Metrics

Samplers

Instrumentation categories grouped by subsystem. Each can be enabled or disabled independently in the agent config.

memory

CPU

Usage

CPU time by state, softirq breakdown

expand_more
cpu_usagePer-CPU nanoseconds by state: user, system
softirqPer-CPU interrupt count by kind: hi, timer, net_tx, net_rx, block, irq_poll, tasklet, sched, hrtimer, rcu
softirq_timePer-CPU nanoseconds spent in softirq, same kinds
cgroup_cpu_usagePer-cgroup nanoseconds by state: user, system

Frequency

APERF, MPERF, TSC cycle counters

expand_more
cpu_aperfPer-CPU actual performance cycles
cpu_mperfPer-CPU maximum performance cycles
cpu_tscPer-CPU timestamp counter cycles

Performance Counters

Cycles, instructions, branch predictions, cache, TLB

expand_more
cpu_cyclesPer-CPU cycle count
cpu_instructionsPer-CPU retired instructions
cpu_branch_instructionsPer-CPU branch instructions
cpu_branch_missesPer-CPU branch mispredictions
cpu_dtlb_missPer-CPU data TLB misses (with op: load, store on Intel)
cpu_l3_accessPer-CPU L3 cache accesses
cpu_l3_missPer-CPU L3 cache misses
cpu_tlb_flushPer-CPU TLB flush count by reason: task_switch, remote_shootdown, local_shootdown, etc.
cgroup_cpu_cyclesPer-cgroup cycle count
cgroup_cpu_instructionsPer-cgroup retired instructions

Bandwidth & Migrations

CFS throttling, CPU migration events

expand_more
cpu_coresNumber of online logical cores (gauge)
cpu_migrationsPer-CPU migration count with direction: from, to
cgroup_cpu_bandwidth_*Per-cgroup CFS quota, period, throttled time, period counts
cgroup_cpu_migrationsPer-cgroup CPU migration count
schedule

Scheduler

Runqueue

Scheduling latency, running time, off-CPU time, context switches

expand_more
scheduler_runqueue_latencyHistogram of time tasks wait in the runqueue (ns)
scheduler_runningHistogram of time tasks spend running on CPU (ns)
scheduler_offcpuHistogram of time tasks spend off-CPU (ns)
scheduler_context_switchPer-CPU involuntary context switches
scheduler_runqueue_waitPer-CPU total nanoseconds spent waiting
cgroup_scheduler_*Per-cgroup: runqueue_wait, offcpu, context_switch
storage

Block I/O

Latency

I/O latency distributions by operation type

expand_more
blockio_latencyHistogram (ns) with op: read, write, flush, discard

Requests

Operation counts, bytes transferred, size distributions

expand_more
blockio_operationsCounter with op: read, write, flush, discard
blockio_bytesCounter (bytes) with op: read, write, flush, discard
blockio_sizeHistogram (bytes) with op: read, write, flush, discard
lan

Network

Traffic

Aggregate bytes and packets

expand_more
network_bytesCounter with direction: receive, transmit
network_packetsCounter with direction: receive, transmit

Interfaces

Drops, transmit errors, timeouts

expand_more
network_dropDropped packets counter
network_transmit_busyTransmit busy counter
network_transmit_completeCompleted transmissions counter
network_transmit_timeoutTransmit timeout events

Ethtool (ENA)

AWS EC2 Elastic Network Adapter allowance counters

expand_more
network_ena_bandwidth_allowance_exceededWith direction: receive, transmit
network_ena_pps_allowance_exceededPackets-per-second limit exceeded
network_ena_conntrack_allowance_exceededConnection tracking limit exceeded
network_ena_linklocal_allowance_exceededLink-local traffic limit exceeded
cable

TCP

Traffic

Bytes, packets, and segment size distributions

expand_more
tcp_bytesCounter with direction: receive, transmit
tcp_packetsCounter with direction: receive, transmit
tcp_sizeHistogram (bytes) with direction: receive, transmit

Latency

Connection establishment, packet delivery, jitter, RTT

expand_more
tcp_connect_latencyHistogram (ns) — time to establish connection
tcp_packet_latencyHistogram (ns) — receive-to-read latency
tcp_jitterHistogram (ns) — inter-packet jitter
tcp_srttHistogram (ns) — smoothed round-trip time
tcp_retransmitCounter — retransmitted packets
terminal

Syscall

Counts & Latency

Invocation counts and latency distributions by syscall category

expand_more
syscallCounter with op label (see categories below)
syscall_latencyHistogram (ns) with op label (same categories)
cgroup_syscallPer-cgroup counters with same op labels

Syscall categories (op values):

read write poll lock time sleep socket yield filesystem memory process query ipc timer event other
dynamic_form

Memory

Meminfo & VMStat

System memory gauges and NUMA allocation counters

expand_more
memory_totalGauge (bytes)
memory_freeGauge (bytes)
memory_availableGauge (bytes)
memory_buffersGauge (bytes)
memory_cachedGauge (bytes)
memory_numa_hitCounter — allocations on intended node
memory_numa_missCounter — allocations on non-intended node
memory_numa_foreignCounter — allocations intended for this node, placed elsewhere
memory_numa_interleaveCounter — interleave policy allocations
memory_numa_localCounter — allocations on local node
memory_numa_otherCounter — allocations on remote node
developer_board

GPU

NVIDIA

Memory, power, temperature, clocks, utilization (Linux)

expand_more
gpu_memoryPer-GPU gauge (bytes) with state: free, used
gpu_power_usagePer-GPU gauge (milliwatts)
gpu_energy_consumptionPer-GPU counter (millijoules)
gpu_temperaturePer-GPU gauge (Celsius)
gpu_clockPer-GPU gauge (Hz) with clock: compute, graphics, memory, video
gpu_utilizationPer-GPU gauge (percentage)
gpu_memory_utilizationPer-GPU gauge (percentage)
gpu_pcie_bandwidthPer-GPU gauge (bytes/sec) with direction: receive
gpu_pcie_throughputPer-GPU gauge (bytes/sec) with direction: receive, transmit
gpu_sm_utilizationPer-GPU gauge (%) — Hopper+ only
gpu_sm_occupancyPer-GPU gauge (%) — Hopper+ only
gpu_dram_bandwidth_utilizationPer-GPU gauge (%) — Hopper+ only
gpu_tensor_utilizationPer-GPU gauge (%) with pipe: any, hmma (FP16/BF16/TF32), imma (integer), dfma (FP64) — Hopper+ only

Apple Silicon

Power, clocks, utilization (macOS)

expand_more
gpu_power_usagePer-GPU gauge (milliwatts)
gpu_energy_consumptionPer-GPU counter (millijoules)
gpu_clockPer-GPU gauge (Hz) with clock: graphics
gpu_utilizationPer-GPU gauge (percentage)
speed

Rezolus (self-monitoring)

Resource Usage

Rezolus process CPU, memory, I/O, and context switches

expand_more
rezolus_cpu_usageCounter (ns) with state: user, system
rezolus_memory_usage_resident_set_sizeGauge (bytes)
rezolus_memory_page_reclaimsCounter
rezolus_memory_page_faultsCounter
rezolus_blockio_operationsCounter with op: read, write
rezolus_context_switchCounter with kind: voluntary, involuntary

Metric types

Every metric in a Rezolus recording is one of these three shapes. Column metadata records the type, and the viewer dispatches chart rendering accordingly.

counterUInt64, monotonic

A monotonically non-decreasing total. Rates and deltas are derived at query time via rate() / irate(). Examples: cpu_usage, syscall, blockio_operations.

gaugeInt64, instantaneous

An instantaneous value that can go up or down. Used as-is — no rate derivation. Examples: cpu_cores, memory_used, gpu_utilization.

histogramList<UInt64>, H2 bucketed

A full bucketed distribution per snapshot, not just summary percentiles. Queryable via histogram_quantiles(), histogram_mean(), histogram_sum(), etc. Examples: scheduler_runqueue_latency, syscall_latency, tcp_packet_latency.