up{instance=~".*:9100"}

up{instance=~".*:9101"}

up{instance=~".*:9650"}

up{instance=~".*:9323"}

up{instance=~".*:9598"}

up{job="etcd"}

up{job="minio"}

up{instance=~".*:9630"}

up{instance=~".*:9631"}

up{instance=~".*:9854"}

up{instance=~".*:8008"}

up{job="mongo"}

up{job="mysql"}

up{job="kafka"}

etcd_up < 1

min(etcd_server_has_leader) by (cls) < 1

etcd:cls:quota_usage > 0.90

etcd:ins:network_peer_rt_p95_5m > 0.200

etcd:ins:wal_fsync_rt_p95_5m > 0.050

sum without (type, client_api_version) (increase(etcd_server_client_requests_total{job="etcd"}[1m]))

increase(etcd_server_proposals_committed_total{job="etcd"}[1m])

etcd_mvcc_db_total_size_in_bytes{job="etcd"} / etcd_server_quota_backend_bytes{job="etcd"}

max by (job,cls) (etcd_mvcc_db_total_size_in_bytes{job="etcd"} / etcd_server_quota_backend_bytes{job="etcd"})

etcd_debugging_mvcc_current_revision{job="etcd"} - etcd_debugging_mvcc_compact_revision{job="etcd"}

max by (job,cls) (etcd_debugging_mvcc_current_revision{job="etcd"} - etcd_debugging_mvcc_compact_revision{job="etcd"})

histogram_quantile(0.95, rate(etcd_disk_wal_fsync_duration_seconds_bucket{job="etcd"}[5m]))

histogram_quantile(0.95, rate(etcd_disk_backend_commit_duration_seconds_bucket{job="etcd"}[5m]))

histogram_quantile(0.95, rate(etcd_network_peer_round_trip_time_seconds_bucket{job="etcd"}[5m]))

infra_up < 1

agent_up < 1

up{job="infra"}

up{job!~"infra|node|etcd|minio|pgsql|redis|mongo|mysql"}

kafka_exporter_up < 1

sum without (partition) (rate(kafka_topic_partition_current_offset{}[1m]))

sum without (partition) (rate(kafka_topic_partition_current_offset{}[5m]))

sum(kafka:topic:msg_rate1m) without (topic)

sum(kafka:topic:msg_rate5m) without (topic)

sum(kafka:topic:msg_rate1m) by (job,cls)

sum(kafka:topic:msg_rate5m) by (job,cls)

sum without (partition) (rate(kafka_consumergroup_current_offset{}[1m]))

sum without (partition) (rate(kafka_consumergroup_current_offset{}[5m]))

sum(kafka:topic:msg_rate1m) without (topic)

sum(kafka:topic:msg_rate5m) without (topic)

sum(kafka:topic:msg_rate1m) by (job,cls)

sum(kafka:topic:msg_rate5m) by (job,cls)

minio_up < 1

avg_over_time(minio_cluster_nodes_offline_total{job="minio"}[5m]) > 0

avg_over_time(minio_cluster_disk_offline_total{job="minio"}[5m]) > 0

mysql_up < 1

node_up < 1

haproxy_up < 1

vector_up < 1

docker_up < 1

keepalived_up < 1

node:ins:cpu_usage_1m > 0.70

node:ins:stdload1 > 1

node:ins:mem_avail < 0.10

node:ins:swap_usage > 0.01

node:fs:space_usage > 0.90

node:fs:inode_usage > 0.90

node:ins:fd_usage > 0.90

node:dev:disk_read_rt_1m{device="dfa"} > 0.032 or node:dev:disk_write_rt_1m{device="dfa"} > 0.032

rate(node:ins:tcp_error[1m]) > 1

node:ins:tcp_retrans_ratio1m > 1e-2

node_timex_sync_status != 1

time() - node_boot_time_seconds

sum without (machine, release, version, sysname, domainname) (
  label_replace(node_uname_info, "id", "$1", "nodename", "(.+)") OR
  label_replace(node_uname_info, "id", "$1", "instance", "(.+)\\:\\d+") OR
  label_replace(node_uname_info, "id", "$1", "ins", "(.+)")
)

irate(node_cpu_seconds_total[1m])

sum without (mode) (node:cpu:time_irate1m)

sum without (mode) (node:cpu:time_irate1m{mode="idle"})

1 - node:cpu:idle_time_irate1m / node:cpu:total_time_irate1m

avg_over_time(node:cpu:usage[1m])

avg_over_time(node:cpu:usage[5m])

avg_over_time(node:cpu:usage[15m])

count without (cpu) (node:cpu:usage)

sum by (cls, job) (node:ins:cpu_count)

sum by (job) (node:ins:cpu_count)

avg without (cpu) (node:cpu:usage)

avg by (cls, job) (node:cpu:usage)

sum by (job) (node:cls:cpu_usage * node:cls:cpu_count) / sum by (job) (node:cls:cpu_count)

avg_over_time(node:ins:cpu_usage[1m])

avg_over_time(node:ins:cpu_usage[5m])

avg_over_time(node:ins:cpu_usage[15m])

avg_over_time(node:cls:cpu_usage[1m])

avg_over_time(node:cls:cpu_usage[5m])

avg_over_time(node:cls:cpu_usage[15m])

avg_over_time(node:env:cpu_usage[1m])

avg_over_time(node:env:cpu_usage[5m])

avg_over_time(node:env:cpu_usage[15m])

rate(node_schedstat_timeslices_total[1m])

sum without (cpu) (node:cpu:sched_timeslices)

rate(node_schedstat_waiting_seconds_total[1m])

avg without (cpu) (node:cpu:sched_wait_rate1m)

rate(node_forks_total[1m])

rate(node_intr_total[1m])

rate(node_context_switches_total{}[1m])

node_load1 / node:ins:cpu_count

node_load5 / node:ins:cpu_count

node_load15 / node:ins:cpu_count

sum by (cls, job) (node_load1) / node:cls:cpu_count

sum by (cls, job) (node_load5) / node:cls:cpu_count

sum by (cls, job) (node_load15) / node:cls:cpu_count

sum by (job) (node:cls:stdload1) / node:env:cpu_count

sum by (job) (node:cls:stdload5) / node:env:cpu_count

sum by (job) (node:cls:stdload15) / node:env:cpu_count

node_memory_KernelStack_bytes +
node_memory_VmallocUsed_bytes +
node_memory_HardwareCorrupted_bytes +
node_memory_PageTables_bytes +
node_memory_Slab_bytes +
node_memory_PageTables_bytes

node_memory_MemTotal_bytes -
(node_memory_KernelStack_bytes +
node_memory_VmallocUsed_bytes +
node_memory_HardwareCorrupted_bytes +
node_memory_PageTables_bytes +
node_memory_Slab_bytes +
node_memory_PageTables_bytes) - 
node_memory_Buffers_bytes -
(node_memory_Hugepagesize_bytes * node_memory_HugePages_Total) -
(node_memory_Cached_bytes - node_memory_Mapped_bytes ) -
node_memory_MemFree_bytes

node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes

1 - node_memory_MemAvailable_bytes / node_memory_MemTotal_bytes

1 - node_memory_SwapFree_bytes / node_memory_SwapTotal_bytes

node_memory_Committed_AS_bytes / node_memory_CommitLimit_bytes

1 - sum by (cls, job) (node_memory_MemAvailable_bytes) / sum by (cls, job) (node_memory_MemTotal_bytes)

sum by (job) (node_memory_MemTotal_bytes)

sum by (job) (node_memory_MemAvailable_bytes)

1 - node:env:mem_avail / node:env:mem_total

rate(node_vmstat_pgfault[1m])

rate(node_vmstat_pgmajfault[1m])

rate(node_vmstat_pgpgin[1m])

rate(node_vmstat_pgpgout[1m])

rate(node_vmstat_pswpin[1m])

rate(node_vmstat_pswpout[1m])

rate(node_disk_io_time_seconds_total{device=~"[a-z0-9A-Z-_]+"}[1m])

rate(node_disk_io_time_weighted_seconds_total{device=~"[a-z0-9A-Z-_]+"}[1m])

rate(node_disk_reads_completed_total{device=~"[a-z0-9A-Z-_]+"}[1m])

sum without (device) (node:dev:disk_reads_rate1m)

sum by (cls, job) (node:ins:disk_reads_rate1m)

rate(node_disk_writes_completed_total{device=~"[a-z0-9A-Z-_]+"}[1m])

sum without (device) (node:dev:disk_writes_rate1m)

sum by (cls, job) (node:ins:disk_writes_rate1m)

rate(node_disk_reads_merged_total{device=~"[a-z0-9A-Z-_]+"}[1m])

sum without (device) (node:dev:disk_mreads_rate1m)

sum by (cls, job) (node:ins:disk_mreads_rate1m)

rate(node_disk_writes_merged_total{device=~"[a-z0-9A-Z-_]+"}[1m])

sum without (device) (node:dev:disk_mwrites_rate1m)

sum by (cls, job) (node:ins:disk_mwrites_rate1m)

node:dev:disk_reads_rate1m + node:dev:disk_writes_rate1m

node:ins:disk_reads_rate1m + node:ins:disk_writes_rate1m

node:cls:disk_reads_rate1m + node:cls:disk_writes_rate1m

node:dev:disk_mreads_rate1m / (node:dev:disk_reads_rate1m + node:dev:disk_mreads_rate1m)

node:ins:disk_mreads_rate1m / (node:ins:disk_reads_rate1m + node:ins:disk_mreads_rate1m)

node:cls:disk_mreads_rate1m / (node:cls:disk_reads_rate1m + node:cls:disk_mreads_rate1m)

node:dev:disk_mwrites_rate1m / (node:dev:disk_writes_rate1m + node:dev:disk_mwrites_rate1m)

node:ins:disk_mwrites_rate1m / (node:ins:disk_writes_rate1m + node:ins:disk_mwrites_rate1m)

node:cls:disk_mwrites_rate1m / (node:cls:disk_writes_rate1m + node:cls:disk_mwrites_rate1m)

rate(node_disk_read_bytes_total{device=~"[a-z0-9A-Z-_]+"}[1m])

sum without (device) (node:dev:disk_read_bytes_rate1m)

sum by (cls, job) (node:ins:disk_read_bytes_rate1m)

rate(node_disk_written_bytes_total{device=~"[a-z0-9A-Z-_]+"}[1m])

sum without (device) (node:dev:disk_write_bytes_rate1m)

sum by (cls, job) (node:ins:disk_write_bytes_rate1m)

node:dev:disk_read_bytes_rate1m + node:dev:disk_write_bytes_rate1m

node:ins:disk_read_bytes_rate1m + node:ins:disk_write_bytes_rate1m

node:cls:disk_read_bytes_rate1m + node:cls:disk_write_bytes_rate1m

rate(node_disk_read_time_seconds_total{device=~"[a-z0-9A-Z-_]+"}[1m])

rate(node_disk_write_time_seconds_total{device=~"[a-z0-9A-Z-_]+"}[1m])

node:dev:disk_read_time_rate1m + node:dev:disk_write_time_rate1m

node:dev:disk_read_time_rate1m / node:dev:disk_reads_rate1m

node:dev:disk_write_time_rate1m / node:dev:disk_writes_rate1m

node:dev:disk_io_time_rate1m / node:dev:disk_iops_1m

node:dev:disk_read_bytes_rate1m / node:dev:disk_reads_rate1m

node:dev:disk_write_bytes_rate1m / node:dev:disk_writes_rate1m

node:dev:disk_io_bytes_rate1m / node:dev:disk_iops_1m

node_filesystem_free_bytes{fstype!~"(n|root|tmp)fs.*"}

node_filesystem_avail_bytes{fstype!~"(n|root|tmp)fs.*"}

node_filesystem_size_bytes{fstype!~"(n|root|tmp)fs.*"}

sum without(device,mountpoint,fstype) (node:fs:free_bytes{fstype!~"fuse.juicefs"})

sum without(device,mountpoint,fstype) (node:fs:avail_bytes{fstype!~"fuse.juicefs"})

sum without(device,mountpoint,fstype) (node:fs:size_bytes{fstype!~"fuse.juicefs"})

sum by (job,cls) (node:ins:free_bytes)

sum by (job,cls) (node:ins:avail_bytes)

sum by (job,cls) (node:ins:size_bytes)

sum by(job) (node:ins:free_bytes)

sum by(job) (node:ins:avail_bytes)

sum by(job) (node:ins:size_bytes)

1 - node:fs:avail_bytes / node:fs:size_bytes

1 - node:cls:avail_bytes / node:cls:size_bytes

1 - node:env:avail_bytes / node:env:size_bytes

max without (device, fstype, mountpoint) (node:fs:space_usage)

max by (job, cls) (node:ins:space_usage_max)

max by (job, device, fstype, mountpoint) (node:fs:space_usage)

deriv(node:fs:avail_bytes[1h])

clamp(node:fs:avail_bytes / - node:fs:space_deriv1h, -1, 126230400)

predict_linear(node:fs:avail_bytes[1h], 86400)

node_filesystem_files_free{fstype!~"(n|root|tmp)fs.*"}

node_filesystem_files{fstype!~"(n|root|tmp)fs.*"}

node:fs:inode_total - node:fs:inode_free

1 - (node:fs:inode_free / node:fs:inode_total)

sum without (fstype, device, mountpoint) (node:fs:inode_used{fstype!~"fuse.juicefs"}) /
sum without (fstype, device, mountpoint) (node:fs:inode_total{fstype!~"fuse.juicefs"})

node_filefd_allocated / node_filefd_maximum

rate(node_filefd_allocated[1m])

rate(node_network_transmit_packets_total[1m])

sum without (device) (node:dev:network_tx_pps1m{device!~"lo|bond.*"})

sum by (cls, job) (node:ins:network_tx_pps1m)

rate(node_network_receive_packets_total[1m])

sum without (device) (node:dev:network_rx_pps1m{device!~"lo|bond.*"})

sum by (cls, job) (node:ins:network_rx_pps1m)

rate(node_network_transmit_bytes_total[1m])

sum without (device) (node:dev:network_tx_bytes_rate1m{device!~"lo|bond.*"})

sum by (cls, job) (node:ins:network_tx_bytes_rate1m)

rate(node_network_receive_bytes_total[1m])

sum without (device) (node:dev:network_rx_bytes_rate1m{device!~"lo|bond.*"})

sum by (cls, job) (node:ins:network_rx_bytes_rate1m)

node:dev:network_tx_bytes_rate1m + node:dev:network_rx_bytes_rate1m

node:ins:network_tx_bytes_rate1m + node:ins:network_rx_bytes_rate1m

node:cls:network_tx_bytes_rate1m + node:cls:network_rx_bytes_rate1m

rate(node_netstat_Tcp_InSegs[1m])

rate(node_netstat_Tcp_OutSegs[1m])

rate(node_netstat_Tcp_RetransSegs[1m])

node:ins:tcp_insegs_rate1m + node:ins:tcp_outsegs_rate1m

node:ins:tcp_retranssegs_rate1m / node:ins:tcp_outsegs_rate1m

node_netstat_TcpExt_ListenOverflows +
node_netstat_TcpExt_ListenDrops +
node_netstat_Tcp_InErrs

rate(node:ins:tcp_error[1m])

rate(node_netstat_Tcp_PassiveOpens[1m])

rate(node_netstat_Tcp_ActiveOpens[1m])

rate(node_netstat_Tcp_AttemptFails[1m])

rate(node_netstat_Tcp_EstabResets[1m])

rate(node_netstat_TcpExt_ListenOverflows[1m])

rate(node_netstat_TcpExt_ListenDrops[1m])

rate(node_netstat_Udp_InDatagrams{}[1m])

rate(node_netstat_Udp_OutDatagrams[1m])

abs(node_timex_offset_seconds)

max by (cls, job) (node:ins:time_drift)

max by (cls, job) (node_ntp_offset_seconds) - min by (cls, job) (node_ntp_offset_seconds)

(100 - haproxy_process_idle_time_percent) / 100

avg by (cls, job) (haproxy:ins:usage)

time() - haproxy_process_start_time_seconds

sum by (ins,proxy) (haproxy_server_status{state="UP", proxy!='stats'}) > bool 0

sum by (cls,proxy) (haproxy_server_status{state="UP", proxy!='stats'}) > bool 0

sum without (list) (pgbouncer_list_items{list="pools"})

sum without (list) (pgbouncer_list_items{list="databases"})

sum without (list) (pgbouncer_list_items{list="users"})

sum without (list) (pgbouncer_list_items{list="login_clients"})

sum without (list) (pgbouncer_list_items{list="free_clients"})

sum without (list) (pgbouncer_list_items{list="used_clients"})

sum without (list) (pgbouncer_list_items{list="free_servers"})

pgbouncer_database_pool_size{datname!="pgbouncer"} + pgbouncer_database_reserve_pool{datname!="pgbouncer"}

pgbouncer_database_pool_size{datname!="pgbouncer"}

pgbouncer_database_reserve_pool{datname!="pgbouncer"}

pgbouncer_database_current_connections{datname!="pgbouncer"}

pgbouncer_database_paused{datname!="pgbouncer"}

pgbouncer_database_disabled{datname!="pgbouncer"}

pgbouncer:db:pool_current_conn / pgbouncer:db:pool_size

pgbouncer:db:pool_current_conn / pgbouncer:db:pool_max_conn

pgbouncer_pool_maxwait{datname!="pgbouncer"} + pgbouncer_pool_maxwait_us{datname!="pgbouncer"} / 1000000

max without(user, pool_mode) (pgbouncer:pool:maxwait)

max without(datname) (pgbouncer:db:maxwait)

max by (job, cls) (pgbouncer:ins:maxwait)

max by (job) (pgbouncer:cls:maxwait)

pgbouncer_pool_active_servers{datname!="pgbouncer"}

sum without(user, pool_mode) (pgbouncer:pool:active_servers)

sum without(datname) (pgbouncer:db:active_servers)

sum by (job, cls) (pgbouncer:ins:active_servers)

sum by (job) (pgbouncer:cls:active_servers)

pgbouncer_pool_idle_servers{datname!="pgbouncer"}

max without(user, pool_mode) (pgbouncer:pool:idle_servers)

sum without(datname) (pgbouncer:db:idle_servers)

sum by (job, cls) (pgbouncer:ins:idle_servers)

sum by (job) (pgbouncer:cls:idle_servers)

pgbouncer_pool_used_servers{datname!="pgbouncer"}

max without(user, pool_mode) (pgbouncer:pool:used_servers)

sum without(datname) (pgbouncer:db:used_servers)

sum by (job, cls) (pgbouncer:ins:used_servers)

sum by (job) (pgbouncer:cls:used_servers)

pgbouncer_pool_tested_servers{datname!="pgbouncer"}

max without(user, pool_mode) (pgbouncer:pool:tested_servers)

sum without(datname) (pgbouncer:db:tested_servers)

sum by (job, cls) (pgbouncer:ins:tested_servers)

sum by (job) (pgbouncer:cls:tested_servers)

pgbouncer_pool_login_servers{datname!="pgbouncer"}

max without(user, pool_mode) (pgbouncer:pool:login_servers)

sum without(datname) (pgbouncer:db:login_servers)

sum by (job, cls) (pgbouncer:ins:login_servers)

sum by (job) (pgbouncer:cls:login_servers)

pgbouncer_pool_active_clients{datname!="pgbouncer"}

max without(user, pool_mode) (pgbouncer:pool:active_clients)

sum without(datname) (pgbouncer:db:active_clients)

sum by (job, cls) (pgbouncer:ins:active_clients)

sum by (job) (pgbouncer:cls:active_clients)

pgbouncer_pool_waiting_clients{datname!="pgbouncer"}

max without(user, pool_mode) (pgbouncer:pool:waiting_clients)

sum without(datname) (pgbouncer:db:waiting_clients)

sum by (job, cls) (pgbouncer:ins:waiting_clients)

sum by (job) (pgbouncer:cls:waiting_clients)

rate(pgbouncer_stat_total_sent{datname!="pgbouncer"}[1m])

sum without (datname) (pgbouncer:db:tx_rate1m)

sum by (job, cls) (pgbouncer:ins:tx_rate1m)

sum by (job) (pgbouncer:cls:tx_rate1m)

rate(pgbouncer_stat_total_received{datname!="pgbouncer"}[1m])

sum without (datname) (pgbouncer:db:rx_rate1m)

sum by (job, cls) (pgbouncer:ins:rx_rate1m)

sum by (job) (pgbouncer:cls:rx_rate1m)

rate(pgbouncer_stat_total_xact_count{datname!="pgbouncer"}[1m])

sum without(datname) (pgbouncer:db:xact_rate1m)

sum by (job, cls) (pgbouncer:ins:xact_rate1m)

sum by (job) (pgbouncer:cls:xact_rate1m)

rate(pgbouncer_stat_total_xact_count{datname!="pgbouncer"}[5m])

sum without(datname) (pgbouncer:db:xact_rate5m)

sum by (job, cls) (pgbouncer:ins:xact_rate5m)

sum by (job) (pgbouncer:cls:xact_rate5m)

rate(pgbouncer_stat_total_xact_count{datname!="pgbouncer"}[15m])

sum without(datname) (pgbouncer:db:xact_rate15m)

sum by (job, cls) (pgbouncer:ins:xact_rate15m)

sum by (job) (pgbouncer:cls:xact_rate15m)

rate(pgbouncer_stat_total_xact_time{datname!="pgbouncer"}[1m])

sum without(datname) (pgbouncer:db:xact_time_rate1m)

sum by (job, cls) (pgbouncer:ins:xact_time_rate1m)

sum by (job) (pgbouncer:cls:xact_time_rate1m)

rate(pgbouncer_stat_total_xact_time{datname!="pgbouncer"}[5m])

sum without(datname) (pgbouncer:db:xact_time_rate5m)

sum by (job, cls) (pgbouncer:ins:xact_time_rate5m)

sum by (job) (pgbouncer:cls:xact_time_rate5m)

rate(pgbouncer_stat_total_xact_time{datname!="pgbouncer"}[15m])

sum without(datname) (pgbouncer:db:xact_time_rate15m)

sum by (job, cls) (pgbouncer:ins:xact_time_rate15m)

sum by (job) (pgbouncer:cls:xact_time_rate15m)

pgbouncer:db:xact_time_rate1m / pgbouncer:db:xact_rate1m

pgbouncer:ins:xact_time_rate1m / pgbouncer:ins:xact_rate1m

pgbouncer:cls:xact_time_rate1m / pgbouncer:cls:xact_rate1m

pgbouncer:env:xact_time_rate1m / pgbouncer:env:xact_rate1m

pgbouncer:db:xact_time_rate5m / pgbouncer:db:xact_rate5m

pgbouncer:ins:xact_time_rate5m / pgbouncer:ins:xact_rate5m

pgbouncer:cls:xact_time_rate5m / pgbouncer:cls:xact_rate5m

pgbouncer:env:xact_time_rate5m / pgbouncer:env:xact_rate5m

pgbouncer:db:xact_time_rate15m / pgbouncer:db:xact_rate15m

pgbouncer:ins:xact_time_rate15m / pgbouncer:ins:xact_rate15m

pgbouncer:cls:xact_time_rate15m / pgbouncer:cls:xact_rate15m

pgbouncer:env:xact_time_rate15m / pgbouncer:env:xact_rate15m

rate(pgbouncer_stat_total_query_count{datname!="pgbouncer"}[1m])

sum without(datname) (pgbouncer:db:query_rate1m)

sum by (job, cls) (pgbouncer:ins:query_rate1m)

sum by (job) (pgbouncer:cls:query_rate1m)

rate(pgbouncer_stat_total_query_count{datname!="pgbouncer"}[5m])

sum without(datname) (pgbouncer:db:query_rate5m)

sum by (job, cls) (pgbouncer:ins:query_rate5m)

sum by (job) (pgbouncer:cls:query_rate5m)

rate(pgbouncer_stat_total_query_count{datname!="pgbouncer"}[15m])

sum without(datname) (pgbouncer:db:query_rate15m)

sum by (job, cls) (pgbouncer:ins:query_rate15m)

sum by (job) (pgbouncer:cls:query_rate15m)

rate(pgbouncer_stat_total_query_time{datname!="pgbouncer"}[1m])

sum without(datname) (pgbouncer:db:query_time_rate1m)

sum by (job, cls) (pgbouncer:ins:query_time_rate1m)

sum by (job) (pgbouncer:cls:query_time_rate1m)

rate(pgbouncer_stat_total_query_time{datname!="pgbouncer"}[5m])

sum without(datname) (pgbouncer:db:query_time_rate5m)

sum by (job, cls) (pgbouncer:ins:query_time_rate5m)

sum by (job) (pgbouncer:cls:query_time_rate5m)

rate(pgbouncer_stat_total_query_time{datname!="pgbouncer"}[15m])

sum without(datname) (pgbouncer:db:query_time_rate15m)

sum by (job, cls) (pgbouncer:ins:query_time_rate15m)

sum by (job) (pgbouncer:cls:query_time_rate15m)

pgbouncer:db:query_time_rate1m / pgbouncer:db:query_rate1m

pgbouncer:ins:query_time_rate1m / pgbouncer:ins:query_rate1m

pgbouncer:cls:query_time_rate1m / pgbouncer:cls:query_rate1m

pgbouncer:env:query_time_rate1m / pgbouncer:env:query_rate1m

pgbouncer:db:query_time_rate5m / pgbouncer:db:query_rate5m

pgbouncer:ins:query_time_rate5m / pgbouncer:ins:query_rate5m

pgbouncer:cls:query_time_rate5m / pgbouncer:cls:query_rate5m

pgbouncer:env:query_time_rate5m / pgbouncer:env:query_rate5m

pgbouncer:db:query_time_rate15m / pgbouncer:db:query_rate15m

pgbouncer:ins:query_time_rate15m/ pgbouncer:ins:query_rate15m

pgbouncer:cls:query_time_rate15m / pgbouncer:cls:query_rate15m

pgbouncer:env:query_time_rate15m / pgbouncer:env:query_rate15m

stddev_over_time(pgbouncer:db:xact_rate1m[15m])

stddev_over_time(pgbouncer:ins:xact_rate1m[15m])

stddev_over_time(pgbouncer:cls:xact_rate1m[15m])

stddev_over_time(pgbouncer:env:xact_rate1m[15m])

stddev_over_time(pgbouncer:db:xact_rt_1m[15m])

stddev_over_time(pgbouncer:ins:xact_rt_1m[15m])

stddev_over_time(pgbouncer:cls:xact_rt_1m[15m])

stddev_over_time(pgbouncer:env:xact_rt_1m[15m])

stddev_over_time(pgbouncer:db:query_rate1m[15m])

stddev_over_time(pgbouncer:ins:query_rate1m[15m])

stddev_over_time(pgbouncer:cls:query_rate1m[15m])

stddev_over_time(pgbouncer:env:query_rate1m[15m])

stddev_over_time(pgbouncer:db:query_rt_1m[15m])

stddev_over_time(pgbouncer:ins:query_rt_1m[15m])

stddev_over_time(pgbouncer:cls:query_rt_1m[15m])

stddev_over_time(pgbouncer:env:query_rt_1m[15m])

pgbouncer:ins:query_time_rate1m / on(ins) group_left pg:ins:cpu_count or (0 * pgbouncer_up - 1)

pgbouncer:ins:query_time_rate5m / on(ins) group_left pg:ins:cpu_count or (0 * pgbouncer_up - 1)

pgbouncer:ins:query_time_rate15m / on(ins) group_left pg:ins:cpu_count or (0 * pgbouncer_up - 1)

pgbouncer:cls:query_time_rate1m  / pg:cls:cpu_count or (0 * pg:cls:size - 1)

pgbouncer:cls:query_time_rate5m  / pg:cls:cpu_count or (0 * pg:cls:size - 1)

pgbouncer:cls:query_time_rate15m / pg:cls:cpu_count or (0 * pg:cls:size - 1)

max without (instance) (pg:ins:load1 > ignoring(instance) pgbouncer:ins:load1 or pgbouncer:ins:load1)

max without (instance) (pg:ins:load5 > ignoring(instance) pgbouncer:ins:load5 or pgbouncer:ins:load5)

max without (instance) (pg:ins:load15 > ignoring(instance) pgbouncer:ins:load15 or pgbouncer:ins:load15)

pg:cls:load1 > pgbouncer:cls:load1 or pgbouncer:cls:load1

pg:cls:load5 > pgbouncer:cls:load5 or pgbouncer:cls:load5

pg:cls:load15 > pgbouncer:cls:load15 or pgbouncer:cls:load15

clamp_min(max without (instance) (pg:ins:cpu_usage_1m > pg:ins:saturation1 or pg:ins:saturation1), 0)

clamp_min(max without (instance) (pg:ins:cpu_usage_5m > pg:ins:saturation5 or pg:ins:saturation5), 0)

clamp_min(max without (instance) (pg:ins:cpu_usage_15m > pg:ins:saturation15 or pg:ins:saturation15), 0)

clamp_min(max by (job, cls) (pg:cls:cpu_usage_1m > pg:cls:saturation1 or pg:cls:saturation1), 0)

clamp_min(max by (job, cls) (pg:cls:cpu_usage_5m > pg:cls:saturation5 or pg:cls:saturation5), 0)

clamp_min(max by (job, cls) (pg:cls:cpu_usage_15m > pg:cls:saturation15 or pg:cls:saturation15), 0)

sum(patroni_failsafe_mode_is_active{}) > 0

pg_up < 1

pgbouncer_up < 1

patroni_up < 1

pg_exporter_agent_up < 1

pgbouncer_exporter_agent_up < 1

pg_uptime < 300

pg:cls:partition != 1

ins:pressure1 > 0.70

changes(pg_downstream_count{state="streaming"}[5m]) > 0

pg:ins:lag_bytes > 1048576 or pg:ins:lag_seconds > 1

pg:db:age > 0.80

pgbouncer:db:waiting_clients > 1

pgbouncer:db:query_rt_1m > 0.016

pg:db:conn_usage > 0.70

pg:db:ixact_backends > 1

node:ins:cpu_count + on(ip) group_right() pg_up * 0

sum by (job, cls) (pg:ins:cpu_count)

sum by (job) (pg:cls:cpu_count)

node:ins:cpu_usage_1m + on(ip) group_right() pg_up * 0

node:ins:cpu_usage_1m + on(ip) group_right() pg_up * 0

node:ins:cpu_usage_5m + on(ip) group_right() pg_up * 0

node:ins:cpu_usage_15m + on(ip) group_right() pg_up * 0

sum by (job, cls) (pg:ins:cpu_usage * pg:ins:cpu_count) / pg:cls:cpu_count

sum by (job, cls) (pg:ins:cpu_usage_1m * pg:ins:cpu_count) / pg:cls:cpu_count

sum by (job, cls) (pg:ins:cpu_usage_5m * pg:ins:cpu_count) / pg:cls:cpu_count

sum by (job, cls) (pg:ins:cpu_usage_15m * pg:ins:cpu_count) / pg:cls:cpu_count

sum by (job) (pg:cls:cpu_usage * pg:cls:cpu_count) / pg:env:cpu_count

sum by (job) (pg:cls:cpu_usage_1m * pg:cls:cpu_count) / pg:env:cpu_count

sum by (job) (pg:cls:cpu_usage_5m * pg:cls:cpu_count) / pg:env:cpu_count

sum by (job) (pg:cls:cpu_usage_15m * pg:cls:cpu_count) / pg:env:cpu_count

node_memory_MemTotal_bytes + on(ip) group_right() pg_up * 0

sum by (ip)(node_filesystem_size_bytes{fstype!~"(n|root|tmp|fuse.juice)fs.*"}) + on(ip) group_right() pg_up * 0

max by (ip)(node:fs:space_usage)+ on(ip) group_right() pg_up * 0

pg_up * 2 + (1 - pg_in_recovery)

1 - pg_in_recovery

(pg_in_recovery == 0) * 0 + 1

count by (job, cls, ins) (up{cls!="", ins!="", job='pgsql'})

sum by (job, cls) (group by (job, cls, ins, ip) (up{cls!="", ins!="", job='pgsql'}))

group by (job, cls, ins, ip) (up{cls!="", ins!="", job='pgsql'})

count without(appname,usename,address,pid) (pg_repl_lsn) and (pg_in_recovery == 1)

count by (job, cls) (pg_in_recovery == 0)

count by (job, cls) (count by (cls, ins) (pg_up==1))

count by (job, cls) (max by (cls, ins) (pg_in_recovery{}) == 0)

count by (job, cls) (max by (cls, ins) (pg_in_recovery{}) == 1)

max by (job, cls) (pg_sync_standby_enabled) > bool 0

count by (job, cls) (pg:cls:bridges)

max by (job, cls, ins, ip, appname, usename, address, pid) (pg_repl_sync_state)

max by ( job, cls, ip, appname, sender_host, sender_port, slot_name, state ) (
  label_replace(pg_recv_flush_tli, "appname" , "$1",  "ins", "(.*)" )
)

pg_checkpoint_tli

max by (job, cls) (max_over_time(pg_checkpoint_tli{}[30s]))

max by (job,cls,ins,instance) (label_replace(pg_repl_sync_state, "ins", "$1", "appname", "(.+)"))

(max by (job, cls) (pg:ins:sync_state)) > bool 0

max without (address, appname, usename, pid) (
  label_replace(pg_repl_replay_diff{appname!~'walreceiver|pg_receivewal'}, "ins", "$1", "appname", "(.+)")
)

max by (job, cls) (pg:ins:lag_bytes)

max by (job) (pg:cls:lag_bytes)

max without (address, appname, usename, pid) (
  label_replace(pg_repl_replay_lag{appname!~'walreceiver|pg_receivewal'}, "ins", "$1", "appname", "(.+)")
)

max by (job, cls) (pg:ins:lag_seconds)

max by (job) (pg:cls:lag_seconds)

count by (job, cls) (pg_slot_active > bool 0)

max without (slot_name) (pg_slot_retained_bytes)

max by (job, cls) (pg:ins:slot_retained_bytes)

rate(pg_db_xact_commit[1m])

sum without(datname) (pg:db:xact_commit_rate1m)

sum by (job, cls) (pg:ins:xact_commit_rate1m)

sum by (job) (pg:cls:xact_commit_rate1m)

rate(pg_db_xact_commit[5m])

sum without(datname) (pg:db:xact_commit_rate5m)

sum by (job, cls) (pg:ins:xact_commit_rate5m)

sum by (job) (pg:cls:xact_commit_rate5m)

rate(pg_db_xact_commit[15m])

sum without(datname) (pg:db:xact_commit_rate15m)

sum by (job, cls) (pg:ins:xact_commit_rate15m)

sum by (job) (pg:cls:xact_commit_rate15m)

rate(pg_db_xact_rollback[1m])

sum without(datname) (pg:db:xact_rollback_rate1m)

sum by (job, cls) (pg:ins:xact_rollback_rate1m)

sum by (job) (pg:cls:xact_rollback_rate1m)

rate(pg_db_xact_rollback[5m])

sum without(datname) (pg:db:xact_rollback_rate5m)

sum by (job, cls) (pg:ins:xact_rollback_rate5m)

sum by (job) (pg:cls:xact_rollback_rate5m)

rate(pg_db_xact_rollback[15m])