Skip to content

Automation

By default, nothing Seylo proposes changes your cluster until a person approves it. A few kinds of change can run on their own, but only after an admin allows them, one kind at a time, inside limits, until an end date.

Operate → Automation is the one page that answers two questions for the selected cluster: what may Seylo do on its own here, and what did it do? It holds everything that can change a cluster without a person:

Section What it holds
Nightshift fixes it may apply without asking Which Nightshift fixes may run on their own, and the grants signed for this cluster
Savings it may apply without asking The managed savings grant, read-only
Pause and limits One pause for everything, and the limits every grant has
What Seylo did on its own One log of every change made without a person, and every refusal

Changes waiting for approval are not on this page. They are on Approvals; the button at the top shows how many there are, for example 3 waiting for approval, and opens that page.

The first line counts the last 30 days: 4 changes made on its own in 30 days, or Seylo changed nothing on its own here in 30 days. Under it: how many grants are active, whether managed savings is on, and how many changes were undone or refused.

If the page says Unattended changes are off on this deployment: grants are stored, and nothing uses them yet., no grant acts on its own on this Seylo deployment, however many are signed. Every change still goes through approval.

Nightshift fixes it may apply without asking

Section titled “Nightshift fixes it may apply without asking”

Two cards.

Fixes Nightshift may apply is your organisation’s switch and its table:

  • Let proven fixes run on their own, with an Until date at most 90 days away. While it is off, nothing on this card runs on its own.
  • One row per kind of fix, with its risk, how many times it was applied here and verified, and its Mode: Ask first or Auto once proven. How a fix earns autonomy is on fixes, class by class.
  • HPAs at their maximum: one field, Highest maxReplicas (1 to 100), that bounds every maxReplicas raise and the grants for it. See HPAs at their maximum.
  • Under the table, four fixes that run on their own only when their own box is ticked: memory limits sized from evidence, nearly-full volumes and claims that never bound, which are never earned from approvals, and probe paths, which also need 3 approved here first.

Only an admin can change this card. Each setting, its bound and its default are on what Nightshift may do on its own.

Grants on cluster lists every grant for the selected cluster. A grant lets Seylo make one kind of change on its own: a scope, limits per change, allowed hours, a rate limit and an expiry. Each card shows:

  • the scope: Whole cluster, Namespace name, or namespace/workload;
  • its status: active, suspended or expired;
  • what it allows: each dimension with its limits, for example CPU limit +50% / −10%, and each Nightshift action in plain words, the same label the grant form and Approvals use (hover it for the action’s name), with its ceiling where it has one, for example Add an HPA where load splits across pods ... - at most 8 replicas. A grant that names Add an HPA with no ceiling says that only HPAs at their maximum above bounds it;
  • its hours, its rate limit, its end date, and how many changes it made;
  • Details: what the approver signed, who approved it and when, and its last decisions.

A dimension listed with no limit shows no limit: never acts: a grant without a limit authorises nothing. A grant that has failed or been put back in a row shows 1/3 failed; at 3 it suspends itself.

Show expired includes expired grants. New grant (admin) opens the grant form; every field is described on grants.

Managed savings on this cluster shows the grant behind managed savings: whether it is on, who signed it and when, when it expires, and each lever that is switched on with its limits in the words the admin agreed to.

When it is off, the card says Off: Savings only proposes; every saving waits for a person's approval. The switch itself stays on the Savings tab, where you see what it would save: the card’s button, Let Seylo manage savings or Change on Savings, opens it.

Pause all automation, at the top of the page (releaser), stops every grant on every cluster in your organisation at once: Nightshift’s fixes and managed savings. Give a reason; it is recorded with your name, and shown in a banner to whoever comes next: Paused: nothing changes on its own, on any cluster.

It stops automation, not work. Nightshift and managed savings still propose, and a person can still approve. Resume automation turns it off; each grant still in force is used again from its next decision, with its original limits. A pause for a single cluster shows as Paused for cluster.

Stopping is never harder than starting: pausing needs releaser, while requesting a grant needs admin.

The limits every grant has, whatever it says:

  • At most 50% in one change, inside its hours, rate-limited, and it expires in at most 90 days. An admin requests it; an approver signs it.
  • Every change is checked. It is read back and verified, and put back if it does not hold. Three failures in a row suspend the grant.
  • Other tools come first. A workload Argo CD or Flux owns gets a pull request, never an in-place change; one Helm owns waits for a person.

What Seylo did on its own lists, newest first, every change Seylo made on this cluster without a person, and every time a grant refused one. It covers 30 days, and one change is one line, wherever it was recorded.

Filter it with All, Changes, Undone or failed and Refused; each shows its count.

Each line has:

Part What it says
Outcome Applied, checking, Verified, Undone, Failed, Refused, Pull request, PR rejected, or Restored by a person
Source Nightshift, Managed savings or Grant
When The time of the change or the decision
What The workload, and the change in plain units, such as shop/web · CPU limit 900m → 1.35 cores
Why Why it was done, or why it was refused or undone, in one sentence
Under what The grant, earned autonomy or the managed savings grant that allowed it

Open goes to the full record: the fix, the managed change, or the grant.

A change waiting for approval is not here: it is something Seylo has not done. It is on Approvals.

When the same kind of change keeps waiting for approval on one workload, an admin can let it run on its own next time, from the waiting change:

  1. On Approvals, open the waiting resource or replica change and choose Allow this kind.
  2. The grant form opens pre-filled: the workload’s namespace and name, the one resource the change touches, and the waiting change’s own size as the limit in its direction. The other direction keeps the blank form’s limits, 20% up and 10% down. A box above the form says where each value came from; a request that does not record its value before says so, and the blank form’s limits stay.
  3. Check each value, widen what you mean to, and choose Request approval.

The grant covers that workload only, not its namespace. An approver still signs it, and the change that was waiting stays on Approvals for its own decision.

Role Can
viewer Read the page, the grants and the log
releaser Pause all automation and resume it; suspend and resume a grant
admin Change Fixes Nightshift may apply; request a grant (New grant, Allow this kind)
approver Sign a grant on Approvals. One approval is enough, and an admin may approve their own grant

Until 0.4.9, a separate autoscale engine ran every 15 minutes, worked out its own sizing changes - smaller CPU requests, fewer replicas, lower HPA floors - and filed those no grant covered for approval, again every few days. The same container could be asked to lower its request on one page while Nightshift raised its limit on another.

That engine is retired:

  • Sizing is one judgement, in Savings. Rightsizing on the Savings tab now goes both ways: it lowers a request above use, and raises a CPU or memory limit a container presses on, never both for one resource. You send it for approval, or managed savings applies it inside its grant.
  • Nothing applies a sizing change on its own outside managed savings. A grant that names CPU request, memory request or replicas now covers only the changes Nightshift and Seylo’s remediation make with those dimensions, such as raising a throttled container’s CPU limit.
  • Requests the engine already filed stay on Approvals and are decided as before. Its past decisions stay in the grants’ history.
  • “Grant” is the one word for every standing permission, everywhere in the portal. Some older messages and pages may still say autoscale policy or standing permission; they mean a grant.