HA

GET /ha/metrics/capacity

HA capacity reserve

Returns the cached cluster CPU and RAM N+1 failover-reserve view; an unreachable cloud DB is reported in the body with 200.

Status Codes:
  • 200 OK – capacity reserve view

Response JSON Object:
  • available (boolean)

  • collected_at (string)

  • cpu (any) – capacity_type 1 (MHz = cores*speed)

  • ram (any) – capacity_type 0 (bytes)

  • routing_hosts (integer) – Up + Enabled Routing hosts counted

GET /ha/metrics/ceph

HA Ceph view

Returns the cached Ceph HA posture; an unreachable Ceph is reported in the body with 200.

Status Codes:
Response JSON Object:
  • available (boolean) – false => ceph CLI unreachable/errored => Unknown

  • bytes_avail (integer)

  • bytes_total (integer)

  • bytes_used (integer)

  • collected_at (string) – when this view was produced (staleness visibility; a served value may be up to cephViewTTL old)

  • health_checks[] (string)

  • health_status (string) – raw HEALTH_OK | HEALTH_WARN | HEALTH_ERR

  • mons_in_quorum (integer)

  • mons_total (integer)

  • objects (integer)

  • osds_in (integer)

  • osds_total (integer)

  • osds_up (integer)

  • pgs_active (integer) – sum of pgs_by_state counts whose state contains “active”

  • pgs_total (integer)

  • pools (integer)

  • status (string) – derived: healthy | degraded | critical | unknown

GET /ha/metrics/certs

HA certificates view

Returns the certificate cluster-wide view built from cached metrics.

Status Codes:
Response JSON Object:
  • expiring_count (integer)

  • nodes[].certs[].days_left (integer)

  • nodes[].certs[].file (string)

  • nodes[].certs[].not_after (string)

  • nodes[].hostname (string)

  • nodes[].node_ip (string)

  • total_certs (integer)

GET /ha/metrics/dns

HA DNS view

Returns the Technitium DNS/DHCP cluster-wide view built from cached metrics.

Status Codes:
Response JSON Object:
  • cluster_health (string)

  • nodes[].hostname (string)

  • nodes[].node_ip (string)

  • nodes[].status (string)

  • nodes[].total_leases (integer)

  • nodes[].total_zones (integer)

  • nodes[].version (string)

GET /ha/metrics/forecast

Capacity forecast

Returns the capacity fill forecast from a linear fit over Prometheus history.

Status Codes:
Response JSON Object:
  • available (boolean)

  • collected_at (string)

  • items[].current_pct (number) – most recent observed utilization %

  • items[].days_to_threshold (number) – ETA in days; null=not filling / too few points / ALREADY at-or-over threshold (read CurrentPct vs ThresholdPct to tell an over-threshold breach from a not-filling indeterminate)

  • items[].instance (string) – series instance label (”” for a cluster-wide series)

  • items[].resource (string) – curated forecast key (e.g. “host_mem”, “ceph_used”)

  • items[].slope_pct_per_day (number) – least-squares slope over the window

  • items[].threshold_pct (number) – the fill threshold the ETA targets

  • recommendation (string) – human-readable soonest-to-fill / “no resource trending toward full”

  • threshold_pct (number)

  • window_days (integer)

GET /ha/metrics/keepalived

HA Keepalived view

Returns the Keepalived/VRRP cluster-wide view built from cached metrics.

Status Codes:
Response JSON Object:
  • cluster_health (string)

  • nodes[].hostname (string)

  • nodes[].instances[].eff_priority (integer)

  • nodes[].instances[].interface (string)

  • nodes[].instances[].master_router (string)

  • nodes[].instances[].name (string)

  • nodes[].instances[].priority (integer)

  • nodes[].instances[].state (string) – MASTER | BACKUP | FAULT | UNKNOWN

  • nodes[].instances[].virtual_ip (string)

  • nodes[].instances[].vrid (integer)

  • nodes[].node_ip (string)

GET /ha/metrics/local

Local HA posture

Returns this node’s own cached HA posture without aggregation; polled by peers.

Status Codes:
  • 200 OK – this node’s HA posture

Response JSON Object:
  • certs[].days_left (integer)

  • certs[].file (string)

  • certs[].not_after (string)

  • collected_at (string)

  • envoy_edge.edge_cx_active (integer) – http.edge_public.downstream_cx_active (live gauge)

  • envoy_edge.edge_cx_total (integer) – http.edge_public.downstream_cx_total

  • envoy_edge.edge_rq_2xx (integer) – http.edge_public.downstream_rq_2xx

  • envoy_edge.edge_rq_4xx (integer) – http.edge_public.downstream_rq_4xx

  • envoy_edge.edge_rq_5xx (integer) – http.edge_public.downstream_rq_5xx

  • envoy_edge.edge_rq_total (integer) – http.edge_public.downstream_rq_total

  • envoy_edge.edge_rx_bytes_total (integer) – http.edge_public.downstream_cx_rx_bytes_total

  • envoy_edge.edge_tx_bytes_total (integer) – http.edge_public.downstream_cx_tx_bytes_total

  • envoy_edge.lb_policy (string) – constant “equal” — uniform active-active edge

  • envoy_edge.server_uptime_secs (integer) – server.uptime — drops on an Envoy restart; lets the dashboard detect a counter RESET before delta-ing the *_total counters for req/s (avoids a negative/huge spurious rate)

  • envoy_edge.upstream_cx_active (integer) – cluster.api_backend.upstream_cx_active (live gauge)

  • envoy_edge.upstream_rq_active (integer) – cluster.api_backend.upstream_rq_active (live gauge)

  • errors[] (string)

  • hostname (string)

  • keepalived.instances[].eff_priority (integer)

  • keepalived.instances[].interface (string)

  • keepalived.instances[].master_router (string)

  • keepalived.instances[].name (string)

  • keepalived.instances[].priority (integer)

  • keepalived.instances[].state (string) – MASTER | BACKUP | FAULT | UNKNOWN

  • keepalived.instances[].virtual_ip (string)

  • keepalived.instances[].vrid (integer)

  • mariadb.currency (string) – Currency is the witness split-brain fence verdict (VALID | SUPERSEDED). Per the witness contract VALID means “not proven superseded”, NOT “healthy” — the evaluator gates it by quorum. Empty => witness file unread (Unknown).

  • mariadb.currency_stale_secs (integer)

  • mariadb.master_host (string) – Master_Host from SHOW SLAVE STATUS — this replica’s upstream primary (replication topology display); “” on a primary/unconfigured

  • mariadb.read_only (boolean) – @@global.read_only (1 => fenced/replica)

  • mariadb.replica_configured (boolean) – SHOW SLAVE STATUS returned rows (replication actually set up) — distinguishes a real replica from a read_only DB with NO slave (fresh peer / fenced ex-primary / starved status read). Only when true do the replica_io/sql/seconds series carry meaning.

  • mariadb.replica_io_ok (boolean) – Slave_IO_Running=Yes (only meaningful when ReplicaConfigured)

  • mariadb.replica_sql_ok (boolean) – Slave_SQL_Running=Yes (only meaningful when ReplicaConfigured)

  • mariadb.role (string) – PRIMARY | REPLICA | UNKNOWN

  • mariadb.running (boolean) – mysqladmin ping ok

  • mariadb.seconds_behind (integer) – Seconds_Behind_Master; nil = N/A/NULL

  • mariadb.slave_status_read (boolean) – SHOW SLAVE STATUS executed OK this cycle (empty OR rows). Distinguishes a PROVEN “no replica configured” (read succeeded, empty) from a STARVED/errored read that merely LOOKS empty — so ReplicaConfigured=false is only a divergence-suspect when this is true.

  • mariadb.vip_held (boolean) – holds the :mysql VIP label

  • nfs_secondary.detail (string)

  • nfs_secondary.status (string)

  • node_ip (string)

  • ntp_sources[].address (string)

  • ntp_sources[].name (string) – the name the source was configured with, e.g. pool.ntp.org

  • ntp_sources[].reach (string) – octal reachability register, e.g. 377

  • ntp_sources[].reachable (boolean) – Reach register != 0

  • ntp_sources[].selected (boolean) – chrony’s current best source (^*)

  • ntp_sources[].state (string) – *, +, -, x, ~, ? per chronyc

  • ntp_sources[].stratum (integer)

  • ntp_tracking.frequency_ppm (number) – clock frequency error (signed: - = slow)

  • ntp_tracking.last_offset_seconds (number) – chrony’s last measured correction (signed)

  • ntp_tracking.leap_status (string)

  • ntp_tracking.node (string)

  • ntp_tracking.reference (string) – the source chrony is tracking (name or ref id)

  • ntp_tracking.rms_offset_seconds (number) – long-run offset magnitude

  • ntp_tracking.skew_exceeded (boolean) – |system_offset| exceeds the threshold

  • ntp_tracking.stratum (integer)

  • ntp_tracking.synchronized (boolean) – stratum > 0 and a real reference

  • ntp_tracking.system_offset_seconds (number) – system clock offset from NTP (signed)

  • ntp_tracking.zone (string)

  • partial (boolean) – true => one+ probes were DROPPED by the collect budget, so this snapshot is incomplete: some components are N/A because they were not read, NOT because they are healthy. A partial snapshot must NOT render a node “healthy” (role-independent — the budget fires on timeout regardless of role).

  • plugins_pg.detail (string)

  • plugins_pg.status (string)

  • postgresql.currency (string) – VALID | SUPERSEDED (witness)

  • postgresql.currency_stale_secs (integer)

  • postgresql.replay_lag (string)

  • postgresql.replay_lag_secs (number) – numeric twin of ReplayLag (standby only; nil = N/A) for the textfile gauge

  • postgresql.role (string) – PRIMARY | STANDBY | UNKNOWN

  • postgresql.running (boolean)

  • postgresql.sender_host (string)

  • postgresql.standbys[].app_name (string) – application_name label (masked to “” without pg_read_all_stats)

  • postgresql.standbys[].flush_lag_secs (number)

  • postgresql.standbys[].state (string) – streaming | catchup | startup | backup

  • postgresql.standbys[].sync_state (string) – sync | async | quorum | potential

  • postgresql.standbys[].write_lag_secs (number) – nil = NULL (e.g. a just-connected standby); never fabricated

  • postgresql.standbys_read (boolean) – pg_stat_replication read succeeded; empty Standbys is then a PROVEN “no standby connected”, not Unknown

  • postgresql.sync_names (string) – raw value (”” = async)

  • postgresql.sync_standby_on (boolean) – synchronous_standby_names non-empty

  • postgresql.vip_held (boolean) – holds the :pg VIP label

  • postgresql.wal_receiver_read (boolean) – the wal-receiver probe succeeded; when false the streaming gauge is Unknown (omitted), never a fabricated 0.

  • postgresql.wal_receiver_streaming (boolean) – standby: WAL receiver connected+streaming (pg_stat_wal_receiver). false on a standby = dead/diverged receiver.

  • postgresql.wal_status (string)

  • quorum.available (boolean) – false => witness file unread => UNKNOWN

  • quorum.fence_reason (string) – “” | QUORUM_LOST | PG_SUPERSEDED — why Fenced (empty iff not fenced)

  • quorum.fenced (boolean) – THIS node’s DB writes are fenced right now — read from the SAME oracle (WriteFenceStatus) that actually gates writes, so the chip can never disagree with reality. Always false on a single-node cluster (Topology single).

  • quorum.stale_secs (integer) – now - file mtime at collection (staleness; witness stamps none inside)

  • quorum.state (string) – DISARMED | QUORATE | LOST | UNKNOWN

  • quorum.topology (string) – single | multi — on “single” the write-fence is DELIBERATELY DISABLED (witnessSingleNode), so a split-brain-guard chip must render INERT there, NOT green (else a false-green on a cluster where the guard cannot fire)

  • resources.cpu_percent (number) – busy% over the interval since the last read (Δnon-idle/Δtotal); nil on the first sample

  • resources.nics[].device (string)

  • resources.nics[].speed_mbps (integer) – node_network_speed_bytes*8/1e6; nil when the driver reports -1/unknown (e.g. a bridge)

  • resources.nics[].up (boolean) – node_network_up (operstate up)

  • resources.ram_percent (number) – (MemTotal-MemAvailable)/MemTotal*100 — instant

  • step_ca.detail (string)

  • step_ca.status (string)

  • systemd.services[].active (boolean)

  • systemd.services[].name (string)

  • systemd.services[].status (string)

  • technitium.status (string)

  • technitium.total_leases (integer)

  • technitium.total_zones (integer)

  • technitium.version (string)

  • uptime_secs (number) – host uptime (/proc/uptime); already the Vault boot-grace clock, now surfaced for the dashboard. 0 => /proc/uptime unreadable

  • vault.detail (string)

  • vault.status (string)

  • vault.version (string)

GET /ha/metrics/mariadb

HA MariaDB view

Returns the MariaDB cluster-wide view built from cached metrics.

Status Codes:
Response JSON Object:
  • cluster_health (string)

  • nodes[].currency (string)

  • nodes[].hostname (string)

  • nodes[].master_host (string) – upstream primary (replication topology) — parity with PostgreSQLNodeView.SenderHost

  • nodes[].node_ip (string)

  • nodes[].read_only (boolean)

  • nodes[].replica_io_ok (boolean)

  • nodes[].replica_sql_ok (boolean)

  • nodes[].role (string)

  • nodes[].running (boolean)

  • nodes[].seconds_behind (integer)

  • nodes[].sync_status (string) – per-replica posture: SYNC | LAG | LOST | UNKNOWN (”” on a primary). UNKNOWN on a starved SHOW SLAVE STATUS (never a fabricated LOST); LAG above dbLagFloorSecs.

  • nodes[].vip_held (boolean)

  • primary_node (string)

GET /ha/metrics/mgmt-capacity

Management server capacity

Returns per management server agent capacity, the N+1-safe cluster maximum and an add-a-server recommendation.

Status Codes:
  • 200 OK – management capacity view

Response JSON Object:
  • available (boolean)

  • binding_server (string) – the MS that saturates first (advisory)

  • collected_at (string)

  • n1_safe (boolean) – total_agents <= survivors’ capacity; null=indeterminate (never a fabricated green)

  • n1_safe_max_agents (integer) – agents the cluster can serve AND survive losing its largest MS; null=indeterminate

  • recommendation (string) – human-readable “add an MS when…” / “N+1-safe up to X”

  • servers[].agent_count (integer) – agents CURRENTLY served (the platform-authoritative, per-MS)

  • servers[].axes[].max_agents (integer) – agents this axis alone allows; null=at/below idle OR observational (never a constraint)

  • servers[].axes[].name (string) – cpu | heap (bindable) | ram (observational)

  • servers[].axes[].target_pct (number) – the ceiling target

  • servers[].axes[].util_pct (number) – observed utilization %

  • servers[].binding_axis (string) – the bindable resource axis that saturates first (cpu|heap; ram is observational, never binds)

  • servers[].headroom (integer) – max_agents - agent_count; null=indeterminate

  • servers[].max_agents (integer) – learned ceiling at the target utilization; null=indeterminate

  • servers[].name (string)

  • servers[].service_ip (string)

  • servers[].state (string)

  • target_utilization_pct (number) – headroom target the ceilings are computed against

  • total_agents (integer) – agents currently served across all MS

GET /ha/metrics/nfs-secondary

HA NFS secondary view

Returns the per-node secondary storage NFS cluster view built from cached metrics.

Status Codes:
Response JSON Object:
  • cluster_health (string)

  • nodes[].detail (string)

  • nodes[].hostname (string)

  • nodes[].node_ip (string)

  • nodes[].status (string)

GET /ha/metrics/nodes

HA per-node metrics

Returns the cached raw metrics of every management node.

Status Codes:
GET /ha/metrics/plugins-pg

HA plugins-PG view

Returns the per-node plugins PostgreSQL cluster view built from cached metrics.

Status Codes:
Response JSON Object:
  • cluster_health (string)

  • nodes[].detail (string)

  • nodes[].hostname (string)

  • nodes[].node_ip (string)

  • nodes[].status (string)

GET /ha/metrics/postgresql

HA PostgreSQL view

Returns the PostgreSQL cluster-wide view built from cached metrics.

Status Codes:
Response JSON Object:
  • cluster_health (string)

  • nodes[].currency (string)

  • nodes[].hostname (string)

  • nodes[].node_ip (string)

  • nodes[].replay_lag (string)

  • nodes[].role (string)

  • nodes[].running (boolean)

  • nodes[].sender_host (string)

  • nodes[].standbys[].app_name (string) – application_name label (masked to “” without pg_read_all_stats)

  • nodes[].standbys[].flush_lag_secs (number)

  • nodes[].standbys[].state (string) – streaming | catchup | startup | backup

  • nodes[].standbys[].sync_state (string) – sync | async | quorum | potential

  • nodes[].standbys[].write_lag_secs (number) – nil = NULL (e.g. a just-connected standby); never fabricated

  • nodes[].sync_standby_on (boolean)

  • nodes[].sync_status (string) – per-standby replication posture: SYNC | LOST | UNKNOWN (”” on a primary). No LAG tier — PG lag needs primary-side pg_stat_replication (#15); ReplayLagSecs false-fires on an idle primary.

  • nodes[].vip_held (boolean)

  • nodes[].wal_status (string)

  • primary_node (string)

GET /ha/metrics/quorum

HA quorum view

Returns the cluster-wide quorum verdict view built from cached metrics.

Status Codes:
Response JSON Object:
  • cluster_health (string)

  • lost_nodes (integer)

  • nodes[].hostname (string)

  • nodes[].node_ip (string)

  • nodes[].stale_secs (integer)

  • nodes[].state (string)

  • quorate_nodes (integer)

POST /ha/metrics/refresh

Refresh HA metrics

Invalidates the metrics cache at most once per debounce window and returns the cluster summary.

Status Codes:
Response JSON Object:
  • active_alerts[].detail (string) – the component message / cert file+days

  • active_alerts[].node_ip (string) – for node/cert alerts

  • active_alerts[].severity (string) – critical | degraded | warning

  • active_alerts[].source (string) – node | component | cert

  • active_alerts[].title (string) – node IP / component name / “certificate expiring”

  • cert_alerts[].days_left (integer)

  • cert_alerts[].file (string)

  • cert_alerts[].node_ip (string)

  • cert_alerts[].not_after (string)

  • cluster_health (string) – healthy, degraded, critical

  • collected_at (string)

  • components.keepalived.message (string)

  • components.keepalived.status (string)

  • components.mariadb.message (string)

  • components.mariadb.status (string)

  • components.nfs_secondary.message (string)

  • components.nfs_secondary.status (string)

  • components.plugins_pg.message (string)

  • components.plugins_pg.status (string)

  • components.postgresql.message (string)

  • components.postgresql.status (string)

  • components.quorum.message (string)

  • components.quorum.status (string)

  • components.step_ca.message (string)

  • components.step_ca.status (string)

  • components.technitium.message (string)

  • components.technitium.status (string)

  • components.vault.message (string)

  • components.vault.status (string)

  • healthy_nodes (integer) – reachable AND node-health==”healthy” (survivability KPI); Reachable != Healthy (a reachable degraded node is not counted)

  • lost_nodes (integer) – witness LOST count (a partitioned minority); UNKNOWN/unread counts as neither

  • nodes[].health (string) – healthy, degraded, critical, unreachable

  • nodes[].hostname (string)

  • nodes[].node_ip (string)

  • nodes[].reachable (boolean)

  • quorate_nodes (integer) – witness QUORATE count (the fraction NUMERATOR, e.g. 6/8; denominator is TotalNodes) — same tally as /quorum

  • reachable_nodes (integer)

  • total_nodes (integer)

  • vips[].backup_count (integer)

  • vips[].last_moved_at (string) – LastMovedAt is the timestamp of the last OBSERVED master change for this VIP (keyed by name+VRID). nil until a real A->B failover is seen AFTER this core process started — cold-start and empty<->master transitions are deliberately NOT treated as moves, so a core restart never fabricates a “just moved” badge. Best-effort in-memory recency hint; the durable failover record is app_events (karios_ha_event.sh) surfaced in the HA event feed. See vip_movement.go.

  • vips[].master_count (integer) – nodes reporting MASTER for this VIP; >1 => split-brain VRRP (a partitioned VLAN, both sides claim the VIP)

  • vips[].master_ip (string)

  • vips[].master_node (string)

  • vips[].name (string)

  • vips[].state (string) – MASTER (exactly one) | NO_MASTER (none) | DUAL_MASTER (>1 -> VRRP split)

  • vips[].virtual_ip (string)

  • vips[].vrid (integer)

GET /ha/metrics/step-ca

HA step-ca view

Returns the per-node internal CA (step-ca) cluster view built from cached metrics.

Status Codes:
Response JSON Object:
  • cluster_health (string)

  • nodes[].detail (string)

  • nodes[].hostname (string)

  • nodes[].node_ip (string)

  • nodes[].status (string)

GET /ha/metrics/summary

HA cluster summary

Returns the aggregated cluster health summary built from cached per-node metrics.

Status Codes:
Response JSON Object:
  • active_alerts[].detail (string) – the component message / cert file+days

  • active_alerts[].node_ip (string) – for node/cert alerts

  • active_alerts[].severity (string) – critical | degraded | warning

  • active_alerts[].source (string) – node | component | cert

  • active_alerts[].title (string) – node IP / component name / “certificate expiring”

  • cert_alerts[].days_left (integer)

  • cert_alerts[].file (string)

  • cert_alerts[].node_ip (string)

  • cert_alerts[].not_after (string)

  • cluster_health (string) – healthy, degraded, critical

  • collected_at (string)

  • components.keepalived.message (string)

  • components.keepalived.status (string)

  • components.mariadb.message (string)

  • components.mariadb.status (string)

  • components.nfs_secondary.message (string)

  • components.nfs_secondary.status (string)

  • components.plugins_pg.message (string)

  • components.plugins_pg.status (string)

  • components.postgresql.message (string)

  • components.postgresql.status (string)

  • components.quorum.message (string)

  • components.quorum.status (string)

  • components.step_ca.message (string)

  • components.step_ca.status (string)

  • components.technitium.message (string)

  • components.technitium.status (string)

  • components.vault.message (string)

  • components.vault.status (string)

  • healthy_nodes (integer) – reachable AND node-health==”healthy” (survivability KPI); Reachable != Healthy (a reachable degraded node is not counted)

  • lost_nodes (integer) – witness LOST count (a partitioned minority); UNKNOWN/unread counts as neither

  • nodes[].health (string) – healthy, degraded, critical, unreachable

  • nodes[].hostname (string)

  • nodes[].node_ip (string)

  • nodes[].reachable (boolean)

  • quorate_nodes (integer) – witness QUORATE count (the fraction NUMERATOR, e.g. 6/8; denominator is TotalNodes) — same tally as /quorum

  • reachable_nodes (integer)

  • total_nodes (integer)

  • vips[].backup_count (integer)

  • vips[].last_moved_at (string) – LastMovedAt is the timestamp of the last OBSERVED master change for this VIP (keyed by name+VRID). nil until a real A->B failover is seen AFTER this core process started — cold-start and empty<->master transitions are deliberately NOT treated as moves, so a core restart never fabricates a “just moved” badge. Best-effort in-memory recency hint; the durable failover record is app_events (karios_ha_event.sh) surfaced in the HA event feed. See vip_movement.go.

  • vips[].master_count (integer) – nodes reporting MASTER for this VIP; >1 => split-brain VRRP (a partitioned VLAN, both sides claim the VIP)

  • vips[].master_ip (string)

  • vips[].master_node (string)

  • vips[].name (string)

  • vips[].state (string) – MASTER (exactly one) | NO_MASTER (none) | DUAL_MASTER (>1 -> VRRP split)

  • vips[].virtual_ip (string)

  • vips[].vrid (integer)

HA historical trends

Returns a curated historical trend from the Prometheus TSDB for the requested metric and range.

Query Parameters:
  • metric (string) – host_cpu, host_mem or ceph_used

  • range (string) – 1h, 6h, 24h or 7d (default 24h)

Status Codes:
Response JSON Object:
  • available (boolean)

  • collected_at (string)

  • metric (string) – the curated metric key requested

  • note (string) – set on unavailable/invalid-param (e.g. “prometheus auth required”)

  • range (string) – the range enum requested (1h|6h|24h|7d)

  • series[].labels (object) – the Prometheus series labels (e.g. {“instance”:”node-01”})

  • series[].points[].t (integer) – unix seconds

  • series[].points[].v (number) – value at t (units per the metric; the curated metrics are all percent)

GET /ha/metrics/vault

HA Vault view

Returns the per-node Vault cluster view built from cached metrics.

Status Codes:
Response JSON Object:
  • cluster_health (string)

  • nodes[].detail (string)

  • nodes[].hostname (string)

  • nodes[].node_ip (string)

  • nodes[].status (string)

GET /ha/metrics/vips

HA VIP ownership

Returns the VIP ownership summary built from cached metrics.

Status Codes:
GET /ha/policy

Get HA timing policy

Returns the stored HA-timing knobs, their effective loader-clamped values and when each change applies.

Status Codes:
Response JSON Object:
  • apply_timing (object) – ApplyTiming names which knobs applied immediately vs need the witness restart (which this save already triggered on the writable primary; peers apply on their next 30s reconcile tick from the replicated row).

  • effective (object)

  • stored.currencyGrace (integer)

  • stored.id (integer)

  • stored.mariaDBSvcGrace (integer)

  • stored.pgsvcGrace (integer)

  • stored.quorumLossGrace (integer)

  • stored.updatedAt (string)

  • stored.witnessInterval (integer)

POST /ha/policy

Set HA timing policy

Stores the full HA-timing knob state (omitted knobs are unset), applies it on this node and returns stored and effective values.

Request JSON Object:
  • currency_grace (integer)

  • mariadb_svc_grace (integer)

  • pg_svc_grace (integer)

  • quorum_loss_grace (integer)

  • witness_interval (integer)

Status Codes:
Response JSON Object:
  • apply_timing (object) – ApplyTiming names which knobs applied immediately vs need the witness restart (which this save already triggered on the writable primary; peers apply on their next 30s reconcile tick from the replicated row).

  • effective (object)

  • stored.currencyGrace (integer)

  • stored.id (integer)

  • stored.mariaDBSvcGrace (integer)

  • stored.pgsvcGrace (integer)

  • stored.quorumLossGrace (integer)

  • stored.updatedAt (string)

  • stored.witnessInterval (integer)