Skip to main content
The health system works with two layers of status: the result of a single sample (as shown by a particular test) and the summary state (nodes, connection circuits, the entire fleet), the aggregate state is obtained by combining samples according to the principle of “worst-significant”, this page explains what each status means and how the samples form the overall picture.
For whom: Panel Administrator, SRE. Required rights: Administrator session.

Status of individual samples

Each inspection returns one of the statuses below: “not applicable” status means that no verification is required for this purpose.
The statuses not_required, absent, skipped, not_implemented, unsupported are service states and do not increase the severity of nodes.

WARP status

For WARP, a separate set of states is used: installed, missing, port_closed, unreachable, pending_install, not_required, unknown. If WARP is not required by the circuit assembly plan, the status is not_required, not a failure.

Summary states

Sample statuses are reduced to the general state of the node, circuit or entire park:

Severity and incidents

The level of severity of the incident is deduced from the consolidated state:

How the state of the node develops

The panel combines three signals - the state of the node in Remnawave, the results of health checks and the heartbeat of the agent:
  • critical — Remnawave reports that nodes disconnected; or agent offline; or network reachability unreachable/failed.
  • degraded — Achievability timeout/dns_failed/port_closed; or agent degraded; or service failed/timeout/degraded.
  • healthy - achievability and service ok/not_implemented or service ok with an online agent.
  • unknown is nothing definite yet.
The inaccessibility of the SSH port is fixed for diagnosis, but in itself does not increase severity if the input ports and agent are available. The failure of the input ports translates the node and affected circuits into degradation.

Freshness and Flow version

The draft, the version used and the observed health are displayed separately. The checks relate to the working image; the presence of the applied version does not make Flow healthy without sufficient checks. The absence of a proven client pass can leave the state of unknown or degraded even with available nodes. The late result of the old version does not replace the current one. Use the latest information from the current version to restore it. Diagnosis of Flow.

Quorum of input nodes at the connection scheme

For the connection scheme (Flow), the summary takes into account the roles of nodes - input (entry) and transit (transit) - and works by quorum:
Example: in a balanced circuit with three input nodes, the failure of one of them gives degraded, and not critical - the remaining two hold more than half the capacity.
Losing the required minimum number of healthy transit pool participants can elevate Flow to critical, even if input nodes are available.

Summary throughout the park (global dashboard)

The global state on the dashboard becomes critical when the node, circuit or route becomes critical. The offline agent maximizes the state to degraded; the availability of the node is separately checked by network samples. The state of healthy requires at least one healthy entity and no degraded/critical signals; the empty installation gives unknown.

Causes of failure (classification “why”)

In addition to status, the system captures the cause: the unavailability of the gateway, the fall of the input or transit node, the stop of the service part, the missing agent, the configuration drift, the incorrect conditional connection, the absence or closed WARP port, the inaccessibility or refusal of the authorization of the proxy, the inaccessibility of the source of the CDN, the DNS mismatch, the expired TLS, the failure of the external regional sample, the waiting for target selection and “unknown.” Reconstruction.

Verification

Open Health and press Run check. Summary status and causes are displayed on the dashboard cards.

Rollback

Change operations (confirmation of recovery action, node suppression) and their rollbacks are described on the pages of Restoration and Node Suppression.

Typical errors

unknown means “no data yet,” not “bad,” and after the first run of checks or manual run, the state will be clarified.
Partial loss of inputs can be degraded when the quorum is saved, but the loss of the required minimum of pool participants or sustained end-to-end failure can be critical; see the reason for the particular check.
That’s what it’s designed to be: an offline agent doesn’t raise the global state above degraded. The node’s real inaccessibility will be independently demonstrated by its reachability tests.
Linked pages: Levels of checks, Dashboard health, Recovery, Node Life Cycle.