System Health

What the server is doing, how hard it is working, and what will run out first.

For: System AdminLast checked: 2026-11-13Open this page in Obtempera ↗

Choose a window: 6 hours, 24 hours (the default), 7 days or 30 days. The banner says Healthy, Worth watching or Needs attention, how many minutes ago the last reading was taken and how many readings fall in the window, and the Workers, Customers and documents on the platform. A row of checks follows, each with its state and a line of detail.

The charts and tables

  • The machine — disk used, memory available, CPU load against the number of cores, the database.

  • The load — requests a minute, average and worst response time, server errors.

  • The queues — email waiting, compliance jobs waiting.

  • Background workers (a check in the row) — the five workers that run all the time (the email sender, the compliance and training workers, the document recogniser and the Agent Builder) each record when they last went round. One that has not gone round for its usual time (10 minutes for the training worker, 20 for the compliance worker, 30 for the others) reads Worth watching; three times that, Needs attention, with which one is behind. The overnight monitor watches the same records.

  • Uploads waiting for the malware scan (a check in the row): how many uploaded files are held until their first virus check. Nobody can open an upload, and its reviewer does not see it, until it is checked. The five-minute scan normally clears them; the check reads Needs attention when files are waiting and that scan is not scanning (no scanner is set up or it cannot be reached: it is then recorded as skipped, with the reason, never as a success) or has not succeeded for 30 minutes. 50 or more waiting reads Worth watching, 200 or more Needs attention.

  • Server errors (only when there were any in the window) — by area of the API, with errors, requests and rate, then When: the 20 most recent minutes that had errors. Paths are not kept, because they carry Worker and Customer identifiers; the page gives the log command to read the errors themselves.

  • Where the time goes — the busiest areas of the API, heaviest first, with requests, average and worst time, and errors.

  • Backups — three questions, each with what to do if the answer is no:

    • The application: how old the newest backup is, its size, how many sets are kept, and whether the newest was read back to prove it.
    • The other sites (the training platform, staging and the marketing site): how old their newest backup is and how many sets are kept.
    • Off this server: when the last copy left, its size, where it went, which sets were copied, and whether it was encrypted.

    On a machine that only holds backups pulled down from the server, this section shows those copies instead: the newest, how many sets are held and their size.

Fresh readings

The sampler takes a reading every five minutes, and the page refreshes itself every two. Check now (top right) takes a reading straight away and says Checked just now.; within a minute of the last reading it shows that one instead and says A reading under a minute old is shown.

If the last reading is more than 20 minutes old, the sampler is treated as stopped: the banner says how old the reading is and that everything below is from then, not now, and a message gives the cron line that runs the sampler.

Its limit

It can tell you the machine is struggling. It cannot tell you the machine is down; that needs a check from somewhere else.