntax. Telegraf simply refuses to run without somewhere to send metrics, so the error reads worse than it is. Rather than uncomment the right block among 15,000 lines, I add a drop-in under /etc/telegraf/telegraf.d/, which the systemd unit already loads with -config-directory. A drop-in also stays out of the way of package upgrades, while an edited main file can trigger dpkg’s conffile prompt whenever a new Telegraf ships a changed default config. Open it:
sudo vim /etc/telegraf/telegraf.d/influxdb.conf
Paste the output plus two network inputs. The network inputs are there because Chronograf’s host dashboard has a network panel and the default inputs don’t feed it.
[[outputs.influxdb]]
urls = ["http://127.0.0.1:8086"]
database = "telegraf"
retention_policy = ""
timeout = "5s"
[[inputs.net]]
[[inputs.netstat]]
Test the merged configuration before restarting anything. --test gathers once and prints line protocol to stdout instead of writing it, so a broken plugin shows up here and not in the journal.
sudo telegraf --config /etc/telegraf/telegraf.conf --config-directory /etc/telegraf/telegraf.d --test 2>&1 | grep -E "^(> (cpu,cpu=cpu-total|mem|system|net,|netstat))|E!|W!" | head -8
You’ll get three warnings and no errors. The strict-environment notice only matters if your config uses $VARS, and the skip_processors_after_aggregators one is a heads-up about a default that changes in 1.41. The metric lines prove the inputs work and that the host tag is the machine’s hostname, which is what Chronograf groups hosts by.
2026-11-07T13:39:41Z W! Strict environment variable handling is the new default starting with v1.38.0! If your configuration does not work with strict handling please explicitly add the --non-strict-env-handling flag to switch to the previous behavior!
2026-11-07T13:39:41Z W! Outputs are not used in testing mode!
2026-11-07T13:39:41Z W! [agent] The default value of 'skip_processors_after_aggregators' will change to 'true' with Telegraf v1.41.0! If you need the current default behavior, please explicitly set the option to 'false'!
> net,host=box1,interface=eth0 bytes_recv=312348168u,bytes_sent=7681955u,drop_in=198u,drop_out=0u,err_in=0u,err_out=0u,packets_recv=160163u,packets_sent=96671u,speed=-1i 1794058781000000000
> system,host=box1 load1=0.16,load15=0.06,load5=0.15,n_cpus=2i,n_physical_cpus=2i 1794058781000000000
> system,host=box1 uptime=184u 1794058781000000000
> system,host=box1 uptime_format=" 0:03" 1794058781000000000
> mem,host=box1 active=365248512u,available=3527073792u,available_percent=86.01291938817198,buffered=27115520u,cached=2052308992u,commit_limit=2050314240u,committed_as=894259200u,dirty=811008u,free=1702133760u,high_free=0u,high_total=0u,huge_page_size=2097152u,huge_pages_free=0u,huge_pages_total=0u,inactive=1784115200u,low_free=0u,low_total=0u,mapped=315695104u,page_tables=5050368u,shared=1064960u,slab=127942656u,sreclaimable=72577024u,sunreclaim=55365632u,swap_cached=0u,swap_free=0u,swap_total=0u,total=4100632576u,used=573558784u,used_percent=13.987080611828022,vmalloc_chunk=0u,vmalloc_total=35184372087808u,vmalloc_used=12333056u,write_back=0u,write_back_tmp=0u 1794058781000000000
Because systemd gave up on the unit, clear its failed state before enabling it, then give Telegraf two flush intervals (20 seconds) to create the database. A unit that hit its start limit can refuse another start until the limit window passes, so the reset saves you a confusing second failure.
sudo systemctl reset-failed telegraf
sudo systemctl enable --now telegraf
sleep 25
influx -execute "SHOW DATABASES"
influx -database telegraf -execute "SHOW MEASUREMENTS"
The sleep 25 covers two flush intervals plus a little slack. Telegraf created the telegraf database on its first write. All ten measurements are there, including net and netstat from the drop-in.
name: databases
name
----
_internal
telegraf
name: measurements
name
----
cpu
disk
diskio
kernel
mem
net
netstat
processes
swap
system
Collecting from Docker, Nginx or PostgreSQL is the same pattern with more input blocks. The Telegraf input plugins guide has tested configs for all three, and they write to 1.x unchanged because the output block is the only part that knows which InfluxDB it talks to.
Querying the metrics with InfluxQL
Before touching a dashboard, I like one query from the CLI so I know the data is shaped the way I think it is. This one averages idle CPU in 30-second buckets over the last two minutes.
influx -database telegraf -precision rfc3339 -execute "SELECT mean(usage_idle) FROM cpu WHERE cpu = 'cpu-total' AND time > now() - 2m GROUP BY time(30s)"
The empty buckets aren’t a bug. Telegraf had only been writing for about 30 seconds, so the earlier windows have no points and InfluxQL prints them with no value.
name: cpu
time mean
---- ----
2026-11-07T13:38:00Z
2026-11-07T13:38:30Z
2026-11-07T13:39:00Z
2026-11-07T13:39:30Z
2026-11-07T13:40:00Z 99.47441171762651
The autogen retention policy that Telegraf’s database gets is infinite (0s duration), so on a monitoring box the disk fills slowly until the day it doesn’t. I cap it at 30 days with daily shards. I set the shard duration explicitly because InfluxDB expires whole shard groups at once, which is far cheaper than deleting points, so smaller groups keep disk use close to the 30-day line instead of overshooting by a group’s length.
influx -execute "ALTER RETENTION POLICY autogen ON telegraf DURATION 30d SHARD DURATION 1d DEFAULT"
influx -database telegraf -execute "SHOW RETENTION POLICIES"
The new duration reads 720h0m0s, and the shard group is now 24 hours. The futureWriteLimit and pastWriteLimit columns stay at 0s (no limit) unless you set them.
name duration shardGroupDuration replicaN futureWriteLimit pastWriteLimit default
---- -------- ------------------ -------- ---------------- -------------- -------
autogen 720h0m0s 24h0m0s 1 0s 0s true
Pick the duration from how far back you actually look. Thirty days covers month-over-month comparisons for most small setups. If you need a year, downsample with a continuous query instead of keeping 10-second points for 365 days.
Starting Kapacitor and checking its subscription
Kapacitor’s packaged config already points at InfluxDB on localhost, so there’s nothing to edit for a single-box stack. I still confirmed the [[influxdb]] section before starting it, because a remote InfluxDB or one with auth needs those lines changed.
grep -nA12 "^\[\[influxdb\]\]" /etc/kapacitor/kapacitor.conf | grep -vE "^\S*-\s*#|^\S*-\s*$"
sudo systemctl enable --now kapacitor
kapacitor version
curl -s -o /dev/null -w "%{http_code}\n" http://localhost:9092/kapacitor/v1/ping
influx -execute "SHOW SUBSCRIPTIONS"
The interesting part is the last block. Within seconds of starting, Kapacitor registered a subscription on every retention policy in InfluxDB, and that is how it receives the stream it alerts on.
155:[[influxdb]]
159- enabled = true
160- default = true
161- name = "localhost"
162- urls = ["http://localhost:8086"]
163- username = ""
164- password = ""
165- timeout = 0
Kapacitor OSS 1.8.9 (git: HEAD acdeb951399e424437df095eea22d71422df8dd0)
204
name: _internal
retention_policy name mode destinations
---------------- ---- ---- ------------
monitor kapacitor-c9b66225-af04-4f36-b063-864930672bc8 ANY [http://localhost:9092]
name: telegraf
retention_policy name mode destinations
---------------- ---- ---- ------------
autogen kapacitor-c9b66225-af04-4f36-b063-864930672bc8 ANY [http://localhost:9092]
If SHOW SUBSCRIPTIONS comes back empty, give it a minute, because Kapacitor re-syncs subscriptions every 60 seconds (subscriptions-sync-interval). Still empty after that means Kapacitor can’t reach InfluxDB, and sudo tail -n 50 /var/log/kapacitor/kapacitor.log will say why, because the packaged config logs to that file rather than the journal. With that, every port is in place, so check what listens where.
sudo ss -tlnp | grep -E "8086|8088|8888|9092"
Look at the bind addresses. 8088 is loopback only, while 8086, 8888 and 9092 listen on all interfaces, which is exactly why the firewall section matters.
LISTEN 0 4096 127.0.0.1:8088 0.0.0.0:* users:(("influxd",pid=2998,fd=4))
LISTEN 0 4096 *:9092 *:* users:(("kapacitord",pid=3698,fd=10))
LISTEN 0 4096 *:8888 *:* users:(("chronograf",pid=2538,fd=8))
LISTEN 0 4096 *:8086 *:* users:(("influxd",pid=2998,fd=7))
Anyone on the same network could write fake metrics into 8086 or delete Kapacitor tasks through 9092 right now, since neither has authentication yet.
Locking the ports down with ufw
Only Chronograf needs to be reachable, and only from the network you administer from. Telegraf, Kapacitor and Chronograf all reach InfluxDB over localhost here, so 8086 and 9092 stay closed. This is the plan for each port:
- 22/tcp (OpenSSH): open, or you lose the session
- 8888/tcp (Chronograf): admin subnet only
- 8086/tcp (InfluxDB): closed until remote agents need it
- 9092/tcp (Kapacitor): closed, Chronograf uses localhost
The order of commands matters too. Allow SSH first, because enabling ufw over an SSH session without that rule locks you out. If it happens anyway, the VM console lets you run sudo ufw disable and try again. Set your admin subnet once, then add the rules.
export ADMIN_CIDR="192.168.1.0/24"
sudo ufw allow OpenSSH
sudo ufw allow from "${ADMIN_CIDR}" to any port 8888 proto tcp comment "Chronograf UI"
sudo ufw --force enable
sudo ufw status numbered
You should see three rules and nothing for 8086 or 9092. The v6 line exists because ufw mirrors the OpenSSH rule for IPv6, while the Chronograf rule stays IPv4-only since the subnet you gave it is IPv4.
Status: active
To Action From
-- ------ ----
[ 1] OpenSSH ALLOW IN Anywhere
[ 2] 8888/tcp ALLOW IN 192.168.1.0/24 # Chronograf UI
[ 3] OpenSSH (v6) ALLOW IN Anywhere (v6)
I tested it from another machine on the LAN rather than trusting the rule list. Chronograf answers, and the other two ports time out, which curl reports as 000.
for p in "8888/" "8086/ping" "9092/kapacitor/v1/ping"; do curl -s -o /dev/null -w "$p %{http_code}\n" --max-time 5 http://192.168.1.133:$p; done
That’s the result you want before letting anyone near the UI. A 204 on either of the other ports would mean a leftover allow rule is exposing InfluxDB or Kapacitor, so read sudo ufw status numbered again before moving on.
8888/ 200
8086/ping 000
9092/kapacitor/v1/ping 000
One caveat the firewall can’t fix: Chronograf itself has no login yet, and it proxies queries to InfluxDB and rules to Kapacitor, so anyone in ADMIN_CIDR can run DROP DATABASE from the Data Explorer or delete alert tasks. Keep that subnet tight until you add Chronograf authentication, which a later part of this series covers. When other servers run Telegraf and write here, add one more rule for 8086 scoped to their subnet, and turn on InfluxDB authentication first. The ufw command reference has the syntax for deleting and reordering rules if you get one wrong.
Connecting Chronograf to InfluxDB and Kapacitor
Open http://192.168.1.133:8888, using your own server’s address, and Chronograf greets you with a five-step wizard. Click Get Started. The InfluxDB Connection step is already filled in with localhost and the telegraf database, so I only renamed the connection to “box1 InfluxDB” and then did this:
- Leave Connection URL at
http://localhost:8086 - Keep Telegraf Database Name as
telegraf - Click Add Connection

- 1Connection URL
- 2Telegraf Database Name
- 3Add Connection
Username and password stay empty because auth is off. The URL says localhost on purpose, because Chronograf’s server makes these calls and your browser never does, so the closed 8086 port doesn’t get in its way.
The Dashboards step looks at the measurements it found and suggests only System out of 29 templates. Select System, then click Create 1 Dashboard.

- 1System
- 2Create 1 Dashboard
Next comes Kapacitor. The URL defaults to http://localhost:9092/, which is right for this box, so I named it “box1 Kapacitor”, left Kapacitor URL alone and clicked Continue.

- 1Kapacitor URL
- 2Continue
The wizard ends on Setup Complete with a View All Connections button. If you skip the Kapacitor step, the alerting pages later say no Kapacitor is configured, and you add it from the wrench icon, which opens the Configuration page.
Watching the first host dashboard
The eye icon at the top of the sidebar opens the Host List, built entirely from the system measurement. box1 shows up with a green status dot, its CPU and load, and a system app link that only appears because the System dashboard exists.

Clicking box1 opens the host dashboard. I took this capture straight after the alert test further down, so the CPU Usage panel shows the spike to 100% and System Load climbing towards 2, which is what two busy cores look like.

The Network Mb/s panel is fed by the [[inputs.net]] block from the drop-in. Without it that panel stays empty, and you’d waste time wondering whether Chronograf or Telegraf is broken. Disk used % sits between 15 and 22% across the two filesystems, which is a useful baseline to remember once the retention policy starts trimming data.
For ad-hoc questions, the Data Explorer (the graph icon) takes raw InfluxQL. I pasted the same CPU query as before, widened to 15 minutes, into the InfluxQL query editor and clicked Submit Query.
SELECT mean("usage_idle") FROM "telegraf"."autogen"."cpu" WHERE "cpu" = 'cpu-total' AND time > now() - 15m GROUP BY time(30s)
The status line reads “Success!” and the graph shows idle CPU wobbling between 97.5 and 99.5% on an otherwise quiet VM.

- 1InfluxQL query
- 2Submit Query
From here Send to Dashboard turns a query into a dashboard cell, and that’s how I’d build anything beyond the System template.
Building the first Kapacitor CPU alert
Chronograf’s rule builder writes the TICKscript for you, which makes it the gentlest way into Kapacitor. Hover the Alerting icon in the sidebar, choose Alert Rules, click Build Alert Rule, and name the rule (“box1 CPU busy” in my case). Keep the Threshold type, pick telegraf.autogen, the cpu measurement and the cpu-total tag value, then set the condition:
- Tick the usage_idle field
- Change the condition to less than
- Type
70as the Threshold value - Click Save Rule once the handler below is set

- 1usage_idle
- 2less than
- 3Threshold value
- 4Save Rule
Why idle below 70 and not usage above 30? Telegraf doesn’t store a single “busy” field for cpu-total, and usage_idle is the one number that already accounts for user, system, iowait and steal together. Alerting on usage_user alone would miss a box pinned by iowait, so idle is the safer signal. The green band in the preview graph is the alert zone, and on a quiet box the line never gets near it.
Every rule needs somewhere to send the alert. I chose the log handler with /var/log/kapacitor/alerts.log, because the kapacitor user can already write there and it needs no outside account. In the Alert Handlers box, open Add a Handler, pick log, set File Path for Log File to /var/log/kapacitor/alerts.log, and type the template below into the Message box so your log lines match mine.
{{ .ID }} is {{ .Level }}: usage_idle {{ index .Fields "value" }}
.ID is the rule name, .Level is OK or CRITICAL, and value is the usage_idle reading that tripped it.

“Send alert only when condition state changes” was already ticked on my rule, and I’d keep it, since without it a sustained spike writes a CRITICAL line every 10 seconds. Saving creates a Kapacitor task, and you can inspect it from the shell.
kapacitor list tasks
kapacitor show $(kapacitor list tasks | awk 'NR==2{print $1}') | sed -n '1,60p'
It’s a stream task on telegraf.autogen, and the generated TICKscript is readable enough to learn from. The sed keeps the first 60 lines, so the handler nodes at the end are cut off here. Note outputDB = 'chronograf'. Every state change is written back into InfluxDB, and that’s what powers Chronograf’s Alert History page.
ID Type Status Executing Databases and Retention Policies
chronograf-v1-70d10408-cf12-4101-ab11-554d78e4c63b stream enabled true ["telegraf"."autogen"]
ID: chronograf-v1-70d10408-cf12-4101-ab11-554d78e4c63b
Error:
Template:
Type: stream
Status: enabled
Executing: true
Created: 07 Nov 26 13:46 UTC
Modified: 07 Nov 26 13:46 UTC
LastEnabled: 07 Nov 26 13:46 UTC
Databases Retention Policies: ["telegraf"."autogen"]
TICKscript:
var db = 'telegraf'
var rp = 'autogen'
var measurement = 'cpu'
var groupBy = []
var whereFilter = lambda: ("cpu" == 'cpu-total') AND isPresent("usage_idle")
var name = 'box1 CPU busy'
var idVar = name
var message = '{{ .ID }} is {{ .Level }}: usage_idle {{ index .Fields "value" }}'
var idTag = 'alertID'
var levelTag = 'level'
var messageField = 'message'
var durationField = 'duration'
var outputDB = 'chronograf'
var outputRP = 'autogen'
var outputMeasurement = 'alerts'
var triggerType = 'threshold'
var crit = 70
var data = stream
|from()
.database(db)
.retentionPolicy(rp)
.measurement(measurement)
.groupBy(groupBy)
.where(whereFilter)
|eval(lambda: "usage_idle")
.as('value')
var trigger = data
|alert()
.crit(lambda: "value" < crit)
.stateChangesOnly()
.message(message)
An alert you’ve never seen fire isn’t an alert, so I loaded both cores for 150 seconds with two yes processes and read the log 40 seconds later. Two processes only works because box1 has 2 vCPUs. On a bigger machine start one per core (nproc tells you how many), or idle CPU never drops below 70 and the alert never fires.
nohup timeout 150 sh -c "yes >/dev/null & yes >/dev/null & wait" >/dev/null 2>&1 &
sleep 40
sudo cat /var/log/kapacitor/alerts.log
Kapacitor caught it within one or two Telegraf intervals of the load starting, with usage_idle at 0 and the level going from OK to CRITICAL. The details field is HTML-escaped JSON, which is ugly in a log but harmless, and it doesn’t affect the alert itself.
{"id":"box1 CPU busy","message":"box1 CPU busy is CRITICAL: usage_idle 0","details":"{\u0026#34;Name\u0026#34;:\u0026#34;cpu\u0026#34;,\u0026#34;TaskName\u0026#34;:\u0026#34;chronograf-v1-70d10408-cf12-4101-ab11-554d78e4c63b\u0026#34;,\u0026#34;Group\u0026#34;:\u0026#34;nil\u0026#34;,\u0026#34;Tags\u0026#34;:{\u0026#34;cpu\u0026#34;:\u0026#34;cpu-total\u0026#34;,\u0026#34;host\u0026#34;:\u0026#34;box1\u0026#34;},\u0026#34;ServerInfo\u0026#34;:{\u0026#34;Hostname\u0026#34;:\u0026#34;localhost\u0026#34;,\u0026#34;ClusterID\u0026#34;:\u0026#34;c9b66225-af04-4f36-b063-864930672bc8\u0026#34;,\u0026#34;ServerID\u0026#34;:\u0026#34;3a001703-9672-4074-b243-87f529cd8291\u0026#34;},\u0026#34;ID\u0026#34;:\u0026#34;box1 CPU busy\u0026#34;,\u0026#34;Fields\u0026#34;:{\u0026#34;value\u0026#34;:0},\u0026#34;Level\u0026#34;:\u0026#34;CRITICAL\u0026#34;,\u0026#34;Time\u0026#34;:\u0026#34;2026-11-07T13:47:20Z\u0026#34;,\u0026#34;Duration\u0026#34;:0,\u0026#34;Message\u0026#34;:\u0026#34;box1 CPU busy is CRITICAL: usage_idle 0\u0026#34;}\n","time":"2026-11-07T13:47:20Z","duration":0,"level":"CRITICAL","data":{"series":[{"name":"cpu","tags":{"cpu":"cpu-total","host":"box1"},"columns":["time","value"],"values":[["2026-11-07T13:47:20Z",0]]}]},"previousLevel":"OK","recoverable":true}
Once the load ended, the recovery came through as an OK event. Chronograf’s Alert History page (Alerting, then Alert History) lists both, the red dot at 0 and the green one at 99.3.

The times there are in my browser’s local zone (UTC+3), which is why they read 16:47 while the log says 13:47Z. The same two events are queryable in the chronograf database, and that’s handy for reporting on alert noise later.
influx -database chronograf -precision rfc3339 -execute "SELECT alertName, level, value FROM alerts"
CRITICAL at 13:47:20 and OK at 13:49:50 are 2.5 minutes apart, which matches the 150-second load almost to the second.
name: alerts
time alertName level value
---- --------- ----- -----
2026-11-07T13:47:20Z box1 CPU busy CRITICAL 0
2026-11-07T13:49:50Z box1 CPU busy OK 99.34869739479018
A log file is a fine first handler, but nobody reads it at 3 a.m. Swap it for Slack, email or a webhook in the same Alert Handlers box. A self-hosted push server such as ntfy behind HTTPS works through the post handler, so you keep alerts on your own hardware.
Verifying every service after a reboot
Remember that the packages were enabled but not started? I wanted proof that the stack comes back on its own, so I checked all four units and their health endpoints, rebooted box1, and checked again. This is the check.
for s in influxdb telegraf kapacitor chronograf; do printf "%-11s %s %s\n" $s $(systemctl is-enabled $s) $(systemctl is-active $s); done
curl -s -o /dev/null -w "influxdb /ping %{http_code}\n" http://localhost:8086/ping
curl -s -o /dev/null -w "kapacitor /ping %{http_code}\n" http://localhost:9092/kapacitor/v1/ping
curl -s -o /dev/null -w "chronograf / %{http_code}\n" http://localhost:8888/
All four should be enabled and active, with two 204 pings and a 200 from Chronograf.

After sudo systemctl reboot I ran a shorter version of the check, with uptime in front so the boot time is visible in the output.
uptime; for s in influxdb telegraf kapacitor chronograf; do printf "%-11s %s\n" $s $(systemctl is-active $s); done
Two minutes after boot the same four units were active again with no manual starts. That’s the bare minimum for a box whose whole job is noticing when other things don’t come back.
13:54:15 up 2 min, 1 user, load average: 0.02, 0.03, 0.01
influxdb active
telegraf active
kapacitor active
chronograf active
Telegraf’s journal for that boot holds zero E! lines, so the drop-in survived the restart exactly as written. Kapacitor came back with its task enabled and executing too, because tasks are stored in /var/lib/kapacitor/kapacitor.db rather than held in memory.
Repeating the install on Ubuntu 24.04
Does anything change on 24.04? I ran the same repository and install commands on a second VM, box2, on Ubuntu 24.04.4 LTS, and the short answer is no. The InfluxData repository is distribution-agnostic (debian stable main), so both releases get identical builds.
influxdb:
Candidate: 1.13.2-1
chronograf:
Candidate: 1.11.7-1
kapacitor:
Candidate: 1.8.9-1
telegraf:
Candidate: 1.40.3-1
1.6.7~rc0-2ubuntu0.24.04.3 500
500 http://archive.ubuntu.com/ubuntu noble-updates/universe amd64 Packages
The same quirks showed up as well. Only Chronograf was running after the install, and when I moved the drop-in aside, Telegraf on box2 failed with the identical no outputs found error. With the same influxdb.conf drop-in in place, everything came up and wrote all ten measurements.
inactive inactive inactive active
influxdb enabled active
telegraf enabled active
kapacitor enabled active
chronograf enabled active
cpu disk diskio kernel mem net netstat processes swap system
Kapacitor OSS 1.8.9 (git: HEAD acdeb951399e424437df095eea22d71422df8dd0)
Chronograf 1.11.7 (git: 50662a3e5992b467afcdce362e75afe3620b8a84)
Telegraf 1.40.3 (git: HEAD@41b45388)
InfluxDB 1.13.2 (git: unknown 64ccba8d279e45a625d6b94873209de7335ba6e5)
The only differences between the two boxes were outside the TICK packages themselves.
| Item | Ubuntu 26.04 (box1) | Ubuntu 24.04 (box2) |
|---|---|---|
| Kernel | 7.0.0-43-generic | 6.8.0-106-generic |
| Ubuntu’s own influxdb package | 1.6.7~rc0-2build2 (resolute/universe) | 1.6.7~rc0-2ubuntu0.24.04.3 (noble-updates/universe) |
| InfluxData builds installed | InfluxDB 1.13.2, Chronograf 1.11.7, Kapacitor 1.8.9, Telegraf 1.40.3 | Same |
| Running right after install | Chronograf only | Chronograf only |
| Telegraf without a drop-in | no outputs found | Same |
So the guide applies to both releases without a single per-version branch. That is not something I can say about most third-party repos in the first months of a new LTS, and it’s one more reason I’m comfortable recommending this stack on 26.04 today.
What I still don’t know
This build answers the install question, but a few things came up in the lab that I couldn’t settle in one sitting. I’d rather list them than pretend:
- How Chronograf behaves against InfluxDB 3’s v1 API
- Whether Kapacitor batch tasks can stand in for subscriptions there
- The RAM curve once 20 or more hosts report
- Whether the keyring package ever overwrites a hand-made source file
The first two decide whether TICK has a future beyond InfluxDB 1.x, and I’ll only answer them with measurements in a later part of the tick-stack series on this site. If you run this today, the next things I’d do are turn on InfluxDB authentication before opening 8086 to other agents, and replace the log handler with something that wakes a person up. Until then, the box you have now already watches itself and tells you when it’s in trouble.