Server Monitor (Live Telemetry).

The live server monitor gives a real-time, top-like view of any Cycle host node directly from the portal without SSH or third-party agents required. Open it from the server's dashboard under Infrastructure.

The monitor connects to the server over a live telemetry stream and refreshes automatically. The header shows the server's hostname, when the data was last updated, and a countdown to the next update (roughly every 10 seconds).

The monitor is organized into six tabs:

  • Dashboard
  • CPU
  • Memory
  • Disk
  • Network
  • Processes.

Viewing requires the servers-view capability.

Server Monitor (Live Telemetry) Dashboard

The dashboard is the triage view: every panel summarizes one subsystem, and each corresponds to a tab where you can drill in. The workflow it supports is "is this server healthy, and if not, which tab do I open next?"

Panel

Description

Load Avg

1-minute load average with 5m / 15m values and history sparkline.

CPU

Current CPU utilization, with the number of vCPUs and busy state.

Memory

Percentage of memory used, with used / total GiB.

Disk Busy

Percentage of time the disk was busy, with the peak over the window.

Pressure · Partial Stalls (10s)

CPU, memory, and I/O partial stall percentages with history.

Health Signals

OOM kills, CPU throttling, disk saturation, and direct reclaim — each reports none / no activity when healthy.

Top Processes

The heaviest processes on the host, sortable by CPU%, memory, threads, and disk I/O. Rows expand for more detail.

Server Monitor (Live Telemetry) CPU

The CPU page breaks down every component of current CPU usage and aggregates usage by cgroup.

Panel

Description

CPU Utilization

Total vs. system (kernel) CPU usage over time. Hover for point-in-time values.

CPU Pressure

Partial and full stall percentages across 10s, 60s, and 5m windows.

Load Average

1m / 5m / 15m load with history, plus 1m load expressed as a percentage of available threads — values over 100% indicate more runnable work than the CPU can service.

User vs System Time

The split of active CPU between user space and kernel time.

Scheduling & Throttling

Throttled periods (/s), time throttled (ms/s), involuntary context switches, and run-queue depth.

CPU Per Cgroup

Every workload on the host is broken out by cgroup, mapped to its Cycle container with a link to its environment. The All Cgroups row aggregates the entire host, and platform services (e.g. the compute service) appear alongside user workloads.

Server Monitor (Live Telemetry) Memory

The memory page gives a granular look at memory usage on the system.

Panel

Description

Memory Usage

Host used, workloads (incl. cache), and anon (pinned) memory over time, against total.

Memory Pressure

Partial and full stall percentages across 10s, 60s, and 5m windows.

Memory Composition

Two breakdowns: Host workloads (pinned), system, and available memory; Workloads anon (pinned), file cache, and kernel memory charged to workloads.

Reclaim & Fault Signals

Working-set refaults, major page faults, and direct reclaim, each per second. Non-zero values indicate memory is being reclaimed faster than workloads would like.

Memory Pressure Per Cgroup

Per-instance memory accounting: usage, peak, part/full stall (10s), and an OOM kill count. Each row links to the instance and its environment.

Server Monitor (Live Telemetry) Disk

Live statistics for each physical disk on the host.

Physical Disk Throughput

A combined throughput chart for the host's physical disks. Each device is charted with two series: solid = read, dashed = write. When multiple disks are present, each gets its own color, shown in the legend.

Hover anywhere on the chart to inspect a point in time. The tooltip will show the timestamp along with the read and write rates for every device at that moment. The y-axis scales automatically to the busiest burst in the window, making periodic write spikes easy to spot against otherwise idle baselines.

Sustained throughput near a device's known limits, especially paired with rising I/O pressure or queue depth in the panels, indicates the disk could be a bottleneck rather than any single workload.

I/O Pressure

Part and full stall (10s) per workload, mapped to containers and environments. This can help to pinpoint which container is driving disk contention.

Column

Description

Device

The device name (e.g. vda).

Read / Write

Current read and write throughput.

IOPS

I/O operations per second.

Latency

Average I/O latency.

Queue

Average I/O queue depth.

Utilization

Percentage of time the device was busy servicing requests.

Volumes

The same statistics for every logical volume on the host, plus size (maximum size not the current usage). This includes Cycle's platform volumes (base, logs, agent, and the LVM pool volumes) and per-instance volumes. Use the instance volumes links to directly access the instance they belong to.

Server Monitor (Live Telemetry) Network

The network tab covers both sides of a Cycle host's networking: the physical interfaces the server uses to reach the world, and the overlay networks Cycle builds for environments.

Physical & Gateway Interfaces

Each physical NIC and Cycle-managed bridge (e.g. br-ex, used for gateway traffic) gets its own card showing:

Element

Description

Live rates

Current receive (↓) and transmit (↑) throughput.

Throughput chart

RX and TX history over the monitoring window.

Totals

Cumulative bytes received and transmitted since the counters started.

Drops

Total dropped packets on the interface.

Sustained drops on a physical interface may point to saturation or upstream network issues, while drops on a gateway bridge can indicate misconfigured or rejected egress traffic.

Overlay Networks

Cycle builds an encrypted overlay network per environment, and every instance connects to it through its own virtual interfaces. This table shows traffic through each of them, grouped as a tree:

  • Environment networks (type Environment) — the host-side overlay (e.g. vxl-577) for each environment with instances on this server. The row links to the environment, shows how many instances are attached, and expands to reveal the underlying host interfaces, including the private variant (-priv).
  • Instances (type Instance) — each instance row links to the instance and expands into its individual interfaces: i-pub-… carries the instance's public-path traffic, i-pri-… carries private environment traffic.

Each row reports RX, TX, and drops, where drops include the percentage of that interface's packets dropped, highlighted when significant. This makes it easy to distinguish "the environment network is fine, one instance is misbehaving" from "the whole overlay is dropping traffic." For instance, drops concentrated on an environment's host interfaces (but not on instance interfaces) usually indicate traffic arriving for the environment that no instance is accepting.

Use Flat view to collapse the hierarchy into a single sortable list of all interfaces.

Server Monitor (Live Telemetry) Processes

A live process explorer for the entire host which includes every process in one tree.

Panel

Description

Processes

Total process count, with peak and history.

Threads

Total thread count across all processes, with peak.

Running

Processes currently runnable/on-CPU (the rest are sleeping or blocked), with peak.

Load Avg

Summary load average for 1, 5, 15 minute windows.

Active Processes

The process table mirrors the host's real process hierarchy: Cycle's platform services (e.g. compute-spawnercompute) appear at the top level, and every containerized workload appears under its container-shim parent. If a shim is expanded it will reveal the process tree running inside that instance (e.g. container-shimnode → worker processes).

Column

Description

PID

The host PID of the process.

Command

The binary and its arguments.

CPU% / Memory / Threads / Disk I/O

Live per-process resource usage. All columns are sortable.

As with the other tabs, Flat view replaces the tree with a single sortable list of all processes.

Cookies

Cookies Preferences

We run basic, anonymous analytics by default to measure site traffic. By clicking "Accept," you allow additional cookies for advanced app improvements and tailored advertising. Choose what you share by clicking "Customize."