Agent metrics

Reference for all metrics collected and shipped by the Deckhouse Observability Platform agent (opAgent). Metrics are grouped by topic: base system, disks, network, hardware, processes and containers, virtualization, databases, queues, web servers, JVM, mail, security, custom, and internal metrics.

How to read this reference

Metric names

Inside the agent, metric names use dots (cpu.usage.user). When shipped to the storage, names are converted to a Prometheus-compatible form:

• a dot . is replaced with __ (double underscore): cpu.usage.user → cpu__usage__user;
• a dash - is replaced with ____ (four underscores);
• a name starting with a digit gets the ___ prefix: 1min → ___1min.

All names in this document are given in the form they are stored and queried (that is, with __).

Common labels

Every agent metric automatically gets the following labels:

Label	Value
job, okagent_plugin	name of the plugin that collected the metric (cpu, postgresql, nginx, …)
conf, okagent_plugin_instance	plugin instance: config path, service address, or container
instance	hostname of the server where the agent runs

Metrics from Kubernetes containers additionally get k8s_namespace, k8s_pod, k8s_container, k8s_namespace_pod. Custom labels from the labels: section of a plugin config are added to all its metrics.

The tables below list only metric-specific labels — the common ones are not repeated.

Metric types

Type	Description
gauge	instantaneous value (memory size, connection count)
counter	monotonically increasing counter (bytes since start); use rate() to get a per-second rate
rate	per-second rate already computed by the agent over the scrape interval (~60 seconds); no rate() needed
text	text value (version, MAC address). Stored as a series with the value=<text> label and metric_type="text"
status	status metric: value 1 and value="empty" — everything is fine; value 0 and value=<error text> — a problem

Integration status

All auto-discovered integrations (PostgreSQL, MySQL, Redis, …) publish a status or plugin__status metric: empty text — collection works, error text — what prevents collection (missing permissions, unreachable port, etc.).

Base system metrics

CPU (plugin cpu)

Collected automatically on every server.

Metric	Labels	Type	Description
cpu__logical_cores	—	gauge	Number of logical cores
cpu__physical_sockets	—	gauge	Number of physical CPU sockets
cpu__physical_cores	—	gauge	Total number of physical cores
cpu__usage__user	—	rate	Share of CPU time in user mode, %
cpu__usage__system	—	rate	Share of CPU time in system mode, %
cpu__usage__nice	—	rate	Share of CPU time at modified priority (nice), %
cpu__usage__iowait	—	rate	Share of time waiting for disk I/O, %
cpu__usage__irq	—	rate	Share of time handling hardware interrupts, %
cpu__usage__softirq	—	rate	Share of time handling software interrupts, %
cpu__usage__steal	—	rate	Share of time “stolen” by the hypervisor, %
user, system, nice, idle, iowait, irq, softirq, steal	—	counter	Accumulated CPU time per mode, seconds (legacy names)
cpu__frequency__mhz	core	gauge	Current core frequency, MHz
cpu__governor	core, governor	gauge	Flag (=1) for the active scaling governor of a core (Linux only)
cpu_flags	model_name, flag	gauge	Flag (=1) for a CPU feature flag (avx, sse4, etc.)
cpu__thermal_throttle__core_throttle_count	cpu	counter	Core thermal throttling events (Linux only)
cpu__thermal_throttle__package_throttle_count	cpu	counter	CPU package thermal throttling events (Linux only)

Memory and swap (plugin memory)

Metric	Labels	Type	Description
memory__total	—	gauge	RAM size, bytes
memory__available	—	gauge	Available memory (including reclaimable caches), bytes
memory__usage__used	—	gauge	Used memory (total − free − buffers − cached), bytes
memory__usage__free	—	gauge	Free memory, bytes
memory__usage__buffers	—	gauge	Memory in buffers, bytes
memory__usage__cached	—	gauge	Memory in page cache, bytes
total, available, free, buffers, cached	—	gauge	Same values (legacy names)
memory__errors	type = correctable | uncorrectable	counter	ECC memory errors from EDAC (Linux only, when /sys/devices/system/edac exists)
memory__numa__allocated	node, allocation = local_node | other_node | interleaved	counter	Page allocations per NUMA node (only with ≥2 nodes)
swap__total	—	gauge	Swap size, bytes
swap__space__free	—	gauge	Free swap, bytes
swap__space__used	—	gauge	Used swap, bytes
swap__io__in	—	rate	Swap-in rate, pages/s
swap__io__out	—	rate	Swap-out rate, pages/s
swap_total, swap_free	—	gauge	Swap size and free swap (legacy names)
swap_in, swap_out	—	counter	Accumulated swap page counters (legacy names)

Load average (plugin load_average)

Metric	Labels	Type	Description
load_average__1min	—	gauge	1-minute load average
load_average__5min	—	gauge	5-minute load average
load_average__15min	—	gauge	15-minute load average
___1min, ___5min, ___15min	—	gauge	Same values (legacy names)

OS information (plugin uname)

Metric	Labels	Type	Description
uname	os, machine, distrib, version, kernel, gather_tool	gauge	Flag (=1) with OS metadata: distribution, version, kernel, architecture
host__uptime	—	gauge	Server uptime, seconds
os, machine, distrib, version, kernel	—	text	Separate text metrics with the same values
uname_gather_error	—	gauge	=1 if the distribution could not be determined

Time and NTP (plugin time)

Metric	Labels	Type	Description
time__ntp__status	—	status	NTP poll status: empty — OK, text — error (servers not found, timeout)
time__current_offset, time__offset__ntp	—	gauge	Local clock offset relative to NTP, seconds (can be negative)

NTP servers are taken from /etc/ntp.conf; pool.ntp.org is used as fallback.

Host identity (plugins metadata, machine_identity)

Metric	Labels	Type	Description
machine_identity	machine_id, source = dmi_uuid | dmi_serial | machine_id	gauge	Flag (=1) with a stable machine identifier (DMI UUID, serial number, or /etc/machine-id)
local_metadata__host__hostname_status	—	status	Hostname change tracking: text when it changed
local_metadata__host__group	—	text	Host group from the local agent config (if set)
local_metadata__host__description	—	text	Host description from the local config (if set)

Disks and storage

File systems (plugin disk)

Per mounted partition; label mount_point.

Metric	Labels	Type	Description
bytes_total	mount_point	gauge	Total file system size, bytes
bytes_used, disk__partition__bytes_used	mount_point	gauge	Used, bytes
disk__partition__bytes_available, disk__partition__space__available	mount_point	gauge	Available to the user, bytes
disk__partition__bytes_reserved, disk__partition__space__reserved	mount_point	gauge	Reserved for root, bytes
disk__partition__space__used	mount_point	gauge	Used, bytes
inodes_total	mount_point	gauge	Total inodes
inodes_used, disk__partition__inodes__used	mount_point	gauge	Used inodes
disk__partition__inodes__free	mount_point	gauge	Free inodes
disk__partition__read_write	mount_point	gauge	1 — mounted rw, 0 — ro
fs, disk__partition__fs	mount_point	text	File system type (ext4, xfs, …)
device, disk__partition__device	mount_point	text	Partition block device
mount_options, disk__partition__mount_options	mount_point	text	Mount options
disk__fs__errors	device, fs="ext4"	counter	ext4 error counter from /sys/fs/ext4/*/errors_count (Linux only)

Disk I/O (plugin disk)

Rates are computed by the agent over the scrape interval.

Metric	Labels	Type	Description
disk__device__ops__read	device, pci_bdf	rate	Read operations per second on a block device
disk__device__ops__write	device, pci_bdf	rate	Write operations per second
disk__device__bytes__read	device, pci_bdf	rate	Bytes read per second
disk__device__bytes__write	device, pci_bdf	rate	Bytes written per second
disk__device__io_time__read	device, pci_bdf	rate	Read time, ms per second
disk__device__io_time__write	device, pci_bdf	rate	Write time, ms per second
disk__device__io_time__total	device, pci_bdf	rate	Total I/O time, ms per second (utilization)
disk__partition__ops__read / __write	mount_point	rate	Read/write operations per second per partition (incl. ZFS datasets)
disk__partition__bytes__read / __write	mount_point	rate	Read/write bytes per second per partition

SMART (plugin disk)

Disk health diagnostics. ATA attributes are read directly (SG_IO), NVMe via ioctl, SCSI/SAS via smartctl. All metrics carry the device (e.g. /dev/sda) and pci_bdf labels.

Metric	Type	Description
disk_smart_has_smartctl	gauge	1 — smartctl is installed on the host
disk_smart_device_model	text	Disk model (ATA IDENTIFY)

ATA attributes — family disk_smart_<attr>_value (normalized value 0–255), disk_smart_<attr>_raw (raw value where a parser is defined), disk_smart_<attr>_threshold (failure threshold). Supported <attr>:

<attr> (SMART ID)	Description
raw_read_error_rate (1)	Surface read errors
reallocated_sectors_ct (5)	Reallocated sectors (>0 — degradation)
power_on_hours (9)	Power-on hours
spin_retry_count (10)	Spindle spin-up retries
power_cycle_count (12)	Power cycles
wear_leveling_count (177)	Samsung SSD wear (value = remaining life, %)
program_fail_cnt_total (181) / erase_fail_cnt_total (182)	NAND program/erase failures
end_to_end_error (184)	End-to-end integrity errors
reported_uncorrect (187)	Uncorrectable errors
command_timeout (188)	Command timeouts
temperature_celsius (194)	Temperature, °C (raw)
hardware_ecc_recovered (195)	Corrected ECC errors
reallocated_event_count (196)	Reallocation events
offline_uncorrectable (198)	Uncorrectable sectors from offline scan
udma_crc_error_count (199)	Interface CRC errors (cabling)
percent_lifetime_remain (202) / ssd_life_left (231) / media_wearout_indicator (233)	SSD remaining life, %
total_lbas_written (241) / total_lbas_read (242)	Total sectors written/read

NVMe — all gauge:

Metric	Description
disk_smart_nvme_available_spare	Remaining spare blocks, %
disk_smart_nvme_available_spare_threshold	Spare warning threshold, %
disk_smart_nvme_percentage_used	Endurance used, % (0–255, >100 — beyond rated)
disk_smart_nvme_temperature_celsius	Controller temperature, °C
disk_smart_nvme_media_errors	Uncorrectable media errors
disk_smart_nvme_power_on_hours	Power-on hours
disk_smart_nvme_unsafe_shutdowns	Unsafe power-off events
disk_smart_nvme_power_cycles	Power cycles
disk_smart_nvme_data_units_read / _written	Read/written (1 unit = 500 KiB)
disk_smart_nvme_host_read_commands / _write_commands	Host read/write commands
disk_smart_nvme_critical_warning	Raw critical warning byte (0 — OK)
disk_smart_nvme_critical_warning_spare_below	1 — spare exhausted
disk_smart_nvme_critical_warning_temp_exceeded	1 — overheating
disk_smart_nvme_critical_warning_reliability_degraded	1 — reliability degraded
disk_smart_nvme_critical_warning_read_only	1 — disk went read-only
disk_smart_nvme_critical_warning_backup_failed	1 — backup power failure

SCSI/SAS (via smartctl) — all gauge:

Metric	Description
disk_smart_scsi_temperature_celsius	Temperature, °C
disk_smart_scsi_grown_defects	Grown defect list
disk_smart_scsi_non_medium_errors	Non-medium errors
disk_smart_scsi_read_uncorrected / _write_uncorrected / _verify_uncorrected	Uncorrected read/write/verify errors
disk_smart_scsi_info_exceptions	1 — SMART status “failed”

RAID (plugin raid)

Hardware controllers (MegaRAID/storcli, Adaptec/arcconf, HPE/ssacli) and Linux software md. Linux only. controller_id: megaraid:0, hpe:0, adaptec:1, soft-raid. The state_class label is a normalized state: healthy, rebuild, predictive, failed, offline, unknown (for arrays: optimal, degraded, offline, rebuilding, unknown).

Metric	Labels	Type	Description
raid_controller_info	controller_id, model, serial, firmware, driver	gauge	=1; controller inventory record
raid_controller_state	controller_id, state_class, raw_state	gauge	=1; controller state
raid_controller_temperature_celsius	controller_id	gauge	Controller temperature, °C
raid_controller_bbu	controller_id, state_class, raw_state	gauge	=1; battery (BBU) state
raid_controller_bbu_health_percent	controller_id	gauge	Battery health, 0–100 %
raid_controller_cachevault	controller_id, state_class, raw_state	gauge	=1; CacheVault state (MegaRAID)
raid_array_info	controller_id, array, raid_level, array_type	gauge	=1; array (VD/LD) inventory record
raid_array_state	controller_id, array, state_class, raw_state	gauge	=1; array state (incl. md arrays)
raid_pd_state	controller_id, slot_id, model, serial, wwn, media_type, state_class, raw_state	gauge	=1; state of a physical drive behind the controller
raid_md_degraded	array, controller_id="soft-raid"	gauge	1 — md array degraded
raid_md_rebuild_pct	array, controller_id="soft-raid"	gauge	md rebuild progress, %
plugin__status	controller	status	Empty text — collection OK; text — controller detected via PCI driver but the CLI tool is missing or parsing failed

Network

Network interfaces (plugin net)

Per interface except veth*; label interface.

Metric	Labels	Type	Description
net__interface__bytes__in / __out	interface	rate	Bytes/s received/sent
net__interface__packets__in / __out	interface	rate	Packets/s
net__interface__errors__in / __out	interface	rate	Errors/s
net__interface__drops__in / __out	interface	rate	Dropped packets/s
in_bytes, in_packets, in_errors, in_dropped, out_bytes, out_packets, out_errors, out_dropped	interface	counter	Cumulative counters from /proc/net/dev (legacy names)
net__interface__status, status	interface	gauge	1 — interface is UP, 0 — DOWN
mtu	interface	gauge	Interface MTU
hw_address	interface	text	MAC address
net__ip__status	interface, ip, version = v4 | v6	gauge	=1; flag for an IP address present on the interface
ipv4_addresses, ipv6_addresses	interface	text	JSON lists of interface addresses
net__conntrack__count	—	gauge	Current number of conntrack entries (Linux only)
net__conntrack__max	—	gauge	conntrack table limit

Sockets and TCP (plugin netstat)

Linux only; collected in the host network namespace.

Metric	Labels	Type	Description
netstat__protocol__packets__in / __out	protocol = tcp | udp | icmp | ip	counter	Received/sent packets (segments, datagrams) per protocol
netstat__protocol__packets__errors	protocol, error	counter	Protocol errors: UDP — NoPorts, RcvbufErrors, SndbufErrors, InErrors; TCP — OutRsts, InErrs; IP — InHdrErrors, InAddrErrors, InUnknownProtos, InDiscards, OutNoRoutes, ReasmFails, FragFails
netstat__protocol__packets__retransmits	protocol="tcp"	counter	TCP retransmits
netstat__listen__status	listen_ip, listen_port	gauge	=1 per listening port (with >200 ports they are grouped into ranges)
netstat__listen__current_ack_backlog	listen_ip, listen_port	gauge	Current accept queue backlog
netstat__listen__max_ack_backlog	listen_ip, listen_port	gauge	Maximum backlog
netstat__connections__inbound__count	listen_ip, listen_port, remote_ip, state	gauge	Inbound TCP connections per state (ESTABLISHED, TIME_WAIT, SYN_RECV, …)
netstat__connections__outbound__count	remote_ip, state	gauge	Outbound connections per state
netstat__connections__inbound__with_read_queue / outbound…	same as count	gauge	Connections with a non-empty read queue
netstat__connections__inbound__with_write_queue / outbound…	same as count	gauge	Connections with a non-empty write queue
netstat__connections__inbound__rtt / outbound…	same as count + percentile="95"	gauge	95th percentile of connection RTT, seconds
okagent__netstat__listen__status	—	status	Text too many listen ports when the port list was truncated

Non-local remote addresses are aggregated into remote_ip="~nonlocal".

SNMP devices (plugin netdev, configured)

Polls network equipment via SNMP v2c (IF-MIB). Configured by a YAML file with host and community.

Metric	Labels	Type	Description
in_bytes / out_bytes	interface, description	counter	Bytes received/sent on a device port
in_packets / out_packets	interface, description	counter	Packets
speed	interface, description	gauge	Port speed, bytes/s

Hardware and inventory

Hardware inventory (plugin cmdb)

Full hardware inventory of bare-metal servers: SMBIOS, sysfs, RAID CLI tools. Disabled on virtual machines. All *_info metrics are gauges with value 1; identity is carried by labels (component_key, serial, model, manufacturer, etc.). Rescan every 5 minutes.

Metric	Type	Description
cmdb_system_info	gauge	System (SMBIOS Type 1): manufacturer, model, serial number
cmdb_baseboard_info	gauge	Motherboard (Type 2)
cmdb_bios_info	gauge	BIOS (Type 0)
cmdb_chassis_info	gauge	Chassis (Type 3)
cmdb_cpu_socket_info	gauge	One series per CPU socket
cmdb_cpu_sockets_total / _populated	gauge	Total/populated CPU sockets
cmdb_memory_slot_info	gauge	One series per DIMM slot
cmdb_memory_slots_total / _populated	gauge	Total/populated memory slots
cmdb_memory_total_bytes	gauge	Installed memory, bytes
cmdb_memory_max_capacity_bytes	gauge	Maximum capacity per SMBIOS
cmdb_storage_block_device_info	gauge	One series per block device (/sys/class/block: disks, partitions, LVM, md); label type = physical_disk | raid_volume | partition | lvm | mdraid | unknown
cmdb_storage_block_device_capacity_bytes	gauge	Block device capacity
cmdb_storage_block_device_slave_info	gauge	Storage stack graph edge: device → slave (LV → md → partition → disk)
cmdb_storage_physical_disk_info	gauge	Physical disk (incl. hidden behind RAID); labels attachment = direct | raid | hba, interface, state_class
cmdb_storage_physical_disk_capacity_bytes	gauge	Physical disk capacity
cmdb_storage_controller_info	gauge	RAID/HBA controller
cmdb_storage_raid_volume_info	gauge	RAID volume (controller virtual drive); volume_wwn, os_device_name link it to the OS
cmdb_storage_raid_volume_capacity_bytes	gauge	RAID volume capacity
cmdb_disks_total / cmdb_disks_total_bytes	gauge	Number of disks and their total capacity
cmdb_nic_info / cmdb_nics_total	gauge	Physical network cards
cmdb_gpu_info / cmdb_gpus_total	gauge	Discrete GPUs
cmdb_system_slot_info, cmdb_system_slots_total / _populated	gauge	PCIe expansion slots (SMBIOS Type 9)
cmdb_sata_port_info, cmdb_sata_ports_total / _occupied	gauge	SATA ports
cmdb_psu_info	gauge	Power supply unit (SMBIOS Type 39); labels status = ok | non_critical | critical | non_recoverable | unknown, present, unplugged
cmdb_psu_total / _present / _healthy	gauge	Total/present/healthy PSUs
cmdb_inventory_scan_success	gauge	1 — scan collected data
cmdb_inventory_partial	gauge	1 — some collectors failed
cmdb_inventory_scan_timestamp	gauge	Unix time of the last rescan
cmdb_inventory_generation	gauge	Monotonic counter of hardware composition changes
cmdb_inventory_fingerprint_info	gauge	=1; hash of the current inventory (label fingerprint)
cmdb_inventory_source_available	gauge	Per source (source = smbios | sysfs): 1 — source available

Sensors (plugin sensors)

lm-sensors, Linux only. Labels: device (chip), sensor.

Metric	Type	Description
sensor__temperature__current	gauge	Current sensor temperature, °C
sensor__temperature__high / __critical	gauge	High/critical thresholds, °C (when defined)
sensor__fan__rpm__current	gauge	Fan speed, RPM
sensor__fan__rpm__minimum	gauge	Minimum speed (when defined)

Cloud provider and packages (plugins provider_info, package_info)

Metric	Labels	Type	Description
provider	provider_name, type, virtualization_type	gauge	=1; detected cloud provider (amazon, azure, google, yandex, hetzner, selectel, …)
package_version	package, version_full, version_major, version_minor, version_patch, version_additional	gauge	=1; installed package version (currently tracks sudo); sources — dpkg/dnf/yum

Processes, containers, cgroups

Processes (plugin process_info)

Metrics are aggregated by groups (process, username, container); container = ~host for processes outside containers.

Metric	Extra labels	Type	Description
process__cpu__user / __system	—	rate	CPU consumption of a process group, cores (CPU-seconds per second)
process__max_cpu_percent_per_thread	—	gauge	Maximum load of a single thread in the group, %
process__mem__rss	—	gauge	Total RSS, KB
process__mem__swap	—	gauge	Total swap, KB
process__proc_count	state	gauge	Process count per state (running, sleeping, zombie, …)
process__thread_count	—	gauge	Total thread count
process__disk__ops__read / __write	—	rate	Disk operations of the group, ops/s
process__disk__bytes__read / __write	—	rate	Disk traffic of the group, bytes/s
process__open_files__count	—	gauge	Open file descriptors
process__open_files__max_usage_percent	—	gauge	Maximum fd limit usage in the group, %
process__uptime	—	gauge	Maximum process uptime in the group, seconds

Cgroups (plugin cgroups)

Linux only. Common labels: cgroup_type = systemd | docker | lxc | k8s, cgroup, container; for k8s also k8s_qos and pod labels.

Metric	Extra labels	Type	Description
cgroup__cpu__usage	space = user | system	counter	CPU time consumed by the cgroup, seconds
cgroup__cpu__shares	—	gauge	CPU shares
cgroup__cpu__quota__period	—	gauge	CFS quota period, seconds
cgroup__cpu__quota__normalized	—	gauge	Quota in core fractions (quota/period)
cgroup__cpu__throttling__nr_periods	—	counter	Total CFS periods
cgroup__cpu__throttling__nr_throttled	—	counter	Throttled periods
cgroup__cpu__throttling__throttled_time	—	counter	Throttled time, seconds
cgroup__memory__usage__rss / __cache	—	gauge	Cgroup RSS and page cache, bytes
cgroup__memory__limit	—	gauge	Memory limit, bytes (when set)
cgroup__memory__dirty / __writeback	—	gauge	Dirty/writeback pages, bytes
cgroup__memory__pages__in / __out	—	counter	Pages paged in/out
cgroup__memory__pages__major_faults	—	counter	Major page faults
cgroup__blkio__bytes	operation = read | write, device	counter	Cgroup disk traffic, bytes
cgroup__blkio__ops	operation, device	counter	Disk operations
cgroup__blkio__limit__bps / __ops	operation, device	gauge	Bandwidth/IOPS limits (when set)

Docker volumes (plugin docker)

Metric	Labels	Type	Description
docker__container__disk__space__used	container, mount_point, device	gauge	Used space on a container rw volume, bytes
docker__container__disk__space__available	container, mount_point, device	gauge	Free space on the volume, bytes

Virtualization

Virtual machines (plugins qemu, vmware)

QEMU/KVM is auto-discovered from processes; VMware is configured with a vCenter address. Common VM metric labels: vm_id, vm_name, hypervisor_type, hypervisor_host; for KubeVirt — namespace, pod.

Metric	Extra labels	Type	Description
vm_info	state, guest_os	gauge	=1; VM identity
vm_nic_info	mac, model, netdev	gauge	=1; one series per VM network interface
vm_disk_size_bytes	drive, file, format, bus, device, device_name	gauge	Virtual disk size, bytes
vm_allocated_vcpu	—	gauge	Allocated vCPUs
vm_allocated_mem_bytes	—	gauge	Allocated memory, bytes
vm_cpu_seconds_total	mode = user | system	counter	QEMU process CPU time, seconds
vm_rss_bytes	—	gauge	QEMU process RSS, bytes
vm_net_rx_bytes_total / vm_net_tx_bytes_total	—	counter	VM network traffic, bytes
vm_blk_read_bytes_total / vm_blk_write_bytes_total	device	counter	VM disk traffic, bytes
qemu_host_kvm_module_loaded	—	gauge	1 — KVM module loaded (host metric)
qemu_host_libvirt_socket_present	—	gauge	1 — libvirt socket exists
qemu_host_vm_count	—	gauge	Number of QEMU processes on the host

VMware (vCenter realtime counters):

Metric	Type	Description
vm_vmware_cpu_usage_percent	gauge	VM CPU usage, 0–100 %
vm_vmware_memory_active_bytes	gauge	Active guest memory, bytes
vm_vmware_memory_consumed_bytes	gauge	Host memory consumed by the VM
vm_vmware_net_rx_rate_bps / _tx_rate_bps	gauge	Network traffic, bytes/s
vm_vmware_disk_read_rate_bps / _write_rate_bps	gauge	Disk traffic, bytes/s
vm_vmware_esxi_info	gauge	=1; ESXi host (labels host_name, vendor, model, cpu_model, connection_state, maintenance)
vm_vmware_esxi_cpu_cores	gauge	Physical host cores
vm_vmware_esxi_cpu_usage_percent	gauge	Host CPU usage, %
vm_vmware_esxi_memory_total_bytes / _usage_bytes	gauge	Host memory: total/used

Databases and caches

All plugins in this section are auto-discovered: the agent finds the database process and connects on its own. On access problems it publishes plugin__status with instructions.

PostgreSQL (plugin postgresql)

Metric	Labels	Type	Description
postgresql__db__size	database	gauge	Database size, bytes
postgresql__db__commits / __rollbacks	database	counter	Transaction commits and rollbacks
postgresql__db__deadlocks	database	counter	Deadlocks
postgresql__db__tmp_bytes	database	counter	Written to temp files, bytes
postgresql__db__max_transaction_age	database	gauge	Age of the oldest transaction (XID), primary only
postgresql__db__transactions_left_before_shutdown	database	gauge	XIDs remaining before wraparound shutdown
postgresql__locks__count	database, mode	gauge	Locks per mode
postgresql__connections__count	user, database, state, client_addr	gauge	Sessions per state (active, idle, idle in transaction, waiting, …)
postgresql__connections__max_transaction_age	user, database	gauge	Maximum open transaction age, seconds
postgresql__autovacuum_workers__common / __wraparound	—	gauge	Autovacuum workers (regular and anti-wraparound)
postgresql__bgwriter__buffers_clean / __buffers_backend / __buffers_alloc / __maxwritten_clean / __buffers_backend_fsync	—	counter	bgwriter statistics (PostgreSQL 17 — from pg_stat_io)
postgresql__bgwriter__checkpoints_timed / __checkpoints_req	—	counter	Scheduled/requested checkpoints
postgresql__bgwriter__checkpoints_write_time / __checkpoints_sync_time	—	counter	Checkpoint write/sync time, seconds
postgresql__bgwriter__buffers_checkpoint	—	counter	Buffers written by checkpoints
postgresql__checkpointer__timed / __req / __write_time / __sync_time / __buffers_written	—	counter	Same values under the new names (PostgreSQL 17 pg_stat_checkpointer)
postgresql__replication__role	—	text	master or slave
postgresql__replication__wal__current_xlog_location, postgresql__replication__wal__location__current_lsn	—	gauge	Current WAL position (primary)
postgresql__replication__wal__last_xlog_receive_location, …__location__receive_lsn	—	gauge	Received WAL position (standby)
postgresql__replication__wal__last_xlog_replay_location, …__location__replay_lsn	—	gauge	Replayed WAL position (standby)
postgresql__replication__lag__seconds	—	gauge	Replica time lag, seconds
postgresql__replication__wal__replay__paused	—	gauge	1 — WAL replay paused
postgresql__replication__lag__bytes	slave_ip, application_name	gauge	Replica lag in WAL bytes (from the primary side)
postgresql__replication__session_age	standby_ip, application_name	gauge	Replication session age, seconds
postgresql__replication__phases__lag__bytes	standby_ip, application_name, phase	gauge	Lag per phase: sending_to_standby, writing_standby_wal, flushing_standby_wal, replaying_standby_wal
postgresql__replication__phases__lag__seconds	same	gauge	Lag per phase in time (PostgreSQL ≥ 10)
postgresql__replication__slots__retained_wal__bytes	slot_name, slot_type, database	gauge	WAL retained by a replication slot
postgresql__wal_archiver__success__count / __fail__count	—	counter	Successful/failed WAL archivals
postgresql__wal_archiver__last_success__age	—	gauge	Seconds since the last successful archival
postgresql__wal_archiver__unarchived_files__count / __bytes	—	gauge	Unarchived WAL segments: count and size
postgresql__wal__size__bytes / postgresql__wal__files__count	—	gauge	WAL size and file count on disk

Query statistics (pg_stat_statements, top-50 queries + the ~other aggregate; labels database, user, query) — all rate:

postgresql__query__calls, postgresql__query__rows, postgresql__query__time__total, postgresql__query__time__cpu, postgresql__query__time__disk_read, postgresql__query__time__disk_write, postgresql__query__blocks__hit, postgresql__query__blocks__read, postgresql__query__blocks__written, postgresql__query__blocks__dirtied, postgresql__query__temp_blocks__read, postgresql__query__temp_blocks__written.

Table statistics (top per database; labels database, schema, table) — counters are published with the __rate suffix (rate): postgresql__table__seq_scan__rate, …__idx_scan__rate, …__seq_tuple_read__rate, …__idx_tuple_fetch__rate, …__inserted__rate, …__updated__rate, …__deleted__rate, …__autovacuum_count__rate, …__heap_blks_read__rate, …__heap_blks_hit__rate, …__idx_blks_read__rate, …__idx_blks_hit__rate, …__toast_blks_read__rate, …__toast_blks_hit__rate, …__tidx_blks_read__rate, …__tidx_blks_hit__rate; sizes — gauge: postgresql__table__table_size, postgresql__table__indexes_size.

Settings — the postgresql__settings__<parameter name> family: numeric parameters (e.g. postgresql__settings__max_connections, …__shared_buffers, …__autovacuum_max_workers) — gauge, string ones (e.g. …__wal_level, …__archive_mode, …__data_directory) — text. All recognized pg_settings parameters (~300) are published.

Status metrics: plugin__status (connection), plugin__status__pg_database, …__pg_settings, …__pg_stat_activity, …__pg_stat_replication, …__pg_stat_statements — text describes missing permissions or a missing extension.

PgBouncer (plugin pgbouncer)

Metric	Labels	Type	Description
pgbouncer__total_requests, pgbouncer__total_query_count	database	counter	Processed queries
pgbouncer__total_received / __total_sent	database	counter	Traffic, bytes
pgbouncer__total_query_time / __total_xact_time / __total_wait_time	database	counter	Total query/transaction/wait time, seconds
pgbouncer__total_xact_count	database	counter	Transactions
pgbouncer__databases__pool_size / __reserve_pool / __max_connections	database, upstream_addr	gauge	Pool settings
pgbouncer__databases__paused / __disabled	database, upstream_addr	gauge	0/1 — database paused/disabled
pgbouncer__databases__upstream_addr	database	text	Upstream server address
pgbouncer__pools__limits__pool_size / __reserve_pool	database, user, upstream_addr	gauge	Active pool limits
pgbouncer__server_connections__max / __avg	database, user, state, upstream_addr	gauge	Server connections per state (active, idle, login, tested, used)
pgbouncer__server_connection_maxwait__max	database, user, upstream_addr	gauge	Maximum wait for a server connection, seconds
pgbouncer__clients__count	database, user, state, client_address	gauge	Client connections per state
pgbouncer__clients__wait__max / __sum	database, user, client_address	gauge	Client wait, seconds
pgbouncer__config__<parameter>	—	gauge	Numeric parameters from SHOW CONFIG (e.g. pgbouncer__config__max_client_conn)
plugin__status	—	status	Connection status

MySQL / MariaDB (plugin mysql)

Values of SHOW GLOBAL STATUS / SHOW VARIABLES (counter when the source variable is cumulative):

Family	Metrics
Connections	mysql__connection__max_connections, mysql__connections__aborted_clients, …__aborted_connects, …__connections, mysql__connection__connections_by_state (gauge, label state)
Traffic	mysql__traffic__bytes_received, …__bytes_sent
Commands	mysql__commands__{select,insert,update,delete,replace,insert_select,replace_select,multi_update,multi_delete,call_procedure}
Prepared statements	mysql__prepared_commands__{prepare,execute,dealloc}
Transactions	mysql__transaction_commands__{begin,commit,rollback}
Handler API	mysql__handler__{commit,delete,read_first,read_key,read_next,read_prev,read_rnd,read_rnd_next,rollback,savepoint,savepoint_rollback,update,write,prepare}
InnoDB buffer pool	mysql__innodb__buffer_pool__usage__{data,dirty,free,misc}, mysql__innodb__buffer_pool__operations__{flushed,read_ahead,read_ahead_evicted,read_ahead_rnd,read_ahead_seq,read_requests,reads_from_disk,wait_free,write_requests}, mysql__innodb__buffer_pool__activity__{pages_created,pages_read,pages_written}
InnoDB I/O	mysql__innodb__io__{read_bytes,written_bytes,reads_count,writes_count}, mysql__innodb__{fsyncs,current_pending_fsyncs,current_pending_reads,current_pending_writes}, mysql__innodb__doublewrite__{pages_written,writes_count}
InnoDB log	mysql__innodb__log__{waits,write_requests,physical_writes,fsyncs,pending_fsyncs,pending_writes,written_bytes}
InnoDB rows/locks	mysql__innodb__rows__{inserted,updated,deletes,read}, mysql__innodb__row_lock__{current_waits,total_time,waits}, mysql__innodb__page_size
MyISAM	mysql__myisam__key_cache__{blocks_not_flushed,blocks_unused,blocks_used,reads_requests,reads_from_disk,write_requests,writes_to_disk}
Query cache	mysql__query_cache__{free_blocks,total_blocks,hits,inserts,lowmem_prunes,not_cached,queries_in_cache}
Temp/sort/select	mysql__tmp__created_{disk_tables,files,tables}, mysql__sort__{merge_passes,range,sorted_rows,scan}, mysql__select__{full_join,full_range_join,range,range_check,scan}
Delayed	mysql__delayed__{errors,insert_threads,rows_written,not_flushed_rows}
Other	mysql__queries_count, mysql__questions_count, mysql__slow_queries, mysql__current_open_files, mysql__current_open_tables, mysql__opened_tables, mysql__locks__{immediate,waited}, mysql__threads__{cached,connected,created,running}, mysql__uptime, mysql__version (text)
Galera/wsrep	mysql__wsrep__local_recv_queue__avg, …__local_send_queue__avg, …__flow_control__{received,sent}, …__open_connections, …__open_transactions, …__writesets__received__{count,bytes}, …__writesets__sent__{count,bytes}

Other metrics:

Metric	Labels	Type	Description
mysql__connections__max_transaction_age	user, database	gauge	Maximum open InnoDB transaction age, seconds
mysql_gather_error	source, reason, mysql_error_code	gauge	=1 when collecting a part of the data failed
mysql__replication__slave_io_running / __slave_sql_running	master_host, channel_name	gauge	0/1 — replication thread state
mysql__replication__seconds_behind_master	same	gauge	Replica lag, seconds
plugin__status	—	status	Connection status

Query statistics (performance schema, top-50; labels database, query) — all rate: mysql__query__calls__rate, mysql__query__total_time__rate, mysql__query__time__lock_wait__rate, mysql__query__rows__sent__rate, mysql__query__rows__examined__rate, mysql__query__rows__sorted__rate, mysql__query__tmp_tables_created__rate (+ label table_type = disk | in-memory), mysql__query__full_join__rate, mysql__query__range_check__rate, mysql__query__full_scan__rate, mysql__query__sort__merge__rate.

MongoDB (plugin mongo)

Metrics are collected by passive traffic analysis (sniffer); top-50 queries. Common labels: query, query_type, db, collection.

Metric	Type	Description
mongo__query__calls__rate	rate	Queries per second
mongo__query__total_time__rate	rate	Total query time, seconds per second
mongo__query__traffic__in / __out	rate	Request/response traffic, bytes/s
mongo__query__times__mean / __min / __max	gauge	Mean/min/max latency, seconds
mongo__query__times__percentile	gauge	Latency percentiles (label percentile = 50 | 95 | 99)

Redis (plugin redis)

Metric	Labels	Type	Description
redis__version	—	text	Redis version
redis__uptime	—	gauge	Uptime, seconds
redis__memory__used / __rss / __lua	—	gauge	Memory: allocated/RSS/Lua, bytes
redis__maxmemory	—	gauge	Memory limit, bytes
redis__cpu__user / __system	process = server | children	counter	CPU time, seconds
redis__clients__connected / __blocked	—	gauge	Connected/blocked clients
redis__clients__connections_received	—	counter	Accepted connections
redis__connections__rejected	—	counter	Rejected connections
redis__commands__total_calls	—	counter	Total commands executed
redis__commands__calls / __seconds	command	counter	Calls and total time per command
redis__cache__evicted_keys	—	counter	Evicted keys
redis__keys__hits / __misses	—	counter	Key hits/misses
redis__db__keys / __expires	db	gauge	Keys in the database / with TTL
redis__pubsub__channels / __patterns	—	gauge	Pub/sub channels and patterns
redis__persistence__rdb__changes_since_last_save	—	gauge	Changes since the last RDB save
redis__persistence__rdb__time_since_last_save	—	gauge	Seconds since the last save
redis__persistence__rdb__loading_in_progress / __bgsave_in_progress	—	gauge	0/1 — load/save in progress
redis__persistence__aof__enabled / __rewrite_in_progress	—	gauge	0/1 — AOF enabled / rewrite in progress
redis__persistence__aof__current_size / __buffer_length	—	gauge	AOF and buffer size, bytes
redis__persistence__aof__pending_bio_fsync / __delayed_fsync	—	gauge	Pending fsyncs
redis__replication__connected_slaves	—	gauge	Connected replicas (on the master)
redis__replication__last_io_seconds_ago	—	gauge	Seconds since the last exchange with the master (on a replica)
redis__replication__sync_in_progress / __sync_left_bytes / __link_down_since_seconds	—	gauge	Sync state with the master

Memcached (plugin memcached)

Metric	Labels	Type	Description
memcached__version	—	text	Version
memcached__uptime	—	gauge	Uptime, seconds
memcached__cpu__user / __system	—	counter	CPU time (rusage), seconds
memcached__items__current / __total	—	gauge/counter	Items cached now / stored ever
memcached__bytes__used / __max / __malloced	—	gauge	Used/limit/allocated, bytes
memcached__connections__current / __total	—	gauge/counter	Connections
memcached__evictions / __reclaimed	—	counter	Evictions/reclaims of entries
memcached__expired_unfetched / __evicted_unfetched / __crawler_reclaimed / __lrutail_reflocked	—	counter	LRU eviction details
memcached__traffic__in / __out	—	counter	Traffic, bytes
memcached__threads	—	gauge	Threads
memcached__command__requests / __hits / __misses / __badval	command (get, set, cas, …)	counter	Per-command statistics
memcached__slab__chunk_size	chunk_size	gauge	Slab class chunk size
memcached__slab__chunks__free / __used	chunk_size	gauge	Free/used chunks
memcached__slab__bytes__free / __used	chunk_size	gauge	Same in bytes

Elasticsearch (plugin elasticsearch)

Common label for all metrics — cluster. Cluster metrics are published only by the master node.

Metric	Extra labels	Type	Description
elasticsearch__version	—	text	Version
elasticsearch__cluster__status	—	status	Cluster health: green — value 1, otherwise 0 + status text
elasticsearch__index__settings	index, setting = number_of_replicas | number_of_shards	gauge	Index settings
elasticsearch__shards__count	index, shard_type = primary | replica, shard_state = active | initializing | relocating | unassigned	gauge	Shards by type and state
elasticsearch__index__size	index	gauge	Index size, bytes (top-50 indexes)
elasticsearch__index__docs__count / __deleted	index	gauge	Documents: total / marked deleted
elasticsearch__index__segments__count / __size	index	gauge	Lucene segments: count and memory
elasticsearch__index__operations__count__rate / __time__rate	index, operation = get | query | fetch | scroll | merge | index | delete | flush | refresh | warmer | …	rate	Index operations: frequency and time spent
elasticsearch__index__cache__size / __objects	index, cache_type = query | fielddata | request	gauge	Index caches
elasticsearch__index__cache__hits__rate / __misses__rate / __evictions__rate / __caches__rate	index, cache_type	rate	Cache activity
elasticsearch__index__translog__size / __operations__rate	index	gauge/rate	Translog
elasticsearch__http__current_open__rate	—	rate	HTTP connection openings
elasticsearch__transport__rx_count__rate / __rx_bytes__rate / __tx_count__rate / __tx_bytes__rate	—	rate	Inter-node traffic
elasticsearch__script__compilations__rate / __cache_evictions__rate	—	rate	Script compilations
elasticsearch__thread_pool__threads / __queue / __active	thread_pool	gauge	Thread pools
elasticsearch__thread_pool__rejected__rate / __completed__rate	thread_pool	rate	Rejected/completed tasks
elasticsearch__breaker__limit / __estimated	breaker	gauge	Circuit breakers: limit and estimate, bytes
elasticsearch__breaker__overhead	breaker	gauge	Overhead coefficient
elasticsearch__breaker__tripped__rate	breaker	rate	Breaker trips

Queues, brokers, search

RabbitMQ (plugin rabbitmq)

Via the Management API; top-50 queues. Labels: vhost, queue.

Metric	Extra labels	Type	Description
rabbitmq__queue__messages	—	gauge	Messages in the queue
rabbitmq__queue__memory	—	gauge	Queue memory, bytes
rabbitmq__queue__published__rate	—	rate	Publish rate, messages/s
rabbitmq__queue__delivered__rate	ack = manual_ack | auto_ack	rate	Delivery rate to consumers
plugin__status	—	status	Errors: management plugin disabled, missing permissions, etc.

Kafka (plugin kafka)

Instance-level labels: broker_id, cluster_id. Partition metrics are published by the partition leader broker.

Metric	Labels	Type	Description
kafka__version	—	text	Kafka version
kafka__brokers__count	—	gauge	Brokers in the cluster
kafka__topic__partition__count	topic	gauge	Partitions per topic
kafka__topic__partition__size	topic, partition	gauge	Partition size on disk, bytes
kafka__topic__partition__offset__newest / __oldest	topic, partition	gauge	Partition boundary offsets
kafka__topic__partition__replication__isr_count	topic, partition	gauge	In-sync replicas
kafka__topic__partition__replication__replicas_count	topic, partition	gauge	Replica count
kafka__consumer_group__status	group	status	Group state: empty — stable, otherwise text (preparingrebalance, …)
kafka__consumer_group__members__count	group	gauge	Group members
kafka__consumer_group__protocol / __protocol_type	group	text	Group protocol
kafka__consumer_group__topic__partition__offset__committed	topic, partition, group, member_host, member_id	gauge	Committed group offset (lag = newest − committed)

ZooKeeper (plugin zookeeper)

Metric	Labels	Type	Description
zookeeper__status	—	status	Empty — OK; zookeeper cluster is down / zookeeper node in standalone mode
plugin__status	—	status	Text please add commands in whitelist when the stat/mntr four-letter commands are disabled
zookeeper__version	—	text	Version
zookeeper__role	—	text	Node role (leader, follower)
zookeeper__nodes__current / __expected	—	gauge	Ensemble nodes: actual / per config
zookeeper__znodes__regular / __ephemerals	—	gauge	znode count: regular / ephemeral
zookeeper__watchers	—	gauge	Watcher count
zookeeper__leader_elections	—	gauge	Epoch number (leader election count)
zookeeper__current_transaction_number__<role>	—	gauge	Transaction counter from zxid (<role> — node mode)
zookeeper__client_latency__min / __avg / __max	—	gauge	Client request latency, seconds
zookeeper__packets__received / __sent	—	counter	Packets
zookeeper__outstanding_requests	—	gauge	Requests in flight
zookeeper__outstanding_requests__limit	—	gauge	Limit (globalOutstandingLimit)
zookeeper__connection__packets__queued / __received / __sent	client_ip	rate	Per-client-connection activity

Sphinx (plugin sphinx)

Metric	Labels	Type	Description
sphinx__uptime	—	gauge	searchd uptime
sphinx__connections__count / __overflows	—	counter	Connections / maxed_out rejections
sphinx__agent__connections / __retries	—	counter	Distributed search agent connections
sphinx__query__count	query_type = local | distributed	counter	Queries
sphinx__query__total_time	query_type	counter	Total query time
sphinx__query__time	query_type, stage = cpu | local | wait | disk_read	counter	Time per stage
sphinx__query__disk__ops__read / sphinx__query__disk__bytes__read	query_type="local"	counter	Disk reads
sphinx__command__calls	command	counter	Calls per command type (search, update, …)
sphinx__index__documents__count / __bytes	index, index_type	gauge	Documents in an index: count and size
sphinx__index__memory__used / __limit	index, index_type	gauge	Index memory
sphinx__index__disk__used	index, index_type	gauge	Disk used by the index

Resque (plugin resque, configured)

Requires a YAML config with redis_address and prefix.

Metric	Labels	Type	Description
resque__tasks__pending	queue	gauge	Tasks in the queue
resque__tasks__processed	worker, queue	rate	Tasks processed/s
resque__tasks__failed	worker, queue	rate	Failures/s

Web servers and proxies

Nginx (plugin nginx)

Metrics are built by parsing access logs (the format is detected automatically from the nginx config) and by checking TLS certificates from the config. Common log-metric labels: file (log path), log_format.

Metric	Extra labels	Type	Description
nginx__requests__rate	method, status, cache_status, url	rate	Requests/s; url — top-N URLs with normalization, the rest — ~other
nginx__traffic__rate	url	rate	Bytes served per second
nginx__response_time__percentiles	percentile = 50 | 75 | 95 | 99	gauge	Response time percentiles, seconds
nginx__response_time__histogram	method, cache_status, url, levels, level	rate	Response time histogram (default buckets 0.5 and 1 s)
nginx__upstream_response_time__percentiles	percentile	gauge	Upstream response time percentiles
nginx__upstream_response_time__histogram	same as histogram	rate	Upstream response time histogram
nginx__ssl_certificate__seconds_to_expire	certificate_file	gauge	Seconds until certificate expiry
nginx__ssl_certificate__revoke_status	certificate_file	gauge	OCSP: 1 — good, 0 — revoked
nginx__ssl_certificate__check_status	certificate_file	status	Certificate read/parse error
status	access_log or log_format	status	Configuration problems: unreadable log, unparsed log_format

PHP-FPM (plugin php_fpm)

Via the FCGI status page (pm.status_path). Common label — pool.

Metric	Extra labels	Type	Description
php_fpm__pool__uptime	—	gauge	Pool uptime, seconds
php_fpm__pool__accepted_connections	—	counter	Accepted connections
php_fpm__pool__listen_queue__current / __limit	—	gauge	Request queue: current/limit
php_fpm__pool__processes__count	state = active | idle	gauge	Workers per state
php_fpm__pool__processes__limit_reached	—	counter	Times max_children was reached
plugin__status / status	pool	status	Status page disabled / connection error

Envoy (plugin envoy)

Transparent passthrough of the Envoy admin interface metrics (/stats?format=prometheus): names and labels are forwarded as is (e.g. envoy_cluster_upstream_rq_total, envoy_server_live). Only counters and gauges are forwarded; histograms are skipped.

JVM and Cassandra

JVM (plugin jvm)

For every Java process with JMX enabled (without authentication). Instance-level labels: jvm, user.

Metric	Labels	Type	Description
jvm__memory__heap__used / __max	—	gauge	Heap: used/max, bytes
jvm__memory__nonheap__used / __max	—	gauge	Non-heap memory
jvm__memory__pool__used / __max	pool	gauge	Memory pools (Eden, Old Gen, Metaspace, …)
jvm__gc__counter	collector	counter	Garbage collections per collector
jvm__gc__total_time	collector	counter	GC time, seconds
jvm__threads__current / __daemon	—	gauge	Live/daemon threads
jvm__threads__created	—	counter	Total threads created
jvm__uptime	—	gauge	JVM uptime, milliseconds
status	—	status	JMX disabled, please enable it / JMX authentication enabled, please disable it, etc.

Cassandra (branch of the jvm plugin, published as plugin=cassandra)

Extra labels: cluster, datacenter, rack.

Family	Metrics	Labels
Version	cassandra__version (text)	—
Caches	cassandra__row_cache__{max_bytes,used_bytes,hits,requests,entries}, cassandra__key_cache__{…}, cassandra__counter_cache__{…}	—
Client requests	cassandra__client_requests__{count,unavailables,timeouts,failures,total_seconds}	operation (Read/Write/…), opt. consistency_level
Commit log	cassandra__commitlog__{waiting_on_segment_allocation,pending_tasks,waiting_on_commit,completed_tasks,total_size}	—
Compactions	cassandra__compaction__{completed_tasks,bytes_compacted,completed_compactions,pending_tasks}	—
Storage	cassandra__storage__{bytes_used,exceptions_count,hints_count,hints_in_progress}	—
CQL	cassandra__cql__prepared_statements__{executed,count,evicted}, cassandra__cql__regular_statements__executed	—
Read repair	cassandra__read_repair__{attempted,blocking,background}	—
Inter-node messaging	cassandra__nodes__{messages_received,messages_sent,pending_responses,pending_requests,messages_dropped,timeouts}	node, for dropped opt. type = large | small
Dropped messages	cassandra__dropped_messages__count	message_type
Tables	cassandra__table__{waiting_on_memtable_free,disk_space_used,sstables_count,pending_flushes,snapshots_size,compression_ratio,pending_compactions,speculative_retries,mean_row_size,min_row_size,max_row_size}, cassandra__table__row_cache__{hit,miss,hit_out_of_range}, cassandra__table__ops__{count,total_time} (+operation), cassandra__memtable__{size__heap,size__off_heap,switch_count,columns_count}, cassandra__bloom_filter__{false_positives,disk_space_used}	keyspace, table

Mail

Mail servers (plugin mail)

Auto-discovery of Exim and Postfix. Queue metrics come from direct polling; mail flow metrics from log parsing. Common label — mta = exim | postfix; log metrics also carry file.

Metric	Extra labels	Type	Description
mail__queue__size	queue (postfix: maildrop, hold, incoming, active, deferred; exim: main)	gauge	Messages in the queue
mail__messages__received	—	rate	Messages received/s
mail__messages__delivered	postfix: recipient_domain	rate	Messages delivered/s
mail__messages__deferred	—	rate	Deferred/s
mail__messages__failed	—	rate	Failed (bounced)/s
mail__delivery_errors	response_code (SMTP 4xx/5xx)	rate	Delivery errors per code
mail__delivery_time__histogram	recipient_domain, levels, level	rate	Postfix delivery time histogram (buckets 5 s/1 min/5 min/30 min)
status	—	status	Exim log parser initialization error

The syslog plugin publishes no metrics of its own — it locates log files for Postfix.

Security

File permissions (plugin security_file_permissions)

Metric	Labels	Type	Description
security_file_permissions	file_path, mode	gauge	Existence (0/1) and access mode of critical files: /etc/passwd, /etc/shadow, /etc/group, /etc/gshadow, /etc/hosts, /etc/fstab, etc.
security_file_permissions_errors_total	—	gauge	Number of files whose check failed

Custom metrics

Plugins in this section publish metrics whose names are defined by the user or derived from data. Configured via YAML files in config.d/.

StatsD (plugin statsd)

Built-in StatsD server (UDP). The metric name comes from the StatsD packet; name parts of the form key=value or key_is_value are extracted into labels. Label values starting with / are normalized as URL paths; password is replaced with ~hidden.

StatsD type	Published metrics
gauge (g)	<name> — gauge
counter (c, m)	<name>__rate — rate (events/s)
timer (ms, h)	<name>__lower, <name>__upper (min/max), <name>__rate (events/s), <name>__sum (sum/s), <name>__mean (average), <name>__percentile with label percentile = 50 | 75 | 90 | 95 | 97 | 99; values in seconds
set (s)	not supported

Logparser (plugin logparser)

Turns log lines into metrics per config: regex + a metrics list with types rate, threshold (histogram), percentiles, max, min. The metric name is the name field from the config (as is, dots → __). Labels come from the config; expressions, top-N variables (rare values → ~other), and User-Agent parsing are supported. The file label is added to all metrics.

Metric	Type	Description
<name from config>	rate/gauge	User metric; threshold adds levels, level labels; percentiles adds percentile
logparser__lines__succeeded	rate	Successfully parsed lines/s
logparser__lines__unmatched	rate	Lines that did not match the regex
logparser__lines__failed	rate	Lines with value parse errors

SQL queries (plugins mysql_query, postgresql_query, mssql_query, clickhouse_query)

Runs SQL from the config. The metric name is the metric_name field. The value result column is the value; the remaining columns become labels.

Metric	Type	Description
<metric_name>	gauge	Values of the result rows
status	status	Connection/query error or OK

Redis queries (plugin redis_query)

Runs Redis commands from the config. Metric name: <key>__<command> (e.g. GET mykey → mykey__get); command arguments go into the param label.

Command execution (plugin execute)

Runs an executable and parses stdout (regex with named groups or JSON). The metric name comes from the config (name) or the JSON name field.

Metric	Type	Description
<name>	gauge/text	Values from the command output
status	status	Exit code / parse error
duplicate_metric_count	gauge	Number of duplicated metrics in the output

HTTP checks (plugin http)

Performs an HTTP request; metrics are extracted from the response body with regex groups (metrics[].metric in the config).

Metric	Type	Description
<name from config>	gauge	Value from the response body
http__request_time	gauge	Request duration, seconds
status	status	Request error / response mismatch

Go expvar (plugin expvar)

Reads /debug/vars of a Go application; a fixed set from memstats is published:

go__memory__bytes__alloc, go__memory__bytes__sys, go__memory__total_alloc, go__memory__ops (label operation = lookup | malloc | free), go__memory__heap__alloc, go__memory__heap__sys, go__memory__heap__bytes (label state = used | idle), go__gc__count, go__gc__total_pause, go__gc__max_pause, plus status.

Prometheus exporters (plugin prometheus)

Scrapes any OpenMetrics/Prometheus endpoint and forwards metrics with their original names and labels (no prefixes added). Every metric additionally gets a metric_type label (the exposition type). Histograms are expanded into <name>_count, <name>_sum, <name>_bucket (+le); summaries into <name>_count, <name>_sum, <name> (+quantile). Filtering — metrics_white_list / metrics_black_list by name prefix; the per-endpoint series limit is 5000 (when exceeded, only plugin__status with an error is published).

Service metrics: status (scrape success), plugin__status (label source — config file).

Internal agent metrics

Self-monitoring (plugins heartbeat, okagent_stat)

Metric	Labels	Type	Description
okagent_heartbeat	—	gauge	=1; agent liveness flag
okagent_up	—	gauge	Agent process uptime, seconds
okagent_uuid	—	text	Agent UUID
okagent__version	—	text	Agent version
okagent__memstats__alloc / __sys / __heap__* / __stack__* / …	—	gauge	Agent Go runtime memory (full runtime.MemStats set)
okagent__gc__time / __count	—	counter	Agent GC: total pause time and cycle count
lock_status	agent_random_uuid	gauge	0 — lock file acquired normally; 1 — a second agent instance is running; 2 — lock check error

Internal counters (via the okagent_stat plugin)

Internal counters are published with the okagent__ prefix. Main families:

Metric	Labels	Description
okagent__agent__stale_nans, …__plugin_metrics_swapped, …__frequent_packages, …__multipackage_labelset, …__duplicate_labelset, …__plugin_instances, …__duplicate_ls, …__dns_fallback	per counter	Metric pipeline diagnostics: staleness markers, duplicates, DNS fallbacks
okagent__metric_group__metrics_dropped	reason = timestamp | cardinality | invalid, pname	Dropped metrics: stale timestamp, cardinality overflow, invalid name
okagent__metric_writer__metricpacks_dropped, …__metrics_dropped, …__buffer_input_metrics, …__switch_proxy, okagent__writer__empty_labels	writer, reason	Transport: buffer overflow, proxy switchover
okagent__scheduler__ticks_skipped	obj_name, func_name	Skipped scheduler ticks (a plugin did not finish within the interval)
okagent__process_multiplexer__pushes_skipped	subscriber	Skipped process list deliveries to subscribers
okagent__statsd__packets__udp, …__errors, …__measurements, …__metrics	type, mtype	StatsD server diagnostics
okagent__sniffer__errors, …__skipped_packets, …__non_data_packets, …__duplicate_packets	error, proto	Sniffer diagnostics (MongoDB)
okagent__mongo__errors	error	MongoDB protocol parse errors
okagent__pg__settings__unknown_variables / __missed_variables	instance	Unrecognized/missing pg_settings parameters

Internal delivery metrics (plugin prometheus_internal_scraper)

Publishes everything from the agent’s internal Prometheus registry — primarily the remote write delivery subsystem metrics (odarix_core_delivery_*): queue state, sending, refill, transport. Names are forwarded as is.

Status metric conventions

Metric	Description
status	Status of a plugin instance (configured plugins and some auto-discovered ones)
plugin__status	Integration status: empty text (value 1) — collection works; error text (value 0) — what needs fixing
plugin__status__invalid_names	Appears when a plugin produced metrics with invalid names (they were dropped; the list is in the text)