# Grafana Dashboard YAML format
# This YAML config file has format shown below.
#
# Explanation of entries:
# =======================
# - section: Section heading
# - row: Indicates a new row (future will allow multiples per row)
# - title: Title of the panel
#   - expr: # is the PromQL query - note that variables are frefixed by '$'
#           # In the case of the flag --no-sdp, then this will be removed: 'sdpinst="$sdpinst",'
#   - expr: p4_monitor_by_cmd{sdpinst="$sdpinst",serverid="$serverid"}
#       # legend is to ensure the legend shows relevant fields from metrics labels
#       #   For more information on options - see 'Labels' column in: https://github.com/perforce/p4prometheus#metrics-available
#     legend: "{{cmd}}"
# Example:
# - section: Monitor Tracking
# - row: 1
# - title: Monitor Processes (by cmd)
#   target:
#   - expr: p4_monitor_by_cmd{sdpinst="$sdpinst",serverid="$serverid"}
#     legend: "{{cmd}}"
#
# Extra fields:
#   type: gauge # Note default is counter
#   yformat: s  # good for setting time values (seconds/hours/weeks/...)
#            decbytes - byte values
#            percent - 1-100%
#========================================================================
# Expected customisations:
# * Check the hardcoded values for replication in 'expr' and adjust for your environment:
#   Anything with the value '(_UPDATE_FOR_YOUR_SITE_)'
#   E.g.
#
# - title: Replica Lag for SAMPLE_REPLICA from SAMPLE_MASTER (_UPDATE_FOR_YOUR_SITE_)
#   target:
#   - expr: >-
#       p4_replica_curr_pos{instance="SAMPLE_MASTER_instance",sdpinst="1",servername="SAMPLE_MASTER"} -
#       ignoring (serverid, servername)
#       p4_replica_curr_pos{instance="SAMPLE_MASTER_instance",sdpinst="1",servername="SAMPLE_REPLICA"}

#========================================================================

# The following counters not included (for now)
# p4_cmd_ip_counter
# p4_cmd_ip_cumulative_seconds
# p4_cmd_running
# p4_cmd_user_counter
# p4_cmd_user_cumulative_seconds
# p4_cmd_user_detail_counter
# p4_cmd_user_detail_cumulative_seconds

- section: Disk Space (_UPDATE_FOR_YOUR_SITE_ - check mountpoints)
- row: 1
- title: Free Space Alert - filesys.*.min
  target:
  - expr: node_filesystem_free_bytes{mountpoint="/hxmetadata"} - on (customer, instance) p4_filesys_min{filesys="P4ROOT"}
    legend: "hxmetadata above min"
  - expr: node_filesystem_free_bytes{mountpoint="/hxlogs"} - on (customer, instance) p4_filesys_min{filesys="P4LOG"}
    legend: "hxlogs above min"
  - expr: node_filesystem_free_bytes{mountpoint="/hxdepots"} - on (customer, instance) p4_filesys_min{filesys="depot"}
    legend: "hxdepots above min"
  - expr: p4_filesys_min{customer="$customer", filesys=~"depots|P4ROOT|P4LOG"}
    legend: "min {{filesys}}"
  yformat: decbytes

- title: Filesystem percentage used (_UPDATE_FOR_YOUR_SITE_ - check mountpoints)
  target:
  - expr: >-
      100.0 - 100 *(node_filesystem_avail_bytes{mountpoint=~"/hx.*"})
          / (node_filesystem_size_bytes{mountpoint=~"/hx.*"})
    legend: "{{mountpoint}}"
  yformat: percent

- section: Monitor Tracking
- row: 1
- title: Monitor Processes (by cmd)
  target:
  - expr: p4_monitor_by_cmd{sdpinst="$sdpinst",serverid="$serverid"}
    legend: "{{cmd}}"
  - expr: sum(p4_monitor_by_cmd{sdpinst="$sdpinst",serverid="$serverid"})
    legend: all
- title: Monitor Processes (by user)
  target:
  - expr: p4_monitor_by_user{sdpinst="$sdpinst",serverid="$serverid"}
    legend: "{{user}}"
  - expr: sum(p4_monitor_by_user{sdpinst="$sdpinst",serverid="$serverid"})
    legend: all

- row: 1
- title: p4d process count
  target:
  - expr: p4_process_count
- title: rtv sessions active
  target:
  - expr: p4_rtv_svr_sessions_active

- row: 1
- title: $serverid Time for last SDP checkpoint
  target:
  - expr: p4_sdp_checkpoint_duration{sdpinst="$sdpinst",serverid="$serverid"}
  yformat: s
- title: Uptime
  type: gauge
  target:
  - expr: p4_server_uptime{sdpinst="$sdpinst"}
  yformat: s

- title: $serverid time since last SDP checkpoint
  target:
  - expr: time() - p4_sdp_checkpoint_log_time{sdpinst="$sdpinst",serverid="$serverid"}
  yformat: s

- row: 1
- title: All P4 Cmds Count (rate/10min)
  target:
  - expr: rate(p4_completed_cmds_per_day{sdpinst="$sdpinst",serverid="$serverid"}[10m])
  - expr: sum(rate(p4_cmd_counter{sdpinst="$sdpinst",serverid="$serverid"}[10m]))
- title: p4d log lines read (rate/min)
  target:
  - expr: rate(p4_prom_log_lines_read{sdpinst="$sdpinst",serverid="$serverid"}[1m])

- row: 1
- title: p4prometheus cmds pending
  target:
  - expr: p4_prom_cmds_pending{sdpinst="$sdpinst",serverid="$serverid"}
- title: p4prometheus cmds processed (rate/min)
  target:
  - expr: rate(p4_prom_cmds_processed{sdpinst="$sdpinst",serverid="$serverid"}[1m])

- row: 1
- title: p4prometheus CPU system (rate/min)
  target:
  - expr: rate(p4_prom_cpu_system{sdpinst="$sdpinst",serverid="$serverid"}[1m])
- title: p4prometheus CPU user (rate/min)
  target:
  - expr: rate(p4_prom_cpu_user{sdpinst="$sdpinst",serverid="$serverid"}[1m])

- row: 1
- title: Error Count rates by subsystem/id
  target:
  - expr: rate(p4_error_count{sdpinst="$sdpinst",serverid="$serverid",subsystem!~"[0-9].*"}[1m])
    legend: subsys {{subsystem}}, level {{level}}, id {{error_id}}

- row: 1
- title: P4 cmds by program - count (rate/10min)
  target:
  - expr: rate(p4_cmd_program_counter{sdpinst="$sdpinst",serverid="$serverid"}[10m])
    legend: "{{program}}"
- title: P4 cmds by program - duration (rate/10min)
  target:
  - expr: rate(p4_cmd_program_cumulative_seconds{sdpinst="$sdpinst",serverid="$serverid"}[10m])
    legend: "{{program}}"

- section: Replication
- row: 1
- title: Replica Journal number (from servers -J on master)
  target:
  - expr: p4_replica_curr_jnl{sdpinst="$sdpinst",serverid="$serverid"}
    legend: "{{servername}}"
- title: Replica Journal Pos (from servers -J on master)
  target:
  - expr: p4_replica_curr_pos{sdpinst="$sdpinst",serverid="$serverid"}
    legend: "{{servername}}"

# This is replaced by the metrics below - because this requires subtracting with named instances
# - row: 1
# - title: Replica Lag for SAMPLE_REPLICA from SAMPLE_MASTER (_UPDATE_FOR_YOUR_SITE_)
#   target:
#   - expr: >-
#       p4_replica_curr_pos{sdpinst="$sdpinst",instance="SOME_MASTER_INSTANCE",servername="SAMPLE_MASTER"} -
#       ignoring (serverid, servername)
#       p4_replica_curr_pos{sdpinst="$sdpinst",instance="SOME_REPLICA_INSTANCE",servername="SAMPLE_REPLICA"}

- row: 1
- title: Pull replica journals behind
  target:
  - expr: p4_pull_replica_journals_behind{sdpinst="$sdpinst"}
- title: Replica lag (bytes)
  target:
  - expr: p4_pull_replica_lag{sdpinst="$sdpinst"}
  yformat: decbytes

- row: 1
- title: Pull queue size on replica
  target:
  - expr: p4_pull_queue{sdpinst="$sdpinst"}
- title: rtv_repl_behind_bytes
  target:
  - expr: p4_rtv_rpl_behind_bytes{instance="$sdpinst"}

- row: 1
- title: Pull queue errors
  target:
  - expr: p4_pull_errors{sdpinst="$sdpinst",serverid="$serverid"}
- title: Replication errors
  target:
  - expr: p4_pull_replication_error{sdpinst="$sdpinst",serverid="$serverid"}

- row: 1
- title: Cmd counts per replica IP rate/min
  target:
  - expr: rate(p4_cmd_replica_counter{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "{{replica}}"
- title: Cmd counts per replica IP duration/min
  target:
  - expr: rate(p4_cmd_replica_cumulative_seconds{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "{{replica}}"

- row: 1
- title: Files synced (add/update/delete) per min
  target:
  - expr: rate(p4_sync_files_added{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "added"
  - expr: rate(p4_sync_files_updated{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "updated"
  - expr: rate(p4_sync_files_deleted{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "deleted"
- title: Bytes synced (add/update) per min
  target:
  - expr: rate(p4_sync_bytes_added{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "added"
    yformat: decbytes
  - expr: rate(p4_sync_bytes_updated{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "updated"
    yformat: decbytes

- section: Cmd Count and Duration
- row: 1
- title: Cmds duration (rate/min)
  target:
  - expr: rate(p4_cmd_cumulative_seconds{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "{{cmd}}"
- title: p4 cmds top 10 (rate/min)
  target:
  - expr: sum without (instance, job)(rate(p4_cmd_counter{sdpinst="$sdpinst",serverid="$serverid"}[1m]))
    legend: "{{cmd}}"

- row: 1
- title: Cmds CPU System (rate/min)
  target:
  - expr: rate(p4_cmd_cpu_system_cumulative_seconds{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "{{cmd}}"
- title: Cmds CPU User (rate/min)
  target:
  - expr: rate(p4_cmd_cpu_user_cumulative_seconds{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "{{cmd}}"

- row: 1
- title: Cmd Error Counter (rate/min)
  target:
  - expr: rate(p4_cmd_error_counter{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "{{cmd}}"

- title: Trigger duration (rate/min)
  target:
  - expr: rate(p4_total_trigger_lapse_seconds{sdpinst="$sdpinst",serverid="$serverid"}[1m])
    legend: "{{trigger}}"

- section: Table Locking
- row: 1
- title: P4 Read Locks
  target:
  - expr: p4_locks_db_read{sdpinst="$sdpinst",serverid="$serverid"}
    legend: db
  - expr: p4_locks_cliententity_read{sdpinst="$sdpinst",serverid="$serverid"}
    legend: cliententity
  - expr: p4_locks_meta_read{sdpinst="$sdpinst",serverid="$serverid"}
    legend: meta
- title: P4 Write Locks
  target:
  - expr: p4_locks_db_write{sdpinst="$sdpinst",serverid="$serverid"}
    legend: db
  - expr: p4_locks_cliententity_write{sdpinst="$sdpinst",serverid="$serverid"}
    legend: cliententity
  - expr: p4_locks_meta_write{sdpinst="$sdpinst",serverid="$serverid"}
    legend: meta

- row: 1
- title: p4 read locks held per table (rate/min)
  target:
  - expr: sum without (instance, job) (rate(p4_total_read_held_seconds{sdpinst="$sdpinst",serverid="$serverid"}[1m]))
    legend: "{{table}}"
- title: p4 read locks waiting (rate/min)
  target:
  - expr: sum without (instance, job) (rate(p4_total_read_wait_seconds{sdpinst="$sdpinst",serverid="$serverid"}[1m]))
    legend: "{{table}}"

- row: 1
- title: p4 write locks held per table (rate/min)
  target:
  - expr: sum without (instance, job) (rate(p4_total_write_held_seconds{sdpinst="$sdpinst",serverid="$serverid"}[1m]))
    legend: "{{table}}"
- title: p4 write locks wait per table (rate/min)
  target:
  - expr: sum without (instance, job) (rate(p4_total_write_wait_seconds{sdpinst="$sdpinst",serverid="$serverid"}[1m]))
    legend: "{{table}}"

- row: 1
- title: RTV processes waiting for locks
  target:
  - expr: p4_rtv_db_lockwait{sdpinst="$sdpinst",serverid="$serverid"}
- title: RTV DB I/O record count
  target:
  - expr: p4_rtv_db_io_records{sdpinst="$sdpinst",serverid="$serverid"}

- row: 1
- title: RTV is checkpoint active
  target:
  - expr: p4_rtv_db_ckp_active{sdpinst="$sdpinst",serverid="$serverid"}
- title: RTV checkpoint records processed
  target:
  - expr: p4_rtv_db_ckp_records{sdpinst="$sdpinst",serverid="$serverid"}
- row: 1
- title: RTV replica byte lag
  target:
  - expr: p4_rtv_rpl_behind_bytes{sdpinst="$sdpinst",serverid="$serverid"}
- title: RTV replica journal lag
  target:
  - expr: p4_rtv_rpl_behind_journals{sdpinst="$sdpinst",serverid="$serverid"}

- row: 1
- title: RTV active sessions
  target:
  - expr: p4_rtv_svr_sessions_active{sdpinst="$sdpinst",serverid="$serverid"}
- title: RTV total sessions
  target:
  - expr: p4_rtv_svr_sessions_total{sdpinst="$sdpinst",serverid="$serverid"}

- row: 1
- title: Processes waiting on read locks
  target:
  - expr: p4_locks_db_read{sdpinst="$sdpinst",serverid="$serverid"}
- title: Processes waiting on write locks
  target:
  - expr: p4_locks_db_write{sdpinst="$sdpinst",serverid="$serverid"}

- row: 1
- title: Processes waiting on cliententity read locks
  target:
  - expr: p4_locks_cliententity_read{sdpinst="$sdpinst",serverid="$serverid"}
- title: Processes waiting on cliententity write locks
  target:
  - expr: p4_locks_cliententity_write{sdpinst="$sdpinst",serverid="$serverid"}

- row: 1
- title: Processes waiting on meta_read locks
  target:
  - expr: p4_locks_meta_read{sdpinst="$sdpinst",serverid="$serverid"}
- title: Processes waiting on meta_write
  target:
  - expr: p4_locks_meta_write{sdpinst="$sdpinst",serverid="$serverid"}

- row: 1
- title: Processes blocked count
  target:
  - expr: p4_locks_cmds_blocked{sdpinst="$sdpinst",serverid="$serverid"}