PricingConsole

What we observe

Last updated 15 November 2026

View .md

The charts in the console and the metrics commands show how your Machines, Volumes, databases and streams are used over time. This page says what Tonbo measures to draw them, where it measures it, how long it keeps it, and who can see it.

Measured from outside your programs

Every figure is counted by Tonbo's own software around your workload: on the host that runs a Machine, in the program that serves a Volume to it, at the network edge, or at the gateway in front of your databases and streams. Tonbo counts and times your requests; it does not look at what they carry. It never reads a Machine's memory, a disk's or Volume's contents, or the bodies of your requests to produce a metric.

One Tonbo program does run inside every Machine: tonbo-guest, which starts the Machine and then your program. It tells the host about the Machine's lifecycle: that your program started, that it exited and with which code or signal, that it could not start and at which step, the wake holds you set and the background processes you started that ended, and the steps of a sleep and a resume. When a Machine stops or sleeps, it also reports, for each Tonbo database your program had open, its id and how many of its committed bytes were not yet synced, so that nothing is lost. The commands and file calls you send a Machine return what you asked for. Beyond these, it sends nothing about your files, your memory or your processes.

Machines

  • Awake time and wakes: when each wake started and ended, and whether it was refused, from Tonbo's record of the Machine's wakes.
  • How long a wake took: for a wake a request caused, from the moment the edge received the request to your program's first byte; otherwise from the wake to tonbo-guest's report that your program started. Timed at the edge and on the host.
  • CPU and memory: the CPU time and memory the Machine's virtual machine used, read from the host's accounting of it about every 30 seconds, and averaged over the time it was awake.
  • Network: the bytes on the Machine's network interface, counted on the host, and separately the bytes it exchanged with the internet, which Tonbo's network counts per connection without reading what the connection carries.
  • Disks: how often the Machine hit its shape's disk limit, counted on the host, never inside the Machine: the requests to its Volume that the limit held back, and the time all of its processes were stalled on its own disks at once, with the bytes and operations on those disks. Read with the Machine's other counters, about every 30 seconds, and starting with the hour its recording went live.

To read them:

tonbo machine metrics app --window 7d
curl -sS "https://api.tonbo.dev/v1/machines/app/metrics?window=7d" \
  -H "Authorization: Bearer $TONBO_API_KEY"

Volumes

Stored size: your Volumes' live bytes, the compressed blocks their latest state references (not the size you give a Volume), averaged over each UTC day. It is the quantity Volume storageis billed for, read from the same measurements. A Volume's page also shows, as its current state rather than as a series, whether its writes are being held back and why.

Read and write latency: how long the program that serves a Volume took to answer its Machine's reads and writes, the median and the 99th percentile per UTC hour or day, from a request's arrival at that program to its reply, measured on the host and never inside the Machine. A write held back counts its wait. The account's figure comes from every Volume's timings put together, never an average of percentiles, and a step with no timed request has none; it starts with the hour its recording went live. One Volume's is in tonbo volume metrics.

tonbo metrics volumes --window 30d
curl -sS "https://api.tonbo.dev/v1/metrics/volumes?window=30d" \
  -H "Authorization: Bearer $TONBO_API_KEY"

Databases and streams

  • Stored size: the log and snapshots a database or stream keeps, sampled by its region's meter.
  • Writes and reads: write units, bytes and records written, and the reads and live reads the gateway answered with success and the bytes they streamed, counted at the gateway.
  • Server latency: how long Tonbo's servers took to answer your appends and catch-up reads, the median and the 99th percentile, from a request's arrival at the gateway to its response headers. Your network adds to this; live reads are not timed.
tonbo database metrics notes --window 24h
curl -sS "https://api.tonbo.dev/v1/databases/notes/metrics?window=24h" \
  -H "Authorization: Bearer $TONBO_API_KEY"
tonbo stream metrics events --window 24h
curl -sS "https://api.tonbo.dev/v1/streams/events/metrics?window=24h" \
  -H "Authorization: Bearer $TONBO_API_KEY"

Who can see them

Any member of the account can read every metric. They carry no money, so they need no more than reading the account. A key narrowed to some databases or streams reads only theirs, and a narrowed key never reads a Machine's. What things cost is in Usage & billing, which owners and admins read.

What Tonbo's operators see

Beyond these metrics, Tonbo's operators can see the records Tonbo keeps to run your resources (their names, settings, states, wakes and usage) and the logs of Tonbo's own services, which they read through a reviewed workflow. Your content is not among them: Tonbo processes it only to run the service for you and does not read it for its own purposes (see the Privacy Policy, section 1).

Metrics and billing are separate

No metric is ever used to bill you. Billing has meters of its own: a Machine's CPU time, memory and scratch disk from the host's accounting, its bytes sent to the internet, a Volume's stored size, and a database's or stream's write units and stored size. Some charts show the same quantities, read from the same measurements, but nothing a chart draws flows back into a bill.

How long Tonbo keeps them

Hourly series answer for at least the last 8 days and daily series for at least the last 92. Server latency is kept exactly that long: 8 days by the hour and 92 days by the day.